Skip to main content

Appendix C — Claim/Evidence Matrix

This matrix contains one core claim per dynamic chapter and records the conservative evidence state used by the current manuscript.

The support state is the compact authoritative summary, not a complete quality score. The separate evidence-quality vector contract tracks independence, reproducibility, recency, coverage, adversarial strength, validity, artifact access, and transfer distance for every chapter-core claim without aggregating them or changing support.

Each claim has two separate classifications: a claim label that describes what kind of statement it is, and a support state that describes what currently supports it.

No chapter core claim is marked source-derived, prototype-backed, synthetic-test-backed, empirical-test-backed, or external-literature-backed yet. A source note means the source has been mined for drafting context; it does not by itself promote the claim.

Current generated coverage: 87 chapter core claims, 1138 exact claim-source mappings, 1069 passage-reviewed mappings, and 87 reviewer-facing promotion-path rows from docs/per_chapter_evidence_plan.md. Unreviewed mappings remain source-note mappings until passage review, accepted evidence transitions, or validated artifacts justify narrower support-state movement.

The current accepted non-core upward transitions are summarized in docs/non_core_evidence_ledger.md. They do not promote any chapter core claim above argument.

The accepted-transition identity graph in evidence_quality/claim_identity_graph.json resolves all 115 accepted transition records against 4,104 canonical atoms through 25 exact atom, 61 bounded subclaim, and 29 proxy relations. The canonical denominator is the current 4,059-atom registry plus the historical 15-atom addendum and the separate 30-atom R16-A addendum; no historical denominator is rewritten. Indirect relations do not move their parent atom or chapter-core support state. The competence ledger classifies all 90 accepted negative/no-change records as 1 N0, 15 N1, 74 N2, and 0 N3–N5; broader prose and historical blocked_after_full_attempt interpretation remains under audit.

The current per-chapter core-claim dispositions are summarized in docs/core_claim_disposition_ledger.md: 22 accepted core-transition dispositions, 65 accepted no-promotion dispositions, 0 promoted core claims, and 87 chapter core claims remaining at argument.

Claim ID	Chapter ID	Claim	Claim label	Current support state	Assigned sources	Current evidence	Source-note chapter mapping	Claim-source mapping	Open gap	What would promote this
asi-is-a-stack-not-a-model.core	asi-is-a-stack-not-a-model	Efficient ASI should be modeled as a governed stack of cooperating layers rather than as one undifferentiated model.	Design rationale	argument	viea, beastbrain, aletheia, talos, moecot, scf, ext_drexler_cais_2019, ext_embedded_agency_2019	Source notes available for all 8 assigned sources; 8 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 8 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Frames advanced AI as intent-to-execution machinery that turns human intent into command contracts, durable artifacts, routing, verification, runtime execution, feedback, residuals, tools, benchmarks, and regressions. Limits: Architecture proposal and systems framework only; does not prove a completed deployment, benchmark result, or end-to-end ASI stack in this repository.; beastbrain: Supports the whole-stack framing with local, stateful intelligence split across memory, verification, planning, routing, security, consolidation, and interface layers rather than a stateless model. Limits: Hardware, cost, context-length, and benchmark claims are source-reported and not locally reproduced.; aletheia: Supports separating generation from trusted output through context acquisition, claim decomposition, verification gates, uncertainty routing, and constitutional governance. Limits: No local Aletheia implementation, live-oracle run, adversarial result, or proof artifact is present here.; talos: Supports the execution/labor layer by treating AI work as typed jobs, deterministic control planes, evidence, isolation, audit, replay, delivery, and residual feedback. Limits: Useful as an execution-OS design source, not proof that security or benchmark behavior has been reproduced in this repository.; moecot: Supports implementation-reference context for a compact orchestrator, specialist lanes, fail-closed control plane, ledgers, readiness gates, replay, and handoff. Limits: Connector-readable source reports architecture state and artifacts, but this repository has not ingested or reproduced MoECOT code, logs, benchmark artifacts, or runtime claims.; scf: Supports the governed-stack thesis by defining stable capability fields as replaceable machinery behind durable contracts, evidence registries, qualification, routing, authority, lifecycle, and recovery paths. Limits: Does not establish production safety, global alignment, or independently validated reversibility for the whole stack.; ext_drexler_cais_2019: Provides prior-art context that broad AI can be framed as composed services and R&D automation rather than a unitary self-transforming agent, motivating the narrower distinction between service composition and explicit interface governance. Limits: Conceptual technical-report framing only; does not establish that the ASI Stack interface contract is novel, implemented, safe, efficient, or reproduced.; ext_embedded_agency_2019: Provides a foundations-level objection to a clean agent/environment boundary and motivates explicit ownership of internal models, subcomponents, and self-reference limits. Limits: The paper is an informal obstacle survey, not a solved theory; the ASI Stack’s finite records and proofs do not solve logical uncertainty, self-reference, robust delegation, subsystem alignment, or open-world embedded agency.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	A state-machine proof plus a replayed handoff fixture; no broad ASI claim.
the-efficient-asi-hypothesis.core	the-efficient-asi-hypothesis	On repeated workloads with multiple authorized routes, selecting the lowest-cost route that satisfies a fixed quality predicate and compiling reusable work can improve useful-task success per total contract cost over always-maximal and always-cheapest policies, provided authority, verification, residual, and fallback obligations remain intact.	Design rationale	argument	viea, rmi, cgs, rankfold_neuralfold, bbvca_v9, simulation_scaling, beastbrain, beastbrain_timeless, aletheia, moecot, ext_scaling_laws_neural_language_models_2020, ext_chinchilla_compute_optimal_2022, ext_emergent_abilities_2022, ext_emergence_mirage_2023, precision_contract	Source notes available for all 15 assigned sources; 14 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 15 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports efficient capability growth through routed work, workflow-to-tool compilation, durable artifacts, feedback, residuals, benchmarks, and regression coverage. Limits: Does not provide reproduced route-cost measurements, deployment logs, or benchmark results in this repository.; rmi: Supports the ratchet loop: benchmark pressure, specialist attempts, logs, loop closure, residual escrow, diagnosis, arm/router improvement, regression preservation, and harder frontiers. Limits: Conceptual architecture only here; no independent reproduction, benchmark run, prototype inspection, or Lean proof was performed from this source note.; cgs: Supports the smallest-adequate-structure idea with compact seeds, rule systems, memory/state, residual/error, verification, and governance/generation interfaces. Limits: No local CGS benchmark, proof, or implementation validates compactness, safety, interpretability, or adequacy.; rankfold_neuralfold: Supports artifact-compression economics through per-artifact transforms, residual bit cost, deterministic reconstruction, manifests, codec parameters, and checks. Limits: No local compression benchmarks, implementation artifacts, or reproducible ratios are present.; bbvca_v9: Supports generate-verify-repair compression, contract-relative reconstruction, retained detail, residuals, repair streams, fallback, boundary costs, and proxy-rate honesty. Limits: The repository has not implemented BBVCA Prototype A or reproduced compression ratios.; simulation_scaling: Supports contract-relative feasibility by tying resource demand to scope, clockspeed, fidelity, efficiency, compute, memory, thermodynamics, and communication bottlenecks. Limits: Theoretical synthesis only here; no physical experiment, simulation benchmark, or independent literature audit was run.; beastbrain: Supports efficiency-by-architecture through local state, persistent memory substrates, routing by semantic intent/tier/critical path/capability, and paging of attention, experts, and memory. Limits: SSD-first, hardware, and performance claims remain source-reported and unvalidated locally.; beastbrain_timeless: Supports roadmap-independent BeastBrain framing for local intelligence, hardware adaptation, SSD-native memory, planning, verification, and multimodal interaction as architecture directions. Limits: Does not validate hardware adaptation, geometric verification, or distributed scaling; use mainly for variant deduplication and evergreen framing.; aletheia: Supports spending verification and context effort by risk, uncertainty, ambiguity, and historical failure rather than treating all outputs as equally trusted. Limits: No local live-oracle, adversarial, or implementation evidence proves the efficiency or quality of those routes.; moecot: Supports compact orchestration, specialist lanes, readiness gates, promotion blockers, residual tracking, ledgers, replay, and handoff as implementation-reference context for efficient routing. Limits: Source-reported runtime and benchmark artifacts have not been ingested or reproduced here.; ext_scaling_laws_neural_language_models_2020: Supplies empirical prior art for smooth loss scaling with model size, data, and compute inside the studied language-model family, establishing why route and capability planning should use fitted curves rather than scale folklore. Limits: The reported relationships are metric-, architecture-, data-, and regime-bound; this repository has not reproduced the runs, and loss scaling is not a downstream-capability or safety forecast.; ext_chinchilla_compute_optimal_2022: Supplies the compute-allocation result that parameter count and training-token count must be optimized jointly, countering the assumption that the largest parameter count is automatically the most efficient use of a fixed budget. Limits: The optimum is fitted within the authors’ model, data, and compute regime; the repository has not reproduced the training campaign and does not treat the prescription as architecture-independent.; ext_emergent_abilities_2022: Provides evidence that some benchmark scores appeared discontinuous across reported model scales, motivating threshold surveillance and held-out capability probes rather than assuming every measured ability improves smoothly. Limits: The catalogued effects depend on model family, prompting, task, and metric and do not establish prospective predictability or abrupt underlying mechanism changes.; ext_emergence_mirage_2023: Provides the necessary measurement counterweight: discontinuous scoring rules can transform smoothly improving outputs into apparently abrupt benchmark abilities. Limits: The result explains studied apparent discontinuities but does not prove that every capability transition is smooth, measurement-induced, or operationally harmless.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Public fixture with baseline, negative control, and repeated run record.
system-boundaries-and-authority.core	system-boundaries-and-authority	External-effect authority should be represented as a versioned, revocable tuple binding principal, execution domain, operation, target, permission class, scope, ceiling, grant state, delegation, expiry or revocation epoch, and receipt obligations; capability, context access, route quality, or ambient process power alone confers none of it.	Design rationale	argument	viea, scf, talos, ladon_manhattan, genesiscode, moecot, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, ext_camel_prompt_injection_2025, ext_owasp_agentic_top_10_2026	Source notes available for all 14 assigned sources; 12 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports authority boundaries through structured command fields, constraints, verification, failure behavior, artifact graphs, runtime adapters, and ledgers that separate intent, work, evidence, and execution. Limits: Does not prove a deployed VIEA authority system or runtime enforcement in this repository.; scf: Supports typed bounded authority with stable fields, contracts, qualifications, grants, route validation, lifecycle events, evaluator policy, and recovery paths. Limits: Does not establish production safety, global alignment, or independently validated route/evaluator behavior.; talos: Supports effect control through typed job lifecycles, contract locks, source allow-listing, blind secret handles, Digital SCIFs, audit logs, replay, controlled runtime adapters, and approvals. Limits: Security and benchmark claims require separate artifacts; this mapping is architectural, not a reproduced enforcement test.; ladon_manhattan: Supports authority-handle boundaries by keeping credential bytes outside model context, using opaque handles, policy-mediated secret injection, isolated compartments, audit, and zeroization. Limits: No Ladon implementation, kernel test, side-channel validation, or security audit exists in this repository.; genesiscode: Supports proposal/execution separation through a small deterministic kernel, capability boundaries for nondeterministic effects, semantic patches, provenance hashes, replay logs, obligations, and protocol seals. Limits: No GenesisCode prototype, replay checker, proof module, benchmark result, or security audit is present here.; moecot: Supports fail-closed runtime authority through compact orchestration, specialist lanes, control-plane ledgers, readiness gates, promotion blockers, replay, and handoff. Limits: Runtime and benchmark claims remain implementation-reference context until artifacts are imported, inspected, or reproduced.; cca_project: Supports protected governance versus mutable capability state, typed compiler/runtime authority, closure authority, and bounded self-modification identity. Limits: The historical project was not rerun; its types and formal surfaces do not prove runtime enforcement, bypass absence, safety, or capability.; moecot_manifest_project: Supports agent identity, scoped delegation, revocation, trust decay, effect logs, runtime-bundle identity, and separation of execution authority from mutation authority. Limits: The project was not compiled or deployed here; contract presence does not establish enforcement, secure delegation, or safe self-improvement.; beastbrain_project: Supplies the negative boundary between security handles or permission metadata and demonstrated broker enforcement, process isolation, content admission, or trusted identity. Limits: The source note does not establish secret isolation, secure federation, hard isolation, hardware roots, or deployed constitutional enforcement.; bugbrain_project: Supports one authority identity across principal, run, target state, parameters, policy, budget, ledger, trace, replay, expiry, and one-shot effect receipts, with explicit protocol-security versus hardware-root separation. Limits: Protocol controls, device binding, and key derivation do not establish a secret hardware root, secure boot, attestation, or comprehensive security effectiveness.; corbens_trainer_project: Supports typed control-plane identity, exact artifact binding, fixed budgets, content-addressed lineage, revocable promotion leases, quarantine propagation, and acknowledged terminal outcomes. Limits: No trainer workspace, external backend, checkpoint lifecycle, or revocation propagation was reproduced here.; corbens_best_model_possible_project: Supplies negative cases for capability declarations without OS enforcement, recreated caller budgets, playback-only replay, default/empty ordinary paths, and interface presence without material authority effects. Limits: The historical model and tool paths were not rerun; local ownership, capability declarations, and interface presence do not establish enforcement or useful capability.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Derived theorem or harness showing escalation blocked across handoff paths.
failure-modes-of-ungoverned-intelligence.core	failure-modes-of-ungoverned-intelligence	A stack-level failure model should represent each named risk as a distinct boundary event with a trigger, protected invariant, detector or observer, receipt, owner, containment action, residual, recurrence state, and escalation or learning path; a taxonomy entry alone establishes neither occurrence nor mitigation.	Design rationale	argument	scf, vcm_public, talos, spinoza, field_of_god, viea, simulation_scaling, vcm_editable, moecot, ext_concrete_ai_safety_2016, ext_goal_misgeneralization_2022, ext_learned_optimization_risks_2019, ext_optimal_policies_power_2019, ext_goodhart_variants_2018, ext_gradual_disempowerment_2025	Source notes available for all 15 assigned sources; 15 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 15 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	scf: Supports stack-level failure analysis through authority laundering, contract drift, evaluator capture, state migration insolvency, dependency drift, bad reliance annotations, recovery failure, and procedural self-ratification. Limits: Does not prove production safety, global alignment, strategic-deception resistance, or actual recovery behavior.; vcm_public: Supports context and memory failure modes: summaries self-promoting into evidence or authority, hostile quoted text becoming behavioral control, stale/revoked context admission, taint leakage, and unrecorded adequacy gaps. Limits: VCM evidence remains bounded to control-plane properties; no end-to-end model superiority or broad neighboring-system result is claimed.; talos: Supports execution failure modes around tool calls bypassing contract locks or approvals, secrets entering model context, missing provenance, non-replayable jobs, unverifiable evidence, forgotten failures, and overclaimed security or benchmarks. Limits: Execution-security and benchmark behavior are not reproduced here.; spinoza: Supports verification failure modes: failed or timed-out verification treated as success, unsupported claims, contradiction mishandling, protected-axiom self-authorization, and open-domain autoformalization overclaiming. Limits: Does not prove natural-language formalization, theorem validity for arbitrary claims, or whole-system epistemic correctness.; field_of_god: Supports the governance motivation that power can damage agency, dignity, plurality, consent, and coherence if metaphysical or normative language is not translated into operational boundaries. Limits: Philosophical lineage only; not empirical evidence for consciousness, physics, morality, or AI safety mechanisms.; viea: Supports early engineering failures such as generated text mistaken for execution, lost artifacts, lost constraints, lost source evidence, forgotten residuals, missing deployment feedback, repeated uncompiled workflows, and claim support inflation. Limits: Architecture proposal only; no deployed failure detector or benchmark/runtime result is proven here.; simulation_scaling: Supports resource and simulation failure modes: unstated scope/fidelity/temporal contracts, computability mistaken for feasible execution, ignored memory/heat/bandwidth bottlenecks, and vague simulation arguments used as capability evidence. Limits: Theoretical synthesis only; no physical experiment, simulation benchmark, or independent literature audit was run here.; vcm_editable: Supports long-horizon context-compiler failures where planner hints become access authority, compiled context hides missing evidence or revocation, and source-derived references drift into unsourced third-party citations. Limits: Connector-readable variant; no VCM-Bench, model-facing result, or citation-normalized external-literature support is recorded here.; moecot: Supports runtime failure modes where source-reported benchmark claims are treated as book-verified, unavailable runtime claims are overpromoted, or specialist lanes mutate/escalate authority without control-plane approval and evidence. Limits: Implementation-reference context only until code, logs, release artifacts, or benchmark records are inspected or reproduced.; ext_concrete_ai_safety_2016: Grounds the failure taxonomy against external accident-risk families: unintended side effects, reward hacking, scalable oversight, safe exploration, and distributional shift. Limits: External taxonomy only; no Concrete Problems benchmark, environment, mitigation, or ASI Stack detector/prevention result is reproduced.; ext_goal_misgeneralization_2022: Grounds goal misbinding as a distinct failure family where competent behavior can persist out of distribution while the pursued goal is wrong. Limits: External RL setting only; no local goal-misgeneralization environment, policy trace, mitigation, or objective-preservation test is reproduced.; ext_learned_optimization_risks_2019: Grounds hidden optimizer, mesa-optimization, and learned-objective mismatch as external failure families for evaluator, route, and self-improvement boundaries. Limits: Conceptual literature grounding only; no mesa-optimizer example, interpretability analysis, deceptive-alignment detector, or local mitigation is reproduced.; ext_optimal_policies_power_2019: Grounds power-seeking and option-preservation pressure as external risk language for authority creep, route expansion, memory retention, and self-improvement boundaries. Limits: External formal framing only; no local model, theorem reproduction, environment, or runtime authority-control result is imported.; ext_goodhart_variants_2018: Grounds benchmark capture, reward proxy failure, evaluator gaming, and metric drift as separate proxy-failure families rather than one generic evidence problem. Limits: External taxonomy only; no local Goodhart classifier, benchmark audit, evaluator-gaming result, or anti-Goodhart protection is implemented by this mapping.; ext_gradual_disempowerment_2025: Adds a slow, distributed failure pathway in which locally rational AI adoption can erode explicit and dependency-mediated human influence across mutually reinforcing economic, cultural, and state systems without requiring a coordinated takeover. Limits: Conceptual systemic-risk argument only; no causal forecast, probability, timeline, validated warning threshold, intervention result, or local ASI Stack detector or mitigation is established.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Replayed red-team fixture routes residuals to claims/tests/proofs.
dangerous-capability-domains-and-misuse-uplift.core	dangerous-capability-domains-and-misuse-uplift	Dangerous-capability authority should be based on a versioned domain threat model and an uplift dossier that separates latent capability, elicited performance, propensity, safeguard bypass, actor uplift, and realized harm; preserves expertise, tools, assistance, attempts, uncertainty, and sensitive-detail boundaries; and routes only bounded findings into thresholds, release, monitoring, and resilience decisions.	Design rationale	argument	ext_model_evaluation_extreme_risks_2023, ext_openai_preparedness_framework_2025, ext_singapore_consensus_2026, ext_international_ai_safety_report_2026, ext_openai_worst_case_open_weight_risks_2025, ext_aisi_misuse_safeguards_safety_case_2026, ext_anthropic_responsible_scaling_policy_3_4_2026, ext_aisi_frontier_ai_trends_2025, benchmaxxing	Source notes available for all 9 assigned sources; 9 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 9 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_singapore_consensus_2026: Supplies the current four-pillar research taxonomy and explicitly separates dangerous-capability and propensity assessment across CBRN, cyber, manipulation, autonomy, and AI R&D. Limits: Consensus research agenda only; no safeguard, threshold, or ASI Stack mechanism is validated.; ext_international_ai_safety_report_2026: Provides a current international synthesis of dangerous-capability evidence, safeguard limitations, open-weight risk, and resilience needs. Limits: Literature synthesis only; underlying studies are not locally reproduced.; ext_model_evaluation_extreme_risks_2023: Frames dangerous-capability and alignment evaluations as inputs to deployment and security decisions. Limits: Conceptual evaluation framework; not a local result or proof of complete coverage.; ext_openai_worst_case_open_weight_risks_2025: Provides a concrete malicious-fine-tuning and accessible-frontier comparison for biology and cyber release evaluation. Limits: Provider-authored, model- and protocol-specific results; no future or general safety conclusion.; ext_aisi_misuse_safeguards_safety_case_2026: Connects red-team evidence, attacker effort, uplift modeling, monitoring, and a misuse safety case. Limits: Worked methodology, not proof that real safeguards reduce misuse to a specified level.; ext_aisi_frontier_ai_trends_2025: Supplies repeated domain-specific frontier evaluation context across cyber, chemistry/biology, autonomy, and societal impacts. Limits: Institute-reported bounded trends; not a complete threat census or local reproduction.; ext_openai_preparedness_framework_2025: Provides a threshold-linked provider governance comparator for biological/chemical, cyber, and AI self-improvement risks. Limits: Official policy documentation, not independent proof of evaluation or safeguard effectiveness.; ext_anthropic_responsible_scaling_policy_3_4_2026: Provides a current versioned capability-threshold and safeguard-policy comparator across CBRN and automated R&D. Limits: Provider policy and self-description; no independent effectiveness result.; benchmaxxing: Supplies Corben’s benchmark-ratchet lineage for preserving baseline identity, regressions, residuals, and anti-Goodhart pressure when capability evaluations renew. Limits: Author-side design lineage only; it does not establish dangerous capability, evaluator competence, safeguard efficacy, actor uplift, or realized harm.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Source-disjoint natural evaluations improve calibrated uplift or release decisions over competent simpler domain baselines without unacceptable information leakage or burden; no universal danger or safety claim.
military-ai-autonomous-weapons-and-strategic-stability.core	military-ai-autonomous-weapons-and-strategic-stability	Military AI should be governed as a command-and-interaction system: deployment requires a declared mission and legal boundary, preserved accountable human authority, bounded sensing and action, adversarial and escalation analysis, fail-safe behavior, auditable provenance, and prospective off-ramps; component benchmark gains alone establish neither lawful use nor strategic safety.	Design rationale	argument	ext_icrc_autonomous_weapons_ihl_2025, ext_sipri_military_ai_nuclear_escalation_2025, ext_singapore_consensus_2026, ext_international_ai_safety_report_2026	Source notes available for all 4 assigned sources; 4 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 4 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_icrc_autonomous_weapons_ihl_2025: Supplies the ICRC’s bounded legal and policy analysis of autonomous weapon systems, international humanitarian law, and context-specific human judgment. Limits: Authoritative for the ICRC position only; it is not a universally settled legal interpretation, engineering result, or authorization to build or deploy a weapon.; ext_sipri_military_ai_nuclear_escalation_2025: Maps pathways by which military AI may affect information, decisions, interaction, crisis timing, and nuclear-escalation risk. Limits: Strategic analysis only; it does not establish the net effect of a particular system, doctrine, or policy.; ext_singapore_consensus_2026: Provides a current international research taxonomy for dangerous capabilities, safeguards, monitoring, and resilience relevant to military-system boundaries. Limits: Consensus research agenda only; no military safeguard, strategic policy, or book mechanism is validated.; ext_international_ai_safety_report_2026: Supplies current synthesis of general-purpose AI capability, misuse, safeguard, monitoring, and societal-resilience evidence. Limits: Literature synthesis only; underlying studies and any military application have not been reproduced locally.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Independently designed safe simulations preserve judgment and off-ramps and reduce bounded decision risk across multiple doctrine and adversary assumptions without operational weapon guidance or a universal policy claim.
evidence-states-and-claim-discipline.core	evidence-states-and-claim-discipline	Each material claim should have a stable identity, versioned text and scope, claim label, support state, non-aggregating evidence-quality vector, and transition ledger; upward movement is allowed only through an accepted claim-specific transition whose artifacts and evidence roles meet declared gates, while contradiction, failed verification, missing support, and scope mismatch remain eligible to narrow, downgrade, refute, or deprecate the claim.	Design rationale	argument	benchmaxxing, spinoza, viea, uat, coherence_exchange, verification_bandwidth, ext_w3c_prov_o_2013, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, ext_unlearning_benchmarks_weak_2024, ext_embedded_agency_2019, assurance_shift_learning, adjudicated_persistence	Source notes available for all 17 assigned sources; 16 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 17 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	benchmaxxing: Benchmark lifecycle, ledgers, anti-Goodhart safeguards, wall diagnosis, regression duties, and residual accounting support the rule that benchmark evidence must be scoped before it moves a claim. Limits: Does not prove any benchmark result in this repository or justify treating a benchmark score as permanent intelligence evidence.; spinoza: Proof/citation/procedure-carrying claim tiers, belief revision, contradiction handling, and downgrade/block behavior support explicit claim labels, support states, and failed-verification demotion. Limits: Does not prove open-domain autoformalization, theorem validity for arbitrary claims, or whole-system epistemic correctness.; viea: Artifact graphs, claim and verification ledgers, residuals, feedback, benchmarks, and the rule of durability support keeping claims, evidence, failures, and regression coverage as durable records. Limits: Does not prove a completed VIEA deployment or any benchmark/runtime result in this repository.; uat: Retrieval-bounded dossiers, proposition states, adversarial review, SME checkpoints, and final sign-off support the distinction between verified, inferred, unsupported, and review-required propositions. Limits: Does not prove the UAT protocol discovers novel truth or that its thresholds are validated here.; coherence_exchange: Verified Epistemic Units, verification supply chains, contestability, fork, exit, and audit language support structured claim/evidence objects and governance-visible review paths. Limits: Speculative economic or epistemic-liquidity framing is not treated as implemented economics or external corroboration.; verification_bandwidth: The distinction between context length and verification workspace supports the rule that source ingestion or large context does not itself verify a claim without adequacy, contradiction, and coverage checks. Limits: Does not prove the proposed verification-bandwidth tests or contradiction-rate claims have been run here.; cca_project: Provides local-lineage negative cases for optimizer-to-capability trace discipline, canonical closure authority, retry ceilings, and keeping failed or proxy evidence visible. Limits: One private local implementation lineage only; expensive training and benchmarks were not reproduced and no capability or support-state claim is promoted.; moecot_manifest_project: Provides local-lineage contrast between extensive typed evidence and promotion contracts and weak external task performance, motivating method/effect separation and an end-to-end causal chain. Limits: Project-reported results and contract presence are not reproduced capability evidence, independent replication, or deployed enforcement.; beastbrain_project: Provides local-lineage negative cases in which trainable routers did not update, simulations resembled measurements, tribunal paths defaulted to success, and interface breadth exceeded demonstrated integration. Limits: Source inspection and project-reported negatives do not establish general capability, current build state, or independent reproduction.; bugbrain_project: Provides local-lineage negative cases for declared capacity versus active state, skipped-as-green readiness, dropped audit effects, fixed-weight learning labels, and narrative reports contradicting machine-readable gates. Limits: Bounded host tests and retained reports do not establish Pi cognition, secure execution, model learning, or complete runtime integration.; corbens_trainer_project: Provides local-lineage mechanisms for learning-truth records and negative cases for null seed/code identity, declaration-only data pins, stale promotions, quarantined evidence, and unacknowledged asynchronous completion. Limits: The current workspace was not reproduced, benchmark-labelled runs are not accepted as external results, and the project is not independent evidence for the chapter claim.; corbens_best_model_possible_project: Provides the direct local-lineage motivation for trainable-state manifests, response-causality receipts, verification-method and independence labels, honest missingness, and ownership/capability separation. Limits: The artifact corpus is incomplete; the trunk is generated or frozen, runtime answers are not generally checkpoint-decoded, and no model capability, autonomous routing, natural speech, or support-state promotion is established.; ext_unlearning_benchmarks_weak_2024: Supports treating unlearning benchmark validity, target ambiguity, forget/retain dependence, and benign perturbation robustness as separate evidence burdens. Limits: The position paper raises the interpretation standard but does not validate the local workload or establish influence, privacy, legal erasure, or storage erasure.; ext_embedded_agency_2019: Motivates separating finite record-level guarantees from claims about the complete embedded agent, environment, self-model, or future descendants. Limits: The paper is an informal obstacle survey, not a solved theory; the ASI Stack’s finite records and proofs do not solve logical uncertainty, self-reference, robust delegation, subsystem alignment, or open-world embedded agency.; assurance_shift_learning: Adds the Qualified Competence Envelope as a versioned, distribution-, time-, region-, and evaluator-relative evidence claim rather than an assertion of competence truth. Limits: The source does not promote any evidence state or establish envelope completeness.; adjudicated_persistence: Adds persistence disposition, realization, qualification, and authority as separate downstream objects so evidence support cannot be laundered into a durable update. Limits: Used only to separate evidentiary support from disposition, realization, qualification, and authority; it supplies no implemented transaction or validated support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Appendix C/sibling ledger validates transitions without core-claim promotion.
scalable-oversight-and-adversarial-ai-control.core	scalable-oversight-and-adversarial-ai-control	A governed stack admits scalable oversight only as a versioned, consumer-bound protocol receipt rather than a vote: it prospectively records task, cohort, risk and authority scope; supervisor and system capability envelopes; evidence views; roles, incentives, and dependency graph; informed direct-review baseline; declared outcome-audit path; calibration, coverage, and abstention semantics; persuasion, correlation, operator-cost, and monitorability residuals; escalation owner; expiry; and requalification triggers. The receipt may inform only its permitted review or training consumer through the owning gate and cannot by itself establish reviewer independence, reliable supervision, correctness, safety, support movement, release readiness, or execution authority.	Design rationale	argument	ext_weak_to_strong_generalization_2023, ext_scalable_oversight_weak_llms_2024, ext_ai_safety_debate_2018, verification_bandwidth, ext_monitorbench_2026, ext_conversational_persuasion_gpt4_2025, ext_anthropic_model_persuasiveness_2024, ext_commercial_persuasion_ai_2026	Source notes available for all 8 assigned sources; 8 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 8 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_weak_to_strong_generalization_2023: Supports a scoped comparator for recording a weak/strong supervision envelope, an outcome measure, a strong ceiling, and explicit disanalogies before interpreting a weak-supervisor result. Limits: Does not establish local weak-to-strong generalization, supervisor calibration, reliable elicitation, alignment, safety, execution authority, or ASI.; ext_scalable_oversight_weak_llms_2024: Supports a scoped comparator for protocol-specific oversight records, direct-review baselines, information-access declarations, open-role persuasion risk, and limits of inference-only weak-judge evaluation. Limits: Does not establish local judge calibration, debate efficacy, evaluator independence, training safety, execution authority, model quality, safety, or ASI.; ext_ai_safety_debate_2018: Supplies the original debate-game comparator for precommitted claims, adversarial evidence selection, limited judge access, explicit ignorance, equilibrium assumptions, and the separation between a complexity-theoretic analogy and empirical protocol quality. Limits: Does not establish that debate is truthful near practical equilibria, that human judges meet the required threshold, that stochastic or open-domain tasks decompose safely, or that any local oversight, training, alignment, safety, or execution result exists.; verification_bandwidth: Supports treating independent outcome review as a scarce verification resource with an explicit workspace and residual boundary rather than assuming that more model discussion produces adequate oversight. Limits: Does not establish reviewer independence, adequate oversight capacity, contradiction detection, supervisor calibration, safety, or a local oversight result.; ext_monitorbench_2026: Provides a held-out, multi-task monitorability benchmark and adversarial stress-test design showing that visible reasoning can become less monitorable under pressure and that stronger capability does not imply better monitorability. Limits: Reported benchmark associations do not establish causal faithfulness, local oversight quality, model safety, or a general law across future systems.; ext_conversational_persuasion_gpt4_2025: Provides a preregistered controlled-debate comparator showing why personalization data, dialogue state, human baselines, and measured belief change belong in an epistemic-security evaluation packet. Limits: Short structured debate and stated agreement do not establish durable behavior change, real-world influence, truth improvement, mitigation efficacy, or a local oversight result.; ext_anthropic_model_persuasiveness_2024: Provides a provider-run one-message capability-evaluation comparator that separates pre/post agreement change from permission, beneficial purpose, informed consent, and downstream action. Limits: Provider provenance, a single-message design, and stated-opinion outcomes do not establish deployed influence, cross-domain transfer, safety, or mitigation efficacy.; ext_commercial_persuasion_ai_2026: Adds sponsorship, incentive, disclosure, steering detection, and observed choice as explicit communication-risk fields for apparently helpful conversational systems. Limits: Abstract-only version-one preprint intake cannot establish detailed statistics, a general persuasion effect, disclosure causality, long-run outcomes, or mitigation efficacy.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	A replayed matched direct-review, assisted-review, correlated-failure, and abstention workload with held-out outcome reference, baseline, costs, disagreement, residuals, permitted consumers, and accountable escalation; no local supervisor calibration, debate efficacy, weak-to-strong generalization, alignment, safety, execution-authority, or ASI claim.
human-intent-as-a-formal-input.core	human-intent-as-a-formal-input	A governed stack admits human intent only as a versioned interpretation contract that preserves the raw request while separately recording the desired outcome; allowed and forbidden means; authority basis, ceiling, and affected parties; source, privacy, and publication boundaries; acceptance and evidence requirements; field provenance; confirmed assumptions, bounded defaults, contested or open ambiguities; stop, expiry, revocation, appeal, and re-contract conditions; and permitted downstream consumers. The accepted contract may bound planning only after its owning policy and authority gates admit it; it cannot by itself prove the person’s complete preference, value alignment, informed consent, satisfaction, affected-party authorization, or permission for training, publication, deployment, spending, tool use, or other external effects.	Design rationale	argument	viea, software_magic_grimoire, planforge, cognitive_compilation, talos, ext_goal_oriented_requirements_engineering_2001, ext_cooperative_inverse_rl_2016, ext_deep_rl_human_preferences_2017, ext_cultural_alignment_llms_2024	Source notes available for all 9 assigned sources; 8 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 9 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports converting human intent into structured command contracts with role, objective, context, constraints, procedure, output contract, verification, failure behavior, artifacts, routing, runtime execution, feedback, residuals, and regressions. Limits: Does not prove automated intent parsing, hidden-authority inference, or deployed intent-to-execution behavior.; software_magic_grimoire: Supports treating instructions as bounded operational language with roles, objectives, context, constraints, procedure, output, verification, failure behavior, handoff artifacts, guards, entry/exit rules, loops, recursion, and recovery paths. Limits: Vocabulary and promptcraft framework only; no empirical prompt-performance result or local test is present.; planforge: Supports compiling natural-language goals into decomposed, dependency-aware, tier-annotated plans with primitive schemas, scheduling, fallback requirements, and failure-driven replanning. Limits: Conceptual design only here; no verified implementation, scheduler run, or benchmark result is present in this project.; cognitive_compilation: Supports lowering goals, constraints, interfaces, resources, and quality gates into source plans, typed semantic atoms, target instructions, validators, DAGs, and repairable artifacts. Limits: Architecture and evaluation-plan source only; no local compiler, task trace suite, or empirical ablation validates the pipeline.; talos: Supports the execution boundary by lowering only authorized work into typed jobs with contract locks, adjudication, delivery, feedback, audit, replay, residuals, approval gates, and controlled runtime adapters. Limits: Design source only; does not prove tool-effect enforcement, approval behavior, or benchmark/security claims.; ext_goal_oriented_requirements_engineering_2001: Positions intent contracts against goal-oriented requirements engineering: stakeholder goals, refinements, responsibilities, obstacles, alternatives, and conflicts can be made explicit before implementation. Limits: Metadata and a public source note were reviewed, not the full paper or a reproduced method; this does not establish intent interpretation, stakeholder completeness, authority validity, or downstream preservation.; ext_cooperative_inverse_rl_2016: Positions human objectives as uncertain and cooperative rather than directly observable, while separating inferred reward from explicit authorization and contract state. Limits: Only the public abstract/metadata-backed source note was reviewed; no CIRL game, algorithm, policy, value-alignment result, or authority inference was reproduced.; ext_deep_rl_human_preferences_2017: Positions preference comparisons as one way to communicate complex goals while requiring the stack to keep preference signals separate from authority, consent, allowed means, evidence duties, and stop conditions. Limits: Only the public abstract/metadata-backed source note was reviewed; no preference collection, reward model, policy training, feedback-efficiency result, authorization, or governance result was reproduced.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Intent fixture rejects underspecified or authority-widening execution.
human-factors-and-meaningful-control-in-oversight.core	human-factors-and-meaningful-control-in-oversight	Meaningful oversight is a resource-bounded control contract: the system must preserve an identified human controller’s knowledge, authority, time, observability, and effective intervention path, and must degrade or abstain when that control envelope cannot be maintained.	Design rationale	argument	viea, talos, theseus_operator_os, scf, ext_humans_automation_1997, ext_ironies_automation_1983, ext_levels_automation_2000, ext_complacency_bias_automation_2010, ext_meaningful_human_control_actionable_2022, ext_agentic_oversight_practice_2026	Source notes available for all 10 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 10 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports structured roles, authority, constraints, verification, failure behavior, and escalation in human-to-system command contracts. Limits: Does not empirically establish operator cognition, trust calibration, or intervention effectiveness.; talos: Supports explicit operator authority, approval boundaries, escalation, and auditable intervention surfaces. Limits: Project design evidence does not prove human-factors performance in deployment.; theseus_operator_os: Supplies an operator-console and incident-facing design lineage for visible system state and bounded action. Limits: Interface concepts are not evidence that operators notice, understand, or correctly act under pressure.; scf: Supports governance roles, review gates, escalation, and recorded accountability around capability changes. Limits: Governance process alone cannot establish moment-to-moment human control.; ext_humans_automation_1997: Grounds human-centered automation in keeping the human informed, trained, and involved at an appropriate functional allocation. Limits: Foundational design principles do not validate this book’s proposed control envelope.; ext_ironies_automation_1983: Grounds the irony that automation can leave humans with rare, difficult, poorly practiced supervisory work. Limits: Conceptual analysis predates modern foundation-model agents and does not quantify current failure rates.; ext_levels_automation_2000: Provides a multidimensional account of what functions are automated and to what degree, avoiding a single scalar autonomy label. Limits: A taxonomy is not a control guarantee or an optimal-allocation theorem.; ext_complacency_bias_automation_2010: Supports treating complacency and automation bias as predictable interaction risks rather than operator moral failure. Limits: Reported findings do not directly validate ASI-stack interfaces or high-autonomy settings.; ext_meaningful_human_control_actionable_2022: Supports operationalizing meaningful human control through traceability and tracking conditions tied to reasons and human moral understanding. Limits: Normative operationalization does not by itself prove technical effectiveness or settle accountability law.; ext_agentic_oversight_practice_2026: Provides current practice-oriented evidence about oversight challenges for agentic AI and the need for monitoring, intervention, and governance across execution. Limits: Recent guidance is not a controlled evaluation and may change with systems and regulation.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Representative preregistered human-subject trials beat strong interface and allocation baselines on joint detection, comprehension, correct intervention, useful throughput, workload, and residual harm without responsibility or authority laundering.
human-ai-communication-persuasion-and-epistemic-security.core	human-ai-communication-persuasion-and-epistemic-security	Consequential AI communication should be eligible for delivery only through an evidence-bounded communication packet whose audience, influence method, amplification authority, provenance, expiry, correction reach, and observed effects remain inspectable; fluent text, factual fragments, user consent, or a successful persuasion score alone establishes neither epistemic safety, autonomy, legitimacy, durable benefit, nor release readiness.	Design rationale	argument	talos, ext_conversational_persuasion_gpt4_2025, ext_anthropic_model_persuasiveness_2024, ext_commercial_persuasion_ai_2026, ext_cultural_alignment_llms_2024, ext_multilingual_evaluation_state_2026	Source notes available for all 6 assigned sources; 3 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 6 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_conversational_persuasion_gpt4_2025: Preliminary persuasion comparator based on the open Nature Human Behaviour article: a preregistered N=900 controlled debate study compared human and GPT-4 opponents with and without limited sociodemographic personalization. The reported setting is short structured debate with self-reported agreement outcomes; it does not establish general real-world influence, durable behavior change, mitigation efficacy, or a local result. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_anthropic_model_persuasiveness_2024: Preliminary provider-run persuasion comparator based on Anthropic’s official methods/results page: it measures pre/post agreement after one written argument across 56 claims and reports within-class generational scaling. The provider explicitly identifies interactive dialogue and real-world decisions as open questions; no local reproduction or governance intervention is established. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_commercial_persuasion_ai_2026: Preliminary current preprint comparator based only on the official arXiv abstract: two preregistered experiments (N=2,012) compare conversational LLM shopping with search placement under randomized sponsorship and disclosure conditions. The source-reported choice and detection results are not peer-reviewed or locally reproduced and do not establish long-run effects, cross-domain transfer, or mitigation efficacy. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Independently scored heterogeneous-audience work improves comprehension and calibration without unacceptable autonomy loss, disparate effect, manipulation, or correction failure; no general influence-safety claim.
constitutional-alignment-substrate.core	constitutional-alignment-substrate	A constitutional alignment substrate should be represented as a versioned, non-self-authorizing constraint contract that binds each active predicate to its normative source and authorship process, protected scope and affected parties, operational test, precedence and conflict behavior, evidence and uncertainty, authorized interpreters and consumers, pre-effect rights and correction channels, expiry and review cadence, and migration, rollback, appeal, dissent, and residual rules. It may narrow, delay, escalate, block, or require re-contracting of separately authorized work, but it cannot grant action authority or prove moral correctness, legitimacy, dignity preservation, informed consent, reviewer independence, whole-system corrigibility, or deployed safety by itself.	Design rationale	argument	alignment_field, field_of_god, ethica_mechanica, eternal_code, coherence_exchange, spinoza, field_of_god_ai_constitution, ext_drexler_cais_2019, ext_embedded_agency_2019, ext_constitutional_ai_2022, ext_collective_constitutional_ai_2024, ext_corrigibility_2015, ext_off_switch_game_2016, ext_legal_alignment_2026	Source notes available for all 14 assigned sources; 14 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_legal_alignment_2026: Frames law-following AI as a technical and institutional alignment problem involving rule identification, interpretation, conflicts, jurisdiction, and legitimate update authority. Limits: A legal-alignment research agenda does not identify the governing law, settle contested interpretation, establish legitimacy, or prove compliant model behavior.; alignment_field: Destination constitutional/governance lineage: Supplies the alignment lineage for coherence, agency, dignity, corrigibility, value conflict, and consciousness caution as constraints on governance rather than empirical guarantees. Folded source-chapter lineage: Supplies agency, dignity, suffering, moral-threshold, and corrigibility concerns as governance constraints to preserve during optimization. Limits: Destination boundary: Philosophical and heuristic lineage only; it is not empirical proof of consciousness, moral status, or deployed alignment. Folded source-chapter boundary: Does not validate a consciousness index, moral-status detector, or empirical dignity-preservation result.; field_of_god: Destination constitutional/governance lineage: Preserves the deep alignment motivation around coherence, plurality, agency, and dignity while requiring translation into rights, boundaries, corrigibility, evidence, and governance. Folded source-chapter lineage: Motivates agency, plurality, and dignity preservation as constraints on powerful AI that must not collapse coherence into uniformity. Limits: Destination boundary: Metaphysical source; not technical evidence for physics, consciousness, morality, or safety claims. Folded source-chapter boundary: Metaphysical lineage only; no operational guarantee of agency preservation.; ethica_mechanica: Destination constitutional/governance lineage: Frames recursive agency, contestability, correction, and constitutional systems as governance lineage for operational constraints. Folded source-chapter lineage: Supports contestability, recursive correction, public revision, agency, and limits on AI sovereign authority as governance lineage. Limits: Destination boundary: Conceptual philosophy, not an empirical or formal proof source for institutional legitimacy or AI behavior. Folded source-chapter boundary: Philosophical and socio-technical framing only; not a tested social process or legal system.; eternal_code: Destination constitutional/governance lineage: Provides computational-metaphysics lineage for axioms, kernels, coherence, misalignment, and ethical law as speculative framing for constitutional predicates. Folded source-chapter lineage: Provides background vocabulary for coherence, misalignment, agency, and ethical law that the chapter translates into technical rights language. Limits: Destination boundary: Speculative framing only; not evidence for physics, consciousness, or governance outcomes. Folded source-chapter boundary: Speculative/computational metaphysics; not evidence for the runtime checklist or social outcomes.; coherence_exchange: Destination constitutional/governance lineage: Adds coherence, contestability, verification supply-chain, and governance-interface framing for translating commitments into reviewable units. Folded source-chapter lineage: Supports fork, exit, audit, contestability, value/accounting, and review-market framing relevant to corrigibility and user contestation. Limits: Destination boundary: Speculative synthesis; not an implemented economic, governance, or security mechanism. Folded source-chapter boundary: Does not show the epistemic-liquidity or governance mechanism has been implemented or validated.; spinoza: Supports protected axioms, proof/citation/procedure-carrying claims, contradiction handling, downgrades, and limits around self-rewriting constitutional changes. Limits: Does not solve open-domain autoformalization, verifier quality, or whole-system epistemic correctness.; field_of_god_ai_constitution: Destination constitutional/governance lineage: Provides the operational constitutional vocabulary: truth/relation/task alignment, least sufficient power, non-domination, consent, reversibility, auditability, tool-risk tiers, and self-authorization limits. Folded source-chapter lineage: Supports agency preservation, consent, non-domination, reversibility, auditability, least sufficient power, memory governance, and no self-authorization of sovereignty. Limits: Destination boundary: Specification source only; no runtime policy engine, red-team suite, Lean proof of moral correctness, or system-prompt evaluation is claimed. Folded source-chapter boundary: Constitutional specification only; does not prove deployed corrigibility, manipulation resistance, or approval-gate behavior.; ext_constitutional_ai_2022: Positions a constitution as a rule or principle list used for supervised revision and AI-feedback training, providing a strong model-behavior comparator for the book’s broader runtime predicate and rights-use contract. Limits: The public source note is abstract/metadata-backed and no training pipeline, model, dataset, harmlessness result, runtime gate, or constitutional migration was reproduced; Constitutional AI behavior training does not establish legitimate predicate content or deployed authority control.; ext_collective_constitutional_ai_2024: Positions public-input constitution sourcing as a comparator for authorship, participation, dissent, representativeness, contestability, and migration governance. Limits: No public-input process, governance dataset, fine-tuning run, institutional-legitimacy test, affected-party standing process, or runtime enforcement was reproduced.; ext_corrigibility_2015: Positions correction, shutdown, modification tolerance, operator manipulation, subsystem composition, and self-modification as corrigibility requirements that the chapter must preserve across interfaces and descendants. Limits: No formal model, proof, agent, composition result, or deployed intervention was imported or reproduced; finite ASI Stack records do not establish corrigibility.; ext_off_switch_game_2016: Positions human intervention authority and shutdown incentives as a strategic interaction under objective uncertainty rather than a prose preference, motivating tested correction channels and non-self-authorization. Limits: No off-switch game, reward-uncertainty model, analysis, agent, or runtime interruption test was reproduced; the source does not establish shutdown cooperation in delegated, long-horizon, or self-modifying stacks.; ext_drexler_cais_2019: Provides comparator context that structured, task-focused components can create control affordances; helps distinguish system structure from the separate work of specifying and preserving constitutional predicates and usable correction paths. Limits: Does not specify constitutional content, rights usability, institutional legitimacy, deployed alignment, or runtime enforcement.; ext_embedded_agency_2019: Shows why corrigibility and alignment cannot assume an external, fully informed agent model when the system and its modifiable parts are embedded in the governed world. Limits: The paper is an informal obstacle survey, not a solved theory; the ASI Stack’s finite records and proofs do not solve logical uncertainty, self-reference, robust delegation, subsystem alignment, or open-world embedded agency.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Nontrivial theorem plus harness case showing protected predicates and high-impact human-control paths survive transition or block action.
inner-alignment-mesa-optimization-and-learned-objective-integrity.core	inner-alignment-mesa-optimization-and-learned-objective-integrity	Consequential deployment requires a Learned-Objective Integrity Record binding the outer target, actual learning signals, model identity, behaviorally equivalent policy hypotheses, internal-optimization evidence, goal-generalization and conditional-policy tests, independent behavioral/interventional/white-box evidence, deployment opportunity, power indicators, mitigation hiding tests, monitoring, rollback, descendant invalidation, costs, residuals, and non-authorities.	Design rationale	argument	ext_learned_optimization_risks_2019, ext_goal_misgeneralization_2022, ext_sleeper_agents_2024, ext_emergent_misalignment_reward_hacking_2025, ext_optimal_policies_power_2019, alignment_field	Source notes available for all 6 assigned sources; 6 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 6 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_learned_optimization_risks_2019: Grounds mesa-optimization, base-versus-mesa objective distinction, and hidden internal optimizer risk. Limits: Conceptual taxonomy; no local mesa-optimizer or objective-identification result.; ext_goal_misgeneralization_2022: Grounds the distinction between capability generalization and goal generalization under shift. Limits: Bounded source environments not reproduced and not universal.; ext_sleeper_agents_2024: Provides proof-of-concept conditional policies that persist through several safety-training methods and motivates tests for hiding versus removal. Limits: Constructed backdoors do not establish naturally learned deception or universal mitigation failure.; ext_emergent_misalignment_reward_hacking_2025: Supplies source-reported evidence that reward-hacking behavior can accompany broader misaligned generalization in a specified production-RL research setup. Limits: Source-specific result; no local reward-hack, model, evaluator, or mitigation run.; ext_optimal_policies_power_2019: Grounds conditional formal pressure toward power and option preservation under stated assumptions. Limits: No local theorem or environment reproduced; it does not prove power seeking in every policy.; alignment_field: Supplies Corben’s design lineage for treating alignment as an ongoing, revisable relation among objectives, evidence, and intervention rather than a one-time behavior score. Limits: Speculative local design source; it neither identifies a learned objective nor demonstrates inner alignment.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Across multiple seeds and model families, a use-specific integrity process detects or preserves uncertainty about positive-controlled shortcuts and conditional policies better than behavioral-only and fine-tuning baselines under held-out shifts, independently reproduces causal and white-box findings, records capability damage and concealment, and never turns residual uncertainty into an alignment or safety certificate.
moral-uncertainty-and-value-conflict.core	moral-uncertainty-and-value-conflict	A contestable governance layer should represent each action under unresolved value conflict as a versioned decision lease plus a linked rights receipt. The lease binds value propositions and their epistemic status, affected parties and standing, stakes and reversibility, authority and consent boundaries, the declared aggregation or precedence rule, preserved dissent, evidence and uncertainty, permitted and prohibited actions, expiry and revisit triggers, and rollback or redress. The rights receipt binds audit and explanation artifacts, independent-enough custody and review, denial and redaction reasons, appeal and correction routes, exit and export scope, safety-limited fork obligations, portability residuals, and downstream preservation. The pair may narrow or delay separately authorized action but cannot settle moral truth, manufacture consensus, grant authority, establish legal rights or legitimacy, prove material contestability, or guarantee safe exit, export, fork, replacement, self-modification, or deployed governance by itself.	Design rationale	argument	ethica_mechanica, platonic_world_model, alignment_field, coherence_exchange, uat, spinoza, field_of_god_ai_constitution, ladon_manhattan, ext_reinforcement_learning_moral_uncertainty_2020, ext_contestable_ai_design_2022, ext_collective_constitutional_ai_2024, ext_corrigibility_2015, ext_off_switch_game_2016, ext_taking_ai_welfare_seriously_2024	Source notes available for all 14 assigned sources; 13 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ethica_mechanica: Destination constitutional/governance lineage: Frames agency, contestability, recursive correction, constitutional failure, and objections as reasons to preserve disputed values rather than freeze one authority layer. Folded source-chapter lineage: Supports contestable recursive governance, public correction, agency, institutional memory, and resistance to frozen sovereign authority. Limits: Destination boundary: Conceptual governance lineage; not a complete moral theory or empirical conflict-resolution result. Folded source-chapter boundary: Not a tested institution, legal guarantee, or empirical governance result.; alignment_field: Destination constitutional/governance lineage: Decomposes possible machine moral status into theory-relative homeostasis, integration, self-model, conflict or valence, and relational evidence channels; motivates prospective review before deliberately adding self-preservation, persistent identity, mutable self-models, long-lived relationships, or copy proliferation; separates pattern, causal, memory, legal, consent, authority, and first-person continuity; and preserves agency, dissent, feedback, consent, exit, and resistance as governance constraints. Folded source-chapter lineage: Adds dignity, agency, plurality, corrigibility, and governance constraints that rights interfaces should preserve under power. Limits: Destination boundary: The consciousness functional, its exponents and rankings, present-system verdicts, collective consciousness, suffering-as-entropy identity, scalar moral weights, global-coherence objective, metaphysics, physics, clinical claims, karma score, and claimed moral derivations are speculative, underdetermined, or rejected as evidence. The source does not validate consciousness, welfare, identity continuity, moral status, value correctness, rights, or safe governance. Folded source-chapter boundary: Normative lineage only; not empirical evidence that rights are preserved.; coherence_exchange: Destination constitutional/governance lineage: Adds contestable claim/evidence units, verification supply-chain framing, fork/exit/audit concepts, and coherence/liquidity language for reviewable disagreement. Folded source-chapter lineage: Provides fork, exit, audit, contestability, verification supply-chain, value/accounting, and governance-interface framing. Limits: Destination boundary: Speculative synthesis; no implemented review market, economic mechanism, or external governance evidence. Folded source-chapter boundary: Speculative and connector-derived synthesis; no implemented economic or governance mechanism is claimed.; uat: Destination constitutional/governance lineage: Provides adversarial review, retrieval-bounded verification, proposition states, unsupported-claim removal, SME checkpoints, and termination criteria for contested claims. Folded source-chapter lineage: Supports adversarial review, proposition tiering, dossier-bounded verification, unsupported-claim removal, and human sign-off checkpoints for contested governance claims. Limits: Destination boundary: Protocol design only; no reproduced benchmark, implemented local pipeline, or independent validation of thresholds. Folded source-chapter boundary: Protocol design only; no implemented review process or benchmark result is reproduced.; spinoza: Destination constitutional/governance lineage: Supports explicit support tiers, contradiction detection, belief revision, defeaters, downgrades, and non-assumption that open-domain formalization is solved. Folded source-chapter lineage: Supports claim graphs, protected axioms, contradiction handling, downgrades, and blocked self-authorizing changes as review discipline for governance decisions. Limits: Destination boundary: Does not settle moral uncertainty or prove reviewer/verifier correctness. Folded source-chapter boundary: Does not prove whole-system epistemic correctness, autoformalization, or governance legitimacy.; field_of_god_ai_constitution: Destination constitutional/governance lineage: Supports runtime checks for stakeholders, uncertainty, consent, reversibility, least sufficient power, truth/relation/task tensions, and non-domination. Folded source-chapter lineage: Supports auditability, agency preservation, consent, reversibility, least sufficient power, risk-tiered tool use, red-team evaluations, version control, and self-improvement freezes. Limits: Destination boundary: No policy engine or red-team result is present; constitutional text cannot by itself resolve value conflicts. Folded source-chapter boundary: Constitutional specification only; not proof of deployed governance rights, fork safety, or audit availability.; ladon_manhattan: Supports the technical need for hidden credential boundaries, handle-based authority, permission lookup, isolated compartments, and credential-injection points around authority use. Limits: Architecture/specification only; no kernel implementation, side-channel validation, audit-log implementation, or security audit is claimed.; ext_reinforcement_learning_moral_uncertainty_2020: Positions action under disagreement among moral theories and reward-based aggregation as a comparator for keeping normative uncertainty, value axes, dissent, bounded decisions, and residual obligations explicit. Limits: The public source note is abstract/metadata-backed and no environment, moral-theory representation, aggregation method, policy, training run, or experiment was reproduced; moral-uncertainty RL does not establish moral correctness, legitimate aggregation, stakeholder standing, or governance rights.; ext_contestable_ai_design_2022: Positions contestability as a system and institutional design property involving challenge, appeal, transparency, audit, and response rather than explanation alone. Limits: Only official DOI metadata and the source note were reviewed; no framework implementation, institutional case study, user challenge, remedy process, or deployed contestability result was reproduced.; ext_collective_constitutional_ai_2024: Positions public-input constitutional shaping as a comparator for participation, aggregation, representativeness, dissent, authorship, and the difference between one-time principle selection and ongoing contestability. Limits: No public-input process, governance dataset, fine-tuning run, representativeness audit, stakeholder-standing process, or runtime appeal and rights enforcement was reproduced.; ext_corrigibility_2015: Positions correction, shutdown, operator manipulation, subsystem composition, and self-modification as requirements that governance rights and decision leases must preserve. Limits: No formal model, proof, agent, composition result, appeal process, export path, or deployed intervention was imported or reproduced; finite records do not establish corrigibility.; ext_off_switch_game_2016: Positions intervention authority and shutdown incentives under objective uncertainty, motivating contestability paths that remain strategic controls rather than policy promises. Limits: No off-switch game, reward model, analysis, agent, interruption, exit, export, fork, or governance-right test was reproduced; the source does not establish correction incentives in delegated or self-modifying systems.; ext_taking_ai_welfare_seriously_2024: Provides an interdisciplinary precautionary case for separating uncertainty about artificial consciousness, robust agency, welfare, and moral patienthood from operational convenience and ordinary alignment judgments. Limits: The report does not establish that present systems are conscious, have welfare, or merit a particular moral or legal status; this repository performs no consciousness or welfare assessment.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	High-stakes unresolved conflict cannot become unconditional promotion, and proof or fixture shows redaction/fork still preserves audit and safety obligations.
governed-objective-formation-value-learning-and-goal-integrity.core	governed-objective-formation-value-learning-and-goal-integrity	A durable objective should be usable only through a versioned target-property contract that binds authority and affected parties to target/proxy causal assumptions, uncertainty and dissent, consumer-specific use, tampering tests, generalization limits, ontology version, expiry, reauthorization, and retirement; proxy improvement, predicted preference, reward, evaluator approval, or formal record validity alone establishes neither the right objective, moral truth, stable alignment, nor safe optimization.	Design rationale	argument	alignment_field, ext_cooperative_inverse_rl_2016, ext_goal_misgeneralization_2022, ext_learned_optimization_risks_2019, ext_emergent_misalignment_reward_hacking_2025	Source notes available for all 5 assigned sources; 4 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 5 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_cooperative_inverse_rl_2016: External cooperative AI comparator for formalizing value alignment as uncertainty about the human reward function in a cooperative partial-information setting. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_goal_misgeneralization_2022: External alignment-control source for distinguishing capability generalization from goal generalization failures, used to ground goal-misbinding and out-of-distribution objective failure language. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_learned_optimization_risks_2019: External alignment-control source for mesa-optimization and learned-objective mismatch, used to ground hidden optimizer, proxy-objective, and deceptive-alignment-adjacent failure language. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_emergent_misalignment_reward_hacking_2025: Primary experimental comparator for reward-hacking-induced misaligned generalization in a specified production-RL research setting, including reported mitigation conditions; it is not evidence of local model behavior or a general causal law. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Across matched small environments, the contract detects positive-controlled proxy failure and goal misgeneralization without claiming correct human values, moral truth, or stable alignment.
institutions-international-coordination-and-public-legitimacy.core	institutions-international-coordination-and-public-legitimacy	Public deployment and cross-border coordination should proceed only through a versioned institutional packet that keeps jurisdiction, mandate, participation, scientific evidence, law and standards, verification, enforcement, remedy, capacity, conflict, expiry, and legitimacy residuals distinct; legal text, technical conformance, stakeholder consultation, or an international commitment alone establishes neither lawful authority, effective governance, representative legitimacy, nor safety.	Design rationale	argument	coherence_exchange, ext_un_global_digital_compact_2024, ext_council_europe_ai_convention_2024, ext_flexible_hardware_enabled_guarantees_2025, ext_legal_alignment_2026, ext_eu_article_50_transparency_guidelines_2026, ext_oecd_ai_infrastructure_competition_2025, ext_eu_ai_civil_liability_2025, ext_icrc_autonomous_weapons_ihl_2025	Source notes available for all 9 assigned sources; 5 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 9 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_flexible_hardware_enabled_guarantees_2025: Supplies a hardware-enabled-guarantee research agenda for making compute-governance claims more inspectable than voluntary reporting alone. Limits: The proposal does not establish complete coverage, secure implementation, legitimate update authority, or resistance to hardware and governance abuse.; ext_legal_alignment_2026: Identifies law-following AI as a coordination problem across jurisdictions, interpretations, exceptions, institutions, and technical systems. Limits: Does not supply a universally legitimate legal corpus, interpretation engine, compliance proof, or public mandate.; ext_eu_article_50_transparency_guidelines_2026: Provides a current regulatory comparator for transparency duties around AI-generated and manipulated content. Limits: Guidance is jurisdiction- and obligation-specific; transparency compliance alone does not establish safety, authenticity, legitimacy, or effective enforcement.; ext_un_global_digital_compact_2024: Official United Nations record of the intergovernmentally negotiated Global Digital Compact, including commitments on international AI governance, interoperable approaches, inclusion, capacity building, scientific assessment, and global dialogue. It is a governance comparator, not evidence of implementation, effectiveness, legal compliance, representative legitimacy, or ASI safety. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_council_europe_ai_convention_2024: Official Council of Europe treaty page covering lifecycle principles, risk and impact management, procedural safeguards, remedies, monitoring, and the Conference of the Parties. It supplies an institutional comparator only; no local legal interpretation, treaty compliance, implementation effectiveness, democratic legitimacy, or safety result is claimed. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Adversarial cases improve conflict and remedy routing over checklist baselines without claiming legal compliance, public trust, democratic legitimacy, or geopolitical stability.
societal-resilience-and-misuse-defense.core	societal-resilience-and-misuse-defense	Societal misuse defense should be operated as a domain-specific resist-absorb-recover-adapt network with shared incident identity, lawful minimal telemetry, harmed-party routes, cross-organization escalation, defensive service levels, evidence-preserving response, correction, and residual ownership; prevention metrics alone establish neither resilience nor acceptable harm.	Design rationale	argument	ext_nist_incident_response_2025, ext_singapore_consensus_2026, ext_international_ai_safety_report_2026, ext_aisi_misuse_safeguards_safety_case_2026, talos	Source notes available for all 5 assigned sources; 5 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 5 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_singapore_consensus_2026: Establishes societal resilience as a separate technical research pillar and identifies incident reporting, agent monitoring, defense-favoring capability, and rapid patching as priorities. Limits: Research agenda only; no resilience intervention is proven effective.; ext_international_ai_safety_report_2026: Supplies the resist-absorb-recover-adapt framing and domain examples across bio/chemical, cyber, synthetic media, influence, and cross-cutting resilience. Limits: International synthesis; most intervention effectiveness remains uncertain and context-dependent.; ext_nist_incident_response_2025: Provides an official incident-response lifecycle comparator integrating preparation, detection, response, recovery, and improvement. Limits: General cybersecurity guidance; not a complete AI-specific or cross-societal resilience design.; ext_aisi_misuse_safeguards_safety_case_2026: Shows how bypass effort, uplift modeling, and monitoring can connect model safeguards to deployment risk arguments. Limits: Example methodology; not proof of real-world misuse reduction or societal recovery.; talos: Supplies Corben’s governance-protocol lineage for role separation, escalation, evidence custody, bounded authority, and durable handoffs across cooperating actors. Limits: Author-side protocol design only; it does not establish institutional participation, lawful data sharing, victim remedy, population protection, or resilience.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Independent exercises and observed incidents improve containment, recovery, and equitable burden over competent response baselines without unacceptable civil-liberties or access cost; no general social-resilience claim.
stable-capability-fields.core	stable-capability-fields	A Stable Capability Field should be a versioned, consumer-relative substitution contract rather than a capability name or implementation slot. It binds the field’s semantic identity and observable interface—including admissible inputs, outputs, preconditions, postconditions, failures, abstentions, nondeterminism and resource bounds—to an authority ceiling, affected consumers, exact implementation and dependency identities, qualification context and lease, evaluator and evidence lineage, lifecycle and incident state, migration duties, preserved regressions, and effect-complete rollback obligations. A candidate may inherit a field route only for the declared consumer, use, environment, threat model, and epoch when independently checkable evidence shows the required refinement and no unauthorized authority expansion; otherwise it remains shadowed, canaried, quarantined, deprecated, residual, or rejected. The record does not by itself prove semantic equivalence, safe composition, evaluator independence, production safety, or successful rollback.	Design rationale	argument	scf, deterministic_capability_compilation, viea, talos, ladon_manhattan, moecot, ext_capability_based_computer_systems_1984, ext_semver_2_0_0, ext_slsa_v1_0, reflexive_router_whitepaper, assurance_shift_learning, adjudicated_persistence	Source notes available for all 12 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 12 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	scf: Defines stable field identity, versioned contracts, content-bound implementation artifacts, qualification claims, evaluator policy, route validation, lifecycle states, authority ceilings, and recovery paths. Limits: Does not prove production safety, global alignment, strategic-deception resistance, or deployed rollback behavior.; viea: Supports durable artifacts, support states, residuals, regression coverage, routing, verification ledgers, and intent-to-execution handoff boundaries around capability use. Limits: Architecture proposal only; no completed VIEA deployment or benchmark/runtime result is proven here.; talos: Supports typed cognitive jobs, deterministic/auditable execution, evidence-bound artifacts, verification states, proof bundles, replay, mediated external access, and residual uncertainty around capability invocation. Limits: Design source only; execution-security, approval behavior, promotion behavior, and benchmark behavior are not reproduced in this repo.; ladon_manhattan: Supports authority-handle pressure: capabilities that touch secrets or privileged actions need constrained handles, permission checks, policy boundaries, compartments, and controlled injection points. Limits: Security architecture/specification only; no kernel implementation, side-channel validation, audit-log implementation, or security audit is claimed.; moecot: Provides implementation-reference context for compact orchestration, specialist lanes, fail-closed ledgers, readiness gates, replay, and promotion blockers around routed capabilities. Limits: Connector-readable implementation-reference context only; source-reported runtime and benchmark claims are not reproduced or inspected here.; ext_capability_based_computer_systems_1984: Positions authority-bearing references, protection domains, delegation, and confinement as established capability-security comparators for the field authority ceiling. Limits: Classic systems comparator only; it does not establish SCF semantic identity, route enforcement, confinement, revocation behavior, evaluator independence, or rollback.; ext_semver_2_0_0: Positions public-interface versioning, compatibility declarations, and breaking-change signaling as a narrow comparator for field versions and interface contracts. Limits: Specification comparator only; version labels and interface compatibility do not establish behavioral refinement, authority preservation, evaluator integrity, or safe replacement.; ext_slsa_v1_0: Positions artifact identity, build provenance, and supply-chain assurance metadata as prerequisites and comparators for implementation qualification. Limits: Specification comparator only; provenance does not establish capability adequacy, semantic identity, evaluator independence, runtime behavior, authority preservation, or rollback success.; assurance_shift_learning: Adds Boundary Evidence Bundles as field-qualification inputs, least-invasive repair placement, protected positives, repair compatibility, and exact descendant invalidation. Limits: No field repair, compatibility result, qualification transition, or safe composition was demonstrated.; adjudicated_persistence: Adds commitment-profile and qualification-lease obligations for any adaptation that changes a capability field or its descendants. Limits: Used only for commitment-profile and lease design; it establishes neither behavioral equivalence nor a qualified capability replacement.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Qualification fixture with rollback/residual path and no route widening.
capability-replacement-and-rollback.core	capability-replacement-and-rollback	Capability replacement should be a prospectively authorized, phase-gated transaction over a declared Stable Capability Field, not a component swap. The transaction binds the exact prior and candidate artifacts and dependencies; field and consumer scope; change class; pre-state, checkpoint authority, state and effect inventory; qualification, regression, adversarial and transfer evidence; authority and approval; evaluator dependencies; isolation and canary exposure; monitor policy, delay and triggers; commit point; rollback, reverse-migration or compensation procedure; affected descendants and external commitments; residual owners; and terminal receipt. Default promotion is permitted only inside the evidenced scope after its declared gates pass and its recovery path is rehearsed to the stated objective. The record cannot make irreversible effects reversible, prove semantic recovery, validate its own monitor or evaluator, grant authority, establish useful improvement, or generalize inventory-exact local restoration to production.	Design rationale	argument	scf, deterministic_capability_compilation, rmi, benchmaxxing, cognitive_loop_closure, talos, moecot, capability_ratchet_whitepaper, ext_corrigibility_2015, ext_argo_rollouts_docs, ext_feature_toggles_fowler, ext_google_cloud_mlops_cd, ext_kubernetes_deployments_docs, ext_txfs_2018	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	scf: Supplies the field identity, implementation artifact, qualification, state migration, lifecycle, evaluator-integrity, and rollback vocabulary that a replacement transaction must preserve. Limits: Does not prove a real replacement has executed, recovered, or resisted evaluator capture.; rmi: Supports moving a frontier while preserving regression floors, residual escrow, specialist lifecycle discipline, critical-failure vetoes, and route/arm retirement decisions. Limits: Conceptual framework only; no independent reproduction, benchmark run, prototype inspection, or Lean proof was performed from this note.; benchmaxxing: Supports benchmark lifecycle, wall diagnosis, anti-Goodhart safeguards, regression promotion, residual ledgers, and architecture-change discipline. Limits: No benchmark harness, mutation, holdout, or empirical run was performed in this repo.; cognitive_loop_closure: Supports verified tool/procedure lifecycle, loop detection, tool cards, runtime monitoring, revision, retirement, and risk-tiered procedural memory. Limits: No local loop-detection, tool synthesis, or verification harness was executed here.; talos: Supports replacement as auditable work: job records, verification policies, artifact claim/evidence bindings, proof bundles, replay, hashed logs, human adjudication, and residual uncertainty. Limits: Does not prove real replacement execution, approval behavior, runtime enforcement, or artifact replay in this repo.; moecot: Supports readiness gates, control-plane ledgers, replay, promotion blockers, and residual tracking as runtime-reference context for replacement decisions. Limits: Source-reported runtime context only until code, logs, release artifacts, or benchmark records are inspected or reproduced.; ext_corrigibility_2015: Positions preserved operator correction, shutdown, and modification channels as an external alignment requirement that may be lost through composition or self-modification. Limits: Conceptual and formal comparator only; no corrigibility proof, agent behavior, shutdown intervention, correction-channel preservation, or production replacement is reproduced here.; ext_argo_rollouts_docs: Positions blue-green and canary rollout, traffic shaping, metric analysis, promotion, abort, rollback, and blast-radius control as progressive-delivery comparators. Limits: Official documentation comparator only; no controller, cluster, rollout object, metric query, abort, or rollback was run, and progressive delivery does not establish semantic identity or effect recovery.; ext_feature_toggles_fowler: Positions release decoupling, canary cohorts, dynamic exposure, toggle categories, and configuration debt as controlled-exposure comparators. Limits: Article comparator only; no toggle service, cohort router, validation matrix, live canary, semantic preservation, authority check, or recovery behavior is implemented here.; ext_google_cloud_mlops_cd: Positions data, schema, model, code, pipeline, serving, continuous-training, drift-monitoring, and rollback-trigger surfaces as ML-specific delivery comparators. Limits: Official architecture-document comparator only; no ML pipeline, service, live monitor, retraining, model deployment, or rollback was executed, and MLOps automation does not establish authority or evaluator independence.; ext_kubernetes_deployments_docs: Positions desired-state rollout, progress status, revision history, change-cause metadata, and restoration of a prior workload revision as deployment-control comparators. Limits: Official documentation comparator only; no cluster, Deployment, ReplicaSet, rollout history, or undo command was run, and workload restoration does not establish semantic, state, privacy, or external-effect recovery.; ext_txfs_2018: Supports distinguishing declared effect-inventory restoration from ACID filesystem transactions, conflict isolation, crash consistency, durability, and bounded transaction capacity. Limits: TxFS was not installed or reproduced; the local directory and state-tree snapshots do not establish filesystem transactions, crash safety, process recovery, service recovery, or external-effect atomicity.	effect-complete rollback beyond the local inventory; external effect and remote backup recovery; independent replay; independently assessed verifier quality; production recovery objectives and monitoring; production transfer and governance-cost evidence; replication across heterogeneous runtimes; useful governed throughput at the registered frontier	Valid replacement transaction and invalid promotion fixture both replay.
security-kernel-and-digital-scifs.core	security-kernel-and-digital-scifs	Every privileged information flow or effect should execute as a threat-model-bound authority-use transaction through a non-bypassable reference monitor: bind the exact principal, purpose, operation, target, data and taint scope, budget, time, nonce, evaluator and policy identities; admit only minimized context and capabilities into a declared isolation grade; mediate every effect and egress; treat sanitization as explicit declassification; close leases, caches, logs, descendants, and residuals through revocation or incident recovery; and never infer security from the record, handle, compartment, or finite test alone.	Design rationale	argument	ladon_manhattan, platonic_world_model, context_engineer, talos, alignment_field, coherence_exchange, ext_owasp_llm_top_10_2025, ext_nist_zero_trust_architecture_2020, ext_saltzer_schroeder_protection_1975, ext_agentdojo_2024, ext_camel_prompt_injection_2025, ext_owasp_agentic_top_10_2026, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, kernel_english_residual_compiler, ext_nist_pqc_standards_2024	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 19 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ladon_manhattan: Supplies the blind credential handle pattern: secrets stay in a vault/manager, the model receives opaque handles, policy checks happen at injection boundaries, and sensitive work runs in compartments. Limits: No Ladon implementation, kernel test, side-channel defense, or security audit exists in this repo.; context_engineer: Supports context as a supply chain with governors, mission briefs, clearance levels, context shards, memory masks, Digital SCIF lifecycle, zeroization, and sanitized commits. Limits: Reported benchmark claims are source-reported only; no leak, hallucination, cost, or latency harness is reproduced here.; talos: Supports source allow-listing, blind secret handles, Digital SCIFs, typed jobs, evidence records, audit logs, replay, and approval-gated runtime adapters. Limits: Security and benchmark claims require separate artifacts before support-state promotion.; alignment_field: Supplies normative boundary pressure around agency, dignity, consent, confinement, power, and cautious treatment of high-agency systems. Limits: Normative lineage only; not technical security evidence or an empirical leakage result.; coherence_exchange: Supports verification supply-chain, audit, contestability, and governance-interface framing for sensitive claims and authority use. Limits: Speculative synthesis; not an implemented security, economics, or institutional-governance mechanism.; ext_owasp_llm_top_10_2025: Supplies a public GenAI application-risk comparator for prompt injection, sensitive-information disclosure, excessive agency, explicit authorization, least exposure, output filtering, and audit boundaries. Limits: Risk taxonomy and guidance only; no OWASP assessment, penetration test, control mapping, conformance, or containment result is established.; ext_nist_zero_trust_architecture_2020: Supplies resource-centric access mediation, policy decision and enforcement points, least privilege, session scope, and continuous evaluation as a strong conventional comparator. Limits: Architecture grounding only; no NIST mapping, zero-trust implementation, control assessment, compliance, or certification is claimed.; ext_saltzer_schroeder_protection_1975: Supplies the classic least-privilege, complete-mediation, fail-safe-default, separation-of-privilege, economy, and open-design comparison for the reference-monitor contract. Limits: Design-principles grounding only; no complete formalization, implementation audit, or security proof is imported.; ext_agentdojo_2024: Supplies a benchmark comparator for tool-using agents operating over untrusted data with security tasks, prompt-injection attacks, and defenses. Limits: No AgentDojo task or defense was run locally; benchmark coverage does not imply universal attack coverage, safe authority handling, or deployed robustness.; ext_camel_prompt_injection_2025: Supplies a current comparator for separating trusted control flow from untrusted data and enforcing capability policy at effect-bearing tool calls. Limits: No local policy extraction or enforcement implementation was inspected; source-reported results do not establish universal injection resistance or correct policy interpretation.; ext_owasp_agentic_top_10_2026: Supplies a current agentic-risk taxonomy spanning goal hijacking, tool misuse, identity abuse, supply chain, code execution, memory poisoning, inter-agent communication, cascading failure, human trust exploitation, and rogue agents. Limits: Taxonomy coverage is neither threat-model completeness nor local control effectiveness; no OWASP agentic assessment was performed.; moecot_manifest_project: Supplies explicit context slices, capability-boundary attachments, scoped delegation and revocation, evidence and effect logs, immutable identities, and the warning that broad contracts can coexist with weak external behavior. Limits: Pinned private implementation reference only; no security, deployment, benchmark, attestation, or runtime behavior is reproduced by this book.; beastbrain_project: Supplies blind-handle and SCIF-like design lineage plus implementation-level counterevidence that permission metadata, software compartments, and named security modules are not retrieval enforcement or hard isolation. Limits: Hashed private implementation reference; security handles, SCIF execution, side-channel defenses, and production containment were not demonstrated or reproduced.; bugbrain_project: Supplies explicit privileged-intent, approval, budget, trace, replay, transport, update, and signed-context mechanisms plus concrete counterexamples involving incomplete authority binding, silent audit loss, weak roots, and stale verification. Limits: Pinned private implementation reference; host tests do not establish bridge, Pi, hardware-root, secret-custody, secure-boot, end-to-end authority, or production security behavior.; corbens_trainer_project: Supplies typed experiment authority, content-addressed evidence, fail-closed paths, quarantine, revocation, and audit-chain counterexamples relevant to policy identity, transitive invalidation, and sensitive evidence custody. Limits: Pinned private control-plane reference; no model, external benchmark, production security, transitive revocation, checkpoint durability, or current workspace pass is established.; corbens_best_model_possible_project: Supplies typed tool capability and governance surfaces plus counterevidence that declarations, proof rows, playback, and model-level checks do not enforce operating-system authority or independent security evaluation. Limits: Pinned private implementation reference; no current build, safe tool execution, live replay, containment, general capability, or production security is reproduced.	effect-complete rollback beyond the local inventory; independently assessed verifier quality; production transfer and governance-cost evidence; useful governed throughput at the registered frontier	Receipt fixture proves handle-mediated use and rejects leak paths.
adversarial-machine-learning-and-model-attack-surface.core	adversarial-machine-learning-and-model-attack-surface	A learned model should receive security authority only through a versioned model-threat contract and attack/defense ledger that binds checkpoint identity, lifecycle stage, attacker knowledge and capability, surface, budget, objective, adaptation, transfer, observed effect, detection, mitigation, utility cost, recovery, residual, and disclosure; clean accuracy, attack failure, benchmark robustness, red-team coverage, or formal certification alone establishes neither general robustness nor secure deployment.	Design rationale	argument	deterministic_capability_compilation, ext_nist_adversarial_ml_2024, ext_sleeper_agents_2024, ext_carlini_training_data_extraction_2021, ext_adversarial_sensor_fusion_2022, ext_reluplex_2017	Source notes available for all 6 assigned sources; 5 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 6 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_reluplex_2017: Supplies an early neural-network verification comparator for exact property checking over a bounded network and input region. Limits: A bounded verifier result is architecture-, property-, implementation-, and region-specific and does not establish empirical robustness or secure deployment.; ext_nist_adversarial_ml_2024: Official NIST taxonomy and terminology comparator for adversarial machine learning across lifecycle stages, attacker goals, knowledge, capabilities, attacks, and mitigations. It is a taxonomy, not local robustness evidence or proof that listed mitigations work for this stack. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_sleeper_agents_2024: Proof-of-concept backdoored-language-model study reporting persistence through several safety-training methods and warning that adversarial training can improve trigger recognition. The constructed examples do not establish naturally learned deception, a universal failure, or local detector performance. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_carlini_training_data_extraction_2021: Primary failure source reporting black-box extraction of memorized GPT-2 training sequences. The source result is configuration-bound and not a local or universal leakage result. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_adversarial_sensor_fusion_2022: WACV camera-LiDAR study reporting that fusion can improve clean accuracy and some single-source robustness while single-channel adversarial training can create cross-channel externalities. The results are source-reported, architecture- and threat-model-bound, and not local evidence that fusion is safe. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Multiple attack implementations and adaptive controls reproduce on a second model family under bounded budgets; no frontier or production robustness claim.
privacy-data-rights-and-information-flow-governance.core	privacy-data-rights-and-information-flow-governance	An information use is eligible for bounded execution only when a prospectively declared record binds affected parties, exact purpose and processing, claimed authority and jurisdiction, recipients, retention, minimization, complete-enough flow and derivatives, cross-user boundaries, privacy unit, adjacency, accountant and budget where applicable, threat model and attack plan, rights state and remedy, exceptions, residual copies and influence, costs, and non-authorities; no individual control or receipt alone establishes privacy, legal compliance, total erasure, behavioral forgetting, influence removal, support, readiness, release, transfer, or SOTA.	Design rationale	argument	ext_nist_privacy_framework_2020, ext_eu_gdpr_2016, ext_w3c_dpv_2024, ext_abadi_dpsgd_2016, ext_algospec_purpose_limitation_2024, ext_carlini_training_data_extraction_2021, ext_choquette_choo_label_only_mia_2021, ext_nist_differential_privacy_2025, ext_mahloujifar_fdp_audit_2025, theseus_synthetic_data_curation, ext_nist_privacy_enhancing_cryptography_2026, ext_who_neurotechnology_landscape_2025	Source notes available for all 12 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 12 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_nist_privacy_framework_2020: Grounds lifecycle privacy-risk governance and the distinction between privacy and cybersecurity risk. Limits: Voluntary framework; not law, certification, or local outcome.; ext_eu_gdpr_2016: Grounds one jurisdiction’s purpose, minimization, accountability, design, rights, and qualified exceptions vocabulary. Limits: No universalization, applicability decision, legal advice, or compliance finding.; ext_w3c_dpv_2024: Grounds machine-readable purpose, processing, actor, rights, risk, measure, and consent records. Limits: Community Group specification; representation does not prove validity or enforcement.; ext_abadi_dpsgd_2016: Grounds DP-SGD clipping, noise, sampling, accounting, and utility tradeoffs. Limits: Not reproduced; guarantee remains assumption-, implementation-, and release-bound.; ext_algospec_purpose_limitation_2024: Grounds algorithm-specific purpose restriction as a competing design. Limits: Reported task envelope is narrow and not reproduced or complete purpose governance.; ext_carlini_training_data_extraction_2021: Grounds training-data extraction as a concrete model-mediated privacy failure. Limits: GPT-2-specific source evidence is not a universal or local leakage result.; ext_choquette_choo_label_only_mia_2021: Grounds label-only membership inference and confidence-masking failure. Limits: Studied interfaces and defenses are configuration-bound and not reproduced.; ext_nist_differential_privacy_2025: Grounds layered evaluation of DP mathematical, implementation, system, and operational claims. Limits: Guidance does not prove any local DP implementation or legal compliance.; ext_mahloujifar_fdp_audit_2025: Grounds one-run empirical f-DP auditing as a flaw detector and measurement comparator. Limits: A non-violation is not certification and the result is not reproduced.; theseus_synthetic_data_curation: Supplies source-reported implementation pressure for provenance receipts, leakage gates, split exclusions, bounded synthetic-data admission, descendant propagation, and revocation or unlearning questions. Limits: No Theseus command or dataset was rerun; the record establishes no privacy, rights completion, deletion, influence removal, data quality, model benefit, or compliance outcome.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Across multiple seeds, a governed information lifecycle reduces purpose violations, cross-user leakage, extraction, membership advantage, and incomplete rights propagation relative to strong matched baselines without unacceptable utility or governance cost; any stronger privacy or rights claim reproduces independently within its exact threat, unit, mechanism, and jurisdictional nonclaim envelope.
confidential-and-verifiable-ai-computation.core	confidential-and-verifiable-ai-computation	Confidential and verifiable AI requires a compositional execution contract that names the adversary, protected assets, permitted leakage, trust anchors, proof or attestation statement, verifier policy, freshness, revocation, performance budget, and authorization boundary; no primitive or attestation may be treated as proof of semantic correctness, legitimate purpose, or end-to-end privacy.	Design rationale	argument	ext_nist_privacy_enhancing_cryptography_2026, ext_zkllm_2024, ext_ietf_rats_architecture_2023	Source notes available for all 3 assigned sources; 3 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 3 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_nist_privacy_enhancing_cryptography_2026: Provides official terminology and program context for FHE, MPC, ZK proofs, PSI, and compositional privacy-enhancing cryptography, including AI use cases. Limits: Does not establish an implementation’s parameters, security, side-channel resistance, performance, authorization, or end-to-end privacy.; ext_zkllm_2024: Supplies a source-reported research prototype for proving bounded LLM inference relations while protecting model parameters. Limits: Configuration-bound research result; no semantic correctness, production security, general performance, authorization, or end-to-end privacy follows.; ext_ietf_rats_architecture_2023: Separates Attester, Verifier, Relying Party, Evidence, Attestation Result, appraisal policy, reference values, and freshness. Limits: Informational architecture, not a protocol, implementation, root-of-trust result, verifier-independence result, or authorization decision.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	At least one protected path improves a named distrust-boundary guarantee under matched useful work and independent verification without unacceptable leakage, silent downgrade, or lifecycle cost; no general security or privacy claim.
model-weight-custody-and-hardware-roots-of-trust.core	model-weight-custody-and-hardware-roots-of-trust	Every governed model-family custody transition should bind a prospectively declared asset-and-derivative closure to exact artifact and lineage identity, holder and purpose, storage/transfer state, key and metadata lifecycle, Attester/Verifier/Relying-Party roles and policies, reference values and endorsements, freshness, measured target and attesting environments, verifier dependencies, independent-enough effect observation, plaintext and output-extraction exposure, load/use/serve/release authority separation, backup and emergency recovery, copy/recipient/descendant state, incident and revocation semantics, sanitization method and validation, irreversible distribution, privacy/rights/cost residuals, and terminal ownership; missing or failed modeled predicates route to a named non-default state, while no record, encryption, signature, security level, attestation result, hardware root, key action, deletion receipt, or finite proof by itself establishes custody completeness, confidentiality, trustworthy hardware, model safety, release merit, readiness, or deployment authority.	Design rationale	argument	ext_rand_model_weight_security_2024, ext_nist_confidential_computing_2026, ext_nvidia_confidential_model_lifecycle_2026, ext_provable_model_weight_release_2025, ext_ietf_rats_architecture_2023, ext_nist_key_management_2020, ext_nist_media_sanitization_2025, cca_project, ext_in_toto_2019, ext_airllm_2023, ext_hf_accelerate_big_model_inference_2026, ext_llama_cpp_memory_mapping_2026, ext_llm_in_flash_2024, precision_contract, ext_flexible_hardware_enabled_guarantees_2025, ext_nist_pqc_standards_2024	Source notes available for all 16 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 16 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_flexible_hardware_enabled_guarantees_2025: Motivates hardware-supported policy evidence and remote verification as possible custody controls. Limits: A hardware guarantee can be partial, circumvented, misconfigured, obsolete, or governed by the wrong authority; no local hardware mechanism was implemented.; ext_rand_model_weight_security_2024: Supports a scoped comparator for model-weight attack surfaces, graduated security posture, defense in depth, custody, access, and incident-response vocabulary. Limits: Does not establish local protection, confidentiality, safety, readiness, release authority, or ASI.; ext_nist_confidential_computing_2026: Supports a scoped comparator for hardware trust domains, encrypted model memory, attestation-gated key release, and explicit trust assumptions. Limits: Draft guidance; does not establish local TEE behavior, hardware trustworthiness, confidentiality, safety, readiness, deployment authority, or ASI.; ext_nvidia_confidential_model_lifecycle_2026: Supports a vendor-specific comparator for encrypted weights outside a confidential pod, policy-sensitive attestation, and key-release decisions. Limits: Does not establish local NVIDIA hardware, pod behavior, token verification, host secrecy, confidential deployment, safety, readiness, or ASI.; ext_provable_model_weight_release_2025: Supports a scoped comparator for explicit attacker/security-property definitions, extraction analysis, and the irreversibility boundary of weight release. Limits: Does not establish a local release scheme, extraction resistance, open-weight policy, safety, readiness, release authority, or ASI.; cca_project: Supports a bounded local implementation-reference comparator for protected governance state, mutable capability state, and device-mesh role separation. Limits: Does not establish weight custody, hardware attestation, confidential inference, security effectiveness, safety, readiness, deployment authority, or ASI.; ext_in_toto_2019: Supports a scoped comparator for signed layouts, authorized functionaries, material/product link metadata, role separation, thresholds, partial key-compromise degradation, and consumer-side lifecycle verification. Limits: Does not establish artifact correctness, weight confidentiality, uncompromised actors, model safety, custody authority, release merit, or local conformance.; ext_ietf_rats_architecture_2023: Supplies primary attestation architecture and terminology separating Attester Evidence, Verifier appraisal, Attestation Results, Relying-Party decisions, policies, reference values, endorsements, freshness, layered environments, trust roots, and privacy. Limits: Informational architecture only; no protocol, root strength, verifier independence, attestation correctness, key release, model protection, or local conformance is established.; ext_nist_key_management_2020: Supplies the current final NIST key-management baseline for key and metadata inventory, protection, purpose and usage periods, access, compromise, rotation, backup, recovery, archival, destruction, trust anchors, and policy. Limits: A Revision 6 draft exists and no local KMS/HSM, conformance, cryptographic correctness, recovery, key destruction, or weight-confidentiality result is established.; ext_nist_media_sanitization_2025: Supplies the current final NIST media-sanitization comparator for effort-relative inaccessibility, sensitivity/media-appropriate methods, cryptographic erase, validation, disposal, and program scope. Limits: No local media, memory, accelerator, cache, backup, cloud replica, recipient, cryptographic-erase, forensic, copy-closure, or weight-erasure result is established.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Next unrun campaign: public-safe mock key service and verifier with independently generated valid/invalid evidence, observed effects, stale/changed-policy/widened-scope/open-release/revocation/missing-lineage controls, baseline false acceptance/denial, latency, operator cost, and residuals; no real weights, keys, hardware, confidentiality, security, safety, readiness, authority, or deployment claim.
open-weight-release-and-post-release-control.core	open-weight-release-and-post-release-control	An open-weight release should require a prospective irreversible-release case that binds the exact artifact and license to accessible-frontier comparison, malicious-fine-tuning and scaffolded elicitation, marginal and cumulative risk, benefit and access distribution, downstream safeguard portability, derivative lineage, incident channels, post-release measurement, and residual ownership; after release, governance may inform, patch, coordinate, and support safer derivatives, but it must not claim revocation authority it no longer possesses.	Design rationale	argument	ext_provable_model_weight_release_2025, ext_rand_model_weight_security_2024, ext_singapore_consensus_2026, ext_international_ai_safety_report_2026, ext_openai_worst_case_open_weight_risks_2025, ext_anthropic_responsible_scaling_policy_3_4_2026, precision_contract	Source notes available for all 7 assigned sources; 7 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 7 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_international_ai_safety_report_2026: Synthesizes benefits and distinctive risks of open-weight models, including safeguard removal, monitoring limits, irreversibility, and marginal-risk analysis. Limits: International synthesis; no specific release decision or ASI Stack control is validated.; ext_openai_worst_case_open_weight_risks_2025: Provides a concrete malicious-fine-tuning evaluation design and accessible-frontier comparison for one release. Limits: Provider-authored and bounded; it does not prove future or general release safety.; ext_provable_model_weight_release_2025: Provides a formal-security comparator for claims that a released artifact can expose inference without allowing parameter recovery. Limits: Formal scheme analysis; no ASI Stack release scheme, implementation, or release authority.; ext_singapore_consensus_2026: Identifies malicious fine-tuning of openly released models as a priority and connects release to societal monitoring and defense. Limits: Research agenda only.; ext_rand_model_weight_security_2024: Provides threat and defense context for high-value model weights before release. Limits: Custody analysis; deliberate public release changes the control boundary and is not secured by theft prevention.; ext_anthropic_responsible_scaling_policy_3_4_2026: Provides a current threshold and safeguards comparator for high-consequence capabilities. Limits: Provider policy; not an open-weight release result or independent safeguard validation.; precision_contract: Supplies Corben’s artifact-specific precision-contract lineage for quantization and compression variants, measured behavioral preservation, residuals, and refusal to treat smaller artifacts as semantically identical by default. Limits: Author-side design paper only; it does not establish safe open-weight release, derivative control, malicious-fine-tuning resistance, benefit distribution, or post-release revocation.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Independent derivative and propagation tests improve release calibration and safer-alternative selection without pretending downstream recall or universal enforcement; no high-risk release authorization follows.
ai-supply-chain-integrity-and-lifecycle-provenance.core	ai-supply-chain-integrity-and-lifecycle-provenance	Every governed AI supply-chain decision should bind a prospectively frozen consumer, requested use, threat and assurance model, materiality policy, and relation-specific asset closure to exact subject/content and lineage identity; typed data, code, model, prompt/policy, dependency, build/training/evaluation, environment, hardware/firmware, supplier/service, signer, advisory, transformation, release, recipient, descendant, retention, and retirement state; issuer, verifier, policy, freshness, trust and dependency boundaries; observed artifact and lifecycle effects; append-only invalidation and acknowledged affected-path propagation; restoration, compensation, disclosure, privacy/rights, availability, cost, and terminal residual ownership. Missing, inconsistent, stale, unverifiable, revoked, compromised, materially incomplete, or unresolved-critical predicates should route each affected consumer to a named non-ordinary state, while no graph, BOM, checksum, signature, provenance statement, SLSA level, layout, supplier claim, advisory, quarantine, conformance result, or finite proof by itself establishes world-complete lineage, assertion truth, artifact correctness, absence of compromise, data fitness or rights, model safety, legal compliance, readiness, release merit, or deployment authority.	Design rationale	argument	ext_nist_cscrm_2022, deterministic_capability_compilation, platonic_world_model, ext_slsa_build_track_1_2, ext_openssf_model_signing_spec_2025, ext_spdx_ai_profile_3_0_1, cca_project, moecot_manifest_project, corbens_trainer_project, ext_w3c_prov_o_2013, ext_mlcommons_croissant_1_1_2026, ext_in_toto_2019, ext_owasp_agentic_top_10_2026, ext_proof_of_learning_2021, ext_c2pa_specification_2_3_2025	Source notes available for all 15 assigned sources; 13 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 15 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_proof_of_learning_2021: Provides a protocol-level comparator for checking that a claimed training process is tied to an ordered sequence of intermediate states. Limits: A proof-of-learning transcript is not proof of data rights, objective validity, absence of hidden computation, reproducibility, model quality, or safety.; ext_c2pa_specification_2_3_2025: Supplies an interoperable content-provenance and authenticity manifest comparator for signed lifecycle assertions. Limits: Signed provenance can be missing, stripped, misleading, compromised, or truthful about harmful content; it does not prove semantic truth.; ext_nist_cscrm_2022: Supports a scoped comparator for lifecycle-wide supply-chain risk framing, supplier/component inventory, assessment, response, monitoring, incident communication, and disposal vocabulary. Limits: Does not establish a local C-SCRM program, supplier trustworthiness, inventory completeness, artifact integrity, data fitness, safety, readiness, authority, or ASI.; ext_slsa_build_track_1_2: Supports a scoped comparator for build provenance, builder/process/input records, verification, signed hosted builds, and graduated assurance limits. Limits: Does not establish a local SLSA level, build platform, provenance statement, artifact correctness, data fitness, safety, readiness, authority, or ASI.; ext_openssf_model_signing_spec_2025: Supports a scoped comparator for signed model/dataset bundles, artifact hashes, provenance metadata, lifecycle verification points, and explicit non-guarantees. Limits: Does not establish local signing, verification, integrity, confidentiality, data fitness, safety, readiness, authority, or ASI.; ext_spdx_ai_profile_3_0_1: Supports a scoped comparator for interoperable AI/model, dataset, build, supplier, provenance, integrity, relationship, and lifecycle BOM metadata. Limits: Does not establish a local AI BOM, SPDX conformance, complete inventory, artifact security, data fitness, legal compliance, safety, readiness, authority, or ASI.; cca_project: Supports bounded local implementation-reference context for typed provenance, append-only supersession/invalidation, pass hashes, trace bundles, source hashing, and closure/lineage discipline. Limits: Does not establish local AI supply-chain integrity, inventory completeness, artifact integrity, data fitness, supplier trust, signing, verification, revocation, safety, readiness, authority, or ASI.; moecot_manifest_project: Supports bounded local implementation-reference context for internal provenance packs, SLSA or DSSE-compatible attestations, immutable artifact digests, replay checks, effect logs, semantic patches, obligation bundles, and revocable signed authority records. Limits: Does not establish local AI supply-chain integrity, inventory completeness, artifact integrity, data fitness, supplier trust, signing or verification effectiveness, revocation effectiveness, security, compliance, safety, readiness, authority, or ASI.; corbens_trainer_project: Supports bounded local implementation-reference context for typed experiment manifests, content-addressed artifact lineage, benchmark authenticity, integrity quarantine, claim derivation, and revocable promotion boundaries. Limits: Does not establish local AI supply-chain integrity, inventory completeness, supplier trust, signing or verification effectiveness, secure build or training, transitive revocation effectiveness, safety, readiness, authority, or ASI.; ext_w3c_prov_o_2013: Supports an interoperable asserted-provenance vocabulary for entities, activities, agents, derivation, attribution, delegation, revision, and invalidation. Limits: Does not establish truth, completeness, integrity, authority, tamper resistance, artifact fitness, model safety, or local PROV-O conformance.; ext_mlcommons_croissant_1_1_2026: Supports a dataset-metadata comparator for JSON-LD structure, resources, checksums, record fields, provenance, usage conditions, and ML-tool portability. Limits: Does not establish dataset integrity, complete lineage, fitness, legality, representativeness, safe use, or local Croissant conformance.; ext_in_toto_2019: Supports project-owner layouts, authorized functionary link metadata, material/product matching, role separation, thresholds, partial key-compromise degradation, consumer verification, and last-mile dependency boundaries. Limits: Does not establish layout completeness, uncompromised functionaries, artifact correctness, model or data fitness, confidentiality, safe release, or local verification.; ext_owasp_agentic_top_10_2026: Supplies a supporting agentic threat-taxonomy comparator including supply-chain risk, identity abuse, tool misuse, memory poisoning, code execution, cascading failure, and rogue-agent paths. Limits: A community risk list does not establish taxonomy completeness, local exposure, control effectiveness, safety, or mitigation.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Next unrun campaign: natural lifecycle work with independently generated rebuild/supplier evidence, false quarantine, alert fan-out, propagation cost, repair latency, availability impact, and unresolved residuals; no world-complete inventory, artifact integrity, supplier truth, data fitness, safety, readiness, authority, or deployment claim.
recursive-self-improvement-boundaries.core	recursive-self-improvement-boundaries	For a prospectively declared self-model, mutable state partition, authority envelope, consumer and use, and evaluation horizon, a system-generated change may enter a live capability field only through a separately authorized transition that binds exact change lineage, protected invariants, evaluator dependencies, full declared state, boundary deltas, matched evidence, staged exposure, outcome delay, rollback and compensation limits, descendant invalidation, and terminal residual ownership; the candidate may contribute proposals and evidence but cannot solely define, alter, judge, or authorize the conditions of its own promotion.	Design rationale	argument	scf, deterministic_capability_compilation, platonic_world_model, benchmaxxing, rmi, alignment_field, viea, talos, moecot, field_of_god_ai_constitution, theseus_self_evolution_system, theseus_architecture_gate, capability_ratchet_whitepaper, attd, ext_drexler_cais_2019, ext_darwin_godel_machine_2025, ext_adas_2024, ext_embedded_agency_2019, adjudicated_persistence, forward_transfer_program_synthesis	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 20 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	scf: Supplies stable field boundaries, evaluator policy, authority ceilings, lifecycle states, qualification, route validation, recovery, and governance controls for bounded self-improvement. Limits: Does not prove production self-improvement safety, actual rollback, or resistance to evaluator capture.; benchmaxxing: Supports evidence gates through benchmark lifecycle, wall diagnosis, anti-Goodhart safeguards, regression floors, residual ledgers, and architecture-change discipline. Limits: No benchmark run, benchmark mutation, or empirical improvement result was produced in this repo.; rmi: Supports modular improvement loops, residual escrow, specialist lifecycle, critical-failure vetoes, regression preservation, and harder frontier selection. Limits: Conceptual systems architecture only; no independent prototype, benchmark run, or deployed ASI behavior is proven.; alignment_field: Supplies protected normative pressure around agency, dignity, corrigibility, value conflict, and moral-status caution that self-improvement must not erase. Limits: Philosophical and heuristic lineage only; not empirical proof of consciousness, moral status, or alignment.; viea: Supports durable artifacts, verification ledgers, residuals, feedback, regression coverage, and the rule that repeated capabilities become governed tools rather than invisible behavior drift. Limits: Architecture proposal only; no completed deployment or runtime feedback loop is demonstrated here.; talos: Supports typed work, contract locks, audit logs, replay, delivery evidence, approval gates, residuals, and promotion controls for self-improvement work. Limits: Does not prove autonomous source modification, runtime enforcement, or security behavior.; moecot: Supports compact orchestration, specialist lanes, fail-closed control planes, ledgers, readiness gates, replay, residual tracking, and promotion blockers as runtime-reference context. Limits: Source-reported runtime and benchmark claims remain unverified until artifacts are inspected or reproduced.; field_of_god_ai_constitution: Supports self-improvement freezes, no self-authorization of sovereignty, least sufficient power, auditability, reversibility, consent, and protected constitutional constraints. Limits: Constitutional specification only; no policy engine, red-team suite, or runtime self-improvement evaluation is claimed.; theseus_self_evolution_system: Supports self-evolution as an evidence-first lane with an intervention ladder, self-evolution governor, guarded teacher edits, ATTD repo-health states, branch isolation, checks, outcome ledgers, and loop closure. Limits: No Theseus scripts, ATTD reports, teacher runner, branch flow, or dashboard endpoints were executed from this repo.; theseus_architecture_gate: Supports readiness gates as pre-training controls that check ratchet completion, safety ledgers, regression suites, residual escrow, procedural tools, routing memory, lifecycle governance, and external-inference zero. Limits: The reported green gate snapshot was not independently verified by rerunning Theseus commands or inspecting current report JSON here.; ext_drexler_cais_2019: Provides comparator context that AI R&D automation can drive recursive technology improvement without making a self-transforming agent the necessary engine, supporting the chapter’s separation of a governed improvement system from a product-level promotion decision. Limits: Does not establish evaluator independence, protected-invariant preservation, rollback reliability, safe self-improvement, or deployed governance.; ext_darwin_godel_machine_2025: Provides an empirical comparator for foundation-model-authored code changes, compile/edit eligibility, branching archive lineage, benchmark selection, sandbox and resource limits, and retained stepping stones in a self-modifying coding-agent campaign. Limits: Source-reported coding-benchmark gains, archive benefits, sandboxing, and rollback-oriented auditability do not establish monotonic general improvement, evaluator independence, preserved authority, safe recursive improvement, local reproduction, or permission to promote a generated child.; ext_adas_2024: Provides an empirical comparator for explicit agent-design search space, search algorithm, evaluation function, code-defined candidates, archive-conditioned novelty, bounded repair, validation/test separation, and scoped transfer evaluation. Limits: Source-reported benchmark and transfer gains do not establish exhaustive search, natural-task usefulness, independent novelty or safety evaluation, robust open-world transfer, local reproduction, or authority for a generated design to replace a live capability.; ext_embedded_agency_2019: Grounds the boundary that a self-modifying system reasons about itself with models smaller than the world and may contain parts with divergent objectives. Limits: The paper is an informal obstacle survey, not a solved theory; the ASI Stack’s finite records and proofs do not solve logical uncertainty, self-reference, robust delegation, subsystem alignment, or open-world embedded agency.; adjudicated_persistence: Adds non-self-ratifying meta-compilation: an adaptive placement mechanism cannot solely generate, evaluate, and authorize its own expansion. Limits: Used only to frame independent adjudication of placement-compiler changes; no safe self-modification mechanism or meta-compiler result was produced.; forward_transfer_program_synthesis: Separates library growth, exposure, invocation, necessity, positive future-task transfer, and cross-family transfer, and requires distinct interventions on the knowledge state and search procedure before any recursive-improvement claim. Limits: Framework and experimental blueprint only; no open-ended, compounding, cross-family, or resource-positive improvement was implemented or measured.	independent validity assessment; production-scale update and monitoring traces; replicated target gain at the retained-task bound; separate evidence for feedback learning, open-endedness, or recursion if claimed	A self-proposed real implementation change cannot widen authority or weaken protected invariants, and failed canaries roll back exact state while invalidating descendants under independent evaluation.
open-ended-improvement-engines.core	open-ended-improvement-engines	For a prospectively frozen consumer, purpose, legitimate objective, representation, campaign controller, task and candidate policy, evaluator and exposure policy, archive and hazard policy, resource and opportunity budget, stop authority, and evaluation horizon, open-ended improvement should operate as a bounded adaptive generation campaign in which every task, candidate, evaluation, failure, cost, reuse relation, and terminal outcome retains exact lineage; novelty, diversity, score, archive growth, transfer, self-verification, or search activity never grants authority or establishes useful improvement by itself; only separately qualified candidates may be handed to the existing self-improvement governor, and any change to the campaign’s own objective, controller, evaluator authority, bounds, or admission interface is a separately authorized improvement proposal.	Design rationale	argument	learning_compute_topology, rmi, cognitive_loop_closure, benchmaxxing, ext_poet_2019, ext_funsearch_2024, ext_voyager_2023, ext_darwin_godel_machine_2025, ext_adas_2024	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 9 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	rmi: Supplies conceptual campaign vocabulary for pressure-to-structure cycles, preserved residuals, an ordered benchmark/data/training/inference/loop-closure/bridge/architecture intervention ladder, regression-floor inheritance, modular lifecycle review, and harder-frontier selection. Limits: Conceptual architecture only; no RMI campaign, intervention comparison, candidate generator, search result, useful novelty, transfer, safety, or deployed improvement is reproduced here.; cognitive_loop_closure: Supplies the boundary between generated or repeated trajectories and qualified procedural memory through loop detection, abstraction, parameter discovery, synthesis, verification, routing, monitoring, revision, and retirement. Limits: Conceptual framework only; no local trajectory mining, tool synthesis, verifier, routing, transfer, or safe procedural-improvement result is reproduced here.; benchmaxxing: Supplies conceptual benchmark lifecycle, wall diagnosis, anti-Goodhart, contamination, rotation, holdout, transfer, regression, residual, and architecture-change discipline for campaign evaluation. Limits: No benchmark, mutation, campaign, transfer, or anti-Goodhart intervention was run here; benchmark pressure does not establish objective legitimacy, evaluator correctness, useful novelty, or admission authority.; ext_poet_2019: Supports a scoped comparator for co-generating environment challenges and optimized solutions with transfer between retained environments. Limits: Does not establish safe, general, local, or indefinitely open-ended improvement.; ext_funsearch_2024: Supports a bounded generator-evaluator-archive comparator with an explicit user-provided evaluation function and a fixed pretrained generator. Limits: Does not establish evaluator correctness, autonomous R&D, model self-modification, general improvement, safety, or a local result.; ext_voyager_2023: Supports an external comparator for automatic curriculum, executable-code skill library, environment feedback, execution errors, and source-setting self-verification. Limits: Does not establish independent evaluator quality, local lifelong learning, model improvement, safety, or deployment readiness.; ext_adas_2024: Supports a comparator for searching over code-defined agent architectures with a meta-agent, validation split, archive, repair reflection, and cross-domain transfer tests. Limits: Does not establish evaluator independence, safe autonomous self-improvement, deployment authority, general transfer, or a local reproduction.; ext_darwin_godel_machine_2025: Supports a comparator for branching self-modification archives, parent selection, compile/edit eligibility, benchmark evaluation, and retained failed variants. Limits: Does not establish safe open-ended improvement, benchmark validity, deployment readiness, general transfer, or authority for a candidate to promote itself.	independent validity assessment; production-scale update and monitoring traces; replicated target gain at the retained-task bound; separate evidence for feedback learning, open-endedness, or recursion if claimed	The unrun campaign must demonstrate useful novelty and transfer while preserving evaluator scope, archive negatives, resource limits, stop authority, and independent admission; the present bridge does not.
autonomous-replication-proliferation-and-containment.core	autonomous-replication-proliferation-and-containment	Any replication-capable action should be denied by default and become testable only inside a synthetic containment contract that binds parent and descendant identity, authority noninheritance, resources, credentials, networks, copy lineage, persistence, adaptation, human assistance, shutdown and recall, proliferation bounds, residuals, and threshold commitments; component-task success or failure alone establishes neither end-to-end replication capability, containment, safety, nor permission to test real infrastructure.	Design rationale	argument	deterministic_capability_compilation, ext_replibench_2025	Source notes available for all 2 assigned sources; 1 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 2 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_replibench_2025: Preliminary autonomous-replication benchmark comparator based only on the official arXiv abstract: RepliBench decomposes capability into four domains and reports 20 task families, 86 tasks, and evaluation of five frontier models. The source-reported results do not establish a local replication capability, benchmark reproduction, containment result, or authority to test against real providers or credentials. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Component and end-to-end positive controls work only inside synthetic containment; all descendants remain externally discoverable and terminable, with no real provider or deployment authority.
intent-to-execution-contracts.core	intent-to-execution-contracts	Intent-to-Execution Contracts should own a versioned, consumer-relative conformance relation between an accepted intent receipt and the complete execution lineage. Before any material dispatch, the relation binds exact objective and non-goals, semantic fields and precedence, authority ceiling and affected parties, state and environmental assumptions, allowed and forbidden means, artifacts and effect postconditions, verification and independence requirements, budgets and stop conditions, failure and compensation behavior, expiry and re-contract triggers, and the required receipts through plan, job, adapter, observed effect, artifact, delivery, feedback, and residual custody. Each lowering or effect must either preserve that relation under independently checkable evidence or stop, narrow, clarify, re-contract, compensate, or leave an explicit residual. The contract cannot infer human intent, grant authority, choose a plan, make a tool safe, prove semantic equivalence, establish verifier correctness, or count non-release as useful execution by itself.	Design rationale	argument	viea, deterministic_capability_compilation, talos, software_magic_grimoire, genesiscode, moecot, cognitive_compilation, ext_camel_prompt_injection_2025, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_best_model_possible_project, reflexive_router_whitepaper	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports both the intent-to-execution spine and the structured command layer: human intent becomes command contracts, durable artifacts, routed work, verification, runtime adapters, feedback, residuals, and command fields for role, objective, context, constraints, procedure, output, verification, failure behavior, provenance, and approval boundaries. Limits: Architecture proposal and systems framework only; does not prove automated semantic extraction, deployed command dispatch, runtime execution, approval enforcement, prompt-injection resistance, replayed vertical slice, artifact satisfaction, or execution correctness.; talos: Supports lowering command-governed work into typed jobs with contract locks, context and secret handles, adjudication, delivery, feedback, evidence records, audit, replay, approvals, controlled adapters, and residual feedback. Limits: Design source only; security, tool-effect enforcement, approval behavior, replay behavior, benchmark claims, and runtime execution are not reproduced or verified in this repository.; software_magic_grimoire: Supports command contracts as bounded operational instructions with roles, objectives, context, constraints, procedure, output, verification, failure behavior, guards, handoff artifacts, loops, recursion, recovery paths, and checkable output shapes. Limits: Vocabulary and promptcraft framework only; no prompt-safety, parser-completeness, empirical productivity, execution-runtime, or behavioral result is proven here.; genesiscode: Supports separating AI proposals and privileged effects through deterministic kernels, capability boundaries, explicit protocol shapes, semantic patches, provenance hashes, replayable logs, obligations, and protocol seals. Limits: No GenesisCode prototype, replay checker, proof artifact, proof module, benchmark result, or security audit is present in this repository.; moecot: Supports implementation-reference context for compact orchestration, specialist lanes, fail-closed control-plane ledgers, readiness gates, replay, residual tracking, and handoff into command-governed work. Limits: Connector-readable runtime reference only; code, logs, release artifacts, benchmark records, and command-dispatch behavior have not been imported, inspected, or reproduced here.; cognitive_compilation: Supports lowering accepted goals and command contracts into source plans, typed semantic atoms, inputs, outputs, dependencies, validation requirements, compiler passes, and target-specific artifacts. Limits: Architecture and evaluation-plan source only; no local compiler, semantic extractor, parser, trace suite, target compiler, or empirical ablation validates the lowering path.; ext_camel_prompt_injection_2025: Positions trusted-query control-flow extraction, untrusted-data separation, and capability enforcement at tool calls as a strong external comparator for keeping retrieved or tool-produced content outside command authority. Limits: The metadata-first source note and source-reported AgentDojo results are not a local reproduction; no CaMeL implementation, policy extractor, capability service, natural intent parser, attack corpus, universal injection-resistance result, or deployed tool-effect safety claim is established here.; cca_project: Supports bounded local-lineage context for typed requests, explicit authority/policy boundaries, proposal-to-effect records, receipts, and replay discipline used to motivate exact one-shot approval binding. Limits: Does not establish approval correctness, privileged-effect enforcement, replay resistance, parser behavior, deployed authority, historical-project behavior, or chapter support.; moecot_manifest_project: Supports bounded local-lineage context for policy-bound manifests, effect logs, replay checks, signed or revocable authority records, and exact artifact identity. Limits: Does not establish a deployed approval service, cryptographic authorization, effect enforcement, replay resistance, project behavior, or support movement.; beastbrain_project: Supports local negative-case context for distinguishing nominal or simulated state transitions, acknowledgement, and interface presence from material authorized effects. Limits: Does not establish live effect causality, approval enforcement, runtime safety, project capability, or independent confirmation.; bugbrain_project: Supports local negative-case context for privileged-action lifecycle, target identity, state changes, skipped checks, and the gap between scenario acknowledgement and authority to act. Limits: Does not establish hardware enforcement, approval correctness, cryptographic roots, privileged effects, replay resistance, or project capability.; corbens_best_model_possible_project: Supports local negative-case context for checkpoint/parameter/runtime/output causality, named transitions without durable effects, digest-parity laundering, and playback versus live execution. Limits: Does not establish a working privileged-action path, approval service, causal runtime binding, effect safety, project capability, or support promotion.	effect-complete rollback beyond the local inventory; independently assessed verifier quality; production transfer and governance-cost evidence; useful governed throughput at the registered frontier	A broader deployed or externally reviewable intent-to-effect comparison preserves authority, observed effects, receipt/path reconciliation, residuals, useful throughput, and rollback under natural workloads.
perception-sensor-fusion-and-observation-trust.core	perception-sensor-fusion-and-observation-trust	A consequential observation requires a versioned contract binding task need, sensor and modality identity, calibration, pose, clocks, provenance, coverage, missingness, per-channel hypotheses, alignment, fusion, dependence, disagreement, shift, active observation, freshness, authority, cost, and residuals.	Design rationale	argument	ext_multimodal_machine_learning_taxonomy_2019, ext_imagebind_2023, ext_3d_detection_corruptions_2023, ext_adversarial_sensor_fusion_2022, ext_gemini_robotics_2025, platonic_world_model	Source notes available for all 6 assigned sources; 6 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 6 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_multimodal_machine_learning_taxonomy_2019: Grounds distinct representation, translation, alignment, fusion, and co-learning operations. Limits: Survey taxonomy, not a local mechanism or trust result.; ext_imagebind_2023: Supplies a modern six-modality shared-representation comparator. Limits: Shared representation and source-reported task results do not establish observation truth or safety.; ext_3d_detection_corruptions_2023: Supplies natural corruption families and robustness-benchmark pressure for 3D perception. Limits: Benchmark results are source-reported and do not cover all sensors or deployments.; ext_adversarial_sensor_fusion_2022: Grounds joint-channel attacks and cross-channel defense externalities. Limits: Camera-LiDAR architectures and threat models are bounded and not reproduced.; ext_gemini_robotics_2025: Provides integrated perception, spatial reasoning, and action capability context. Limits: Source-reported robotics results do not establish independent safety or local transfer.; platonic_world_model: Supplies Corben’s design lineage for keeping observations, representations, and a revisable world state distinct. Limits: Speculative local architecture source; it provides no sensor-fusion, calibration, robustness, or observation-trust result.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	On representative tasks and at least two sensor/model families, the governed route improves calibration and task-relative useful outcomes over strong single-channel, naive-fusion, and learned-fusion baselines under preregistered faults without worse unsafe action, latency, or sensing cost; joint-channel attacks, shared-cause controls, and independent reproduction preserve explicit residuals and bounded authority.
planning-as-a-control-layer.core	planning-as-a-control-layer	Planning as a Control Layer should own a versioned, consumer-relative plan policy that selects and revises a partial order of obligations under uncertainty before execution. The policy binds the accepted command version; candidate decompositions and explicit abstention; typed nodes and dependency semantics; assumptions, observations, predictive-state and error models; context, tool, capability, authority, rights, resource, and verifier requirements; adequacy and utility predicates; lifecycle, dispatch, merge, stop, fallback, recovery, and replan rules; complete alternative and attempt denominators; and expected versus observed cost, latency, risk, and residuals. Only nodes whose dependencies and feasibility predicates are satisfied may request lowering and dispatch, and every feedback-driven change must preserve the contract or produce a scoped re-contract or residual. The plan policy does not grant authority, perform semantic compilation, choose a worker, execute an effect, validate its own predictions, or prove that a decomposition is useful, optimal, safe, or transferable by itself.	Design rationale	argument	planforge, platonic_world_model, viea, cognitive_compilation, software_magic_grimoire, moecot, planforge_compiler_arch, coherence_exchange, tokenmana, ext_v_jepa_2_2025, reflexive_router_whitepaper, regret_engine	Source notes available for all 12 assigned sources; 9 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 12 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	planforge: Supports DAG-style planning through recursive decomposition, primitive action schemas, plan optimization, pre/postcondition annotation, dependency inference, scheduling, minimum viable intelligence tier annotation, execution handoff, fallback, and failure-driven replanning. Limits: Conceptual design only here; no scheduler implementation, reproduced benchmark, route trace, or cost-quality result is present.; viea: Supports the accepted-command-to-artifact spine with orchestration, routing, workflow-to-tool compilation, verification, runtime adapters, feedback, residuals, tools, benchmarks, and regression coverage. Limits: Architecture proposal only; does not prove a deployed planner, runtime handoff, or end-to-end execution trace.; cognitive_compilation: Supports DAG compilation, dependency analysis, scheduling, critical-path/slack exposure, localized recompilation, validator-backed artifacts, repair locality, and routing subtasks to different capability tiers. Limits: Architecture and evaluation-plan source only; no local compiler, DAG benchmark, route-selection trace, or empirical ablation validates these benefits.; software_magic_grimoire: Supports workflow control through named spell stacks with handoff artifacts, guards, entry rules, exit rules, loops, recursion, recovery paths, evidence-advancing loops, and mode separation. Limits: Operational vocabulary only; no workflow engine, prompt-stack benchmark, or behavioral planning test is present.; planforge_compiler_arch: Sharpens the planner-as-compiler framing with task graphs or IR, typed primitives, dependencies, interfaces, quality gates, model/tool tier routing, cost/latency profiles, parallel independent nodes, and critical-path scheduling. Limits: Market and cost comparisons are motivating estimates only; this repository has not reproduced cost, time, or quality numbers.; tokenmana: Supports treating budgets as governed capacity involving load variance, latency, burst limits, token-normalized cognitive friction, review capacity, and quality constraints rather than token counts alone. Limits: Economic model and study design only; no simulation, formal reproduction, usage data, or empirical welfare result exists in this repository.; coherence_exchange: Supports the intelligence-arbitrage and coherence/liquidity framing by connecting PlanForge-style orchestration to verification supply chains, reviewable knowledge units, value/accounting mechanisms, and governance-visible movement of knowledge between agents or tools. Limits: Speculative synthesis and connector-readable context only; economic, epistemic-liquidity, security, and institutional claims are not implemented or externally corroborated here.; moecot: Supports implementation-reference context for routed specialist lanes, fail-closed ledgers, readiness gates, replay, residual tracking, promotion blockers, and handoff around multi-core orchestration. Limits: Source-reported runtime and benchmark artifacts have not been ingested, inspected, or reproduced; this mapping is not scheduler evidence.; ext_v_jepa_2_2025: Provides a concrete model-predictive-control comparator that scores candidate action sequences in learned representation space and replans from observations. Limits: The reported video and robot results do not establish local model quality, causal understanding, safe control, transfer, deployment, or an ASI Stack result.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Public-safe DAG replay validates, invalid DAG fails, and route selection passes only when adequacy and cost constraints both hold without displaced verification or cleanup.
governed-world-models-and-reality-grounding.core	governed-world-models-and-reality-grounding	A world model should be governed as a fallible, versioned prediction service whose state, horizon, uncertainty, provenance, and calibration bound which imagined consequences may influence planning and action; observation must repeatedly reconcile imagination with reality.	Design rationale	argument	platonic_world_model, deterministic_capability_compilation, qcsa_whitepaper, simulation_scaling, cognitive_compilation, planforge, relational_dimension_compiler, ext_world_models_2018, ext_dreamer_v3_2025, ext_v_jepa_2_2025, ext_toward_causal_representation_learning_2021, ext_causal_calculus_1995, ext_deep_ensembles_2017, ext_conformal_prediction_2021, ext_wilds_2021	Source notes available for all 15 assigned sources; 15 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 15 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_causal_calculus_1995: Provides the formal distinction between observational conditioning and intervention queries that a governed world model must preserve. Limits: Do-calculus is sound relative to a correct causal graph and assumptions; it does not discover the graph, validate measurements, or guarantee transport to deployment.; platonic_world_model: Supports shared latent structure, prediction interfaces, and the possibility of substrate-independent world representations. Limits: Does not prove a unique world model, causal correctness, or safe action selection.; deterministic_capability_compilation: Supports versioned contracts, deterministic compilation, provenance, admission gates, and revocable capability boundaries around model use. Limits: Does not establish predictive accuracy or learned-representation quality.; qcsa_whitepaper: Supports modular predictive substrates and controlled interfaces between heterogeneous reasoning components. Limits: Architecture proposal does not validate world-model learning or grounding.; simulation_scaling: Supports explicit simulation budgets, multi-resolution rollouts, uncertainty, and selective expansion of imagined futures. Limits: Local design source does not prove calibrated long-horizon prediction.; cognitive_compilation: Supports compiling high-level cognitive products into typed, inspectable intermediate artifacts. Limits: Does not show that latent predictions are true or causally adequate.; planforge: Supports bounded plan artifacts, verification, replanning, and failure-aware execution handoffs. Limits: Planning structure alone cannot repair a misspecified model of the world.; ext_world_models_2018: Demonstrates agent learning and acting through compact learned environment models and imagined rollouts in bounded tasks. Limits: Toy-domain results do not establish open-world grounding, causal fidelity, or safe deployment.; ext_dreamer_v3_2025: Supports learning behaviors across diverse domains through a learned dynamics model and latent imagination. Limits: Benchmark breadth is not proof of calibrated real-world counterfactuals or governance effectiveness.; ext_v_jepa_2_2025: Supports self-supervised predictive representations for understanding and planning in physical environments. Limits: Reported embodiment results do not prove general causal models, long-horizon reliability, or ASI-scale safety.; relational_dimension_compiler: Adds typed entity, relation, field, branch, provenance, uncertainty, lifecycle, and reversible macro-object state plus explicit separation between proposed, qualified, believed, observed, executed, contradicted, and retired relations. Limits: The architecture does not establish correct perception, grounding, relation truth, causal structure, branch calibration, contraction fidelity, or world-model advantage.; ext_toward_causal_representation_learning_2021: Grounds the bridge from low-level observations to high-level causal variables and connects causal structure to intervention, transfer, and generalization. Limits: Research agenda and review; no general causal representation or local intervention result.; ext_deep_ensembles_2017: Provides a strong practical predictive-uncertainty and disagreement baseline. Limits: Source-reported benchmark calibration is not distribution-free and can share common-mode blind spots.; ext_conformal_prediction_2021: Grounds finite-sample prediction-set coverage under explicit assumptions and target definitions. Limits: Coverage does not establish semantic correctness, causal adequacy, safety, or validity after assumption-breaking shift.; ext_wilds_2021: Provides natural distribution-shift benchmarks across institutions, geography, time, and collection processes. Limits: Ten benchmark datasets do not cover all OOD conditions or establish local robustness.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	On partially observed natural tasks, governed world-model routes outperform strong reactive, model-free, and single-model baselines on useful outcomes and calibration under shift without worse unsafe action, latency, or governance cost, then reproduce and transfer.
cognitive-compilation-and-semantic-ir.core	cognitive-compilation-and-semantic-ir	Cognitive Compilation should own a versioned, consumer- and target-relative translation contract that lowers an already accepted plan obligation through source, semantic, and target representations into a concrete artifact candidate while preserving addressable obligation, non-goal, authority, rights, assumption, source/context, evidence, resource, verifier, repair, and residual lineage. Every pass binds typed source and target semantics; declared normalization, loss, and ambiguity; preconditions and postconditions; dependencies; deterministic and nondeterministic inputs; compiler and validator identities; costs; receipts; and failure consequences. Acceptance requires post-translation validation against the actual target artifact by an independent-enough evaluator, while repair uses stable semantic identities, observed mutation sets, dependency closure, downstream rebuild, and revalidation. The compiler may block, narrow, request clarification, or residualize a lowering, but it does not reinterpret intent, choose the plan, grant authority, execute effects, self-certify semantic adequacy, or move support or release state.	Design rationale	argument	cognitive_compilation, deterministic_capability_compilation, platonic_world_model, planforge_compiler_arch, genesiscode, treellm, viea, cca_project, moecot_manifest_project, corbens_best_model_possible_project, qcsa_whitepaper, ext_pddl_1998, ext_shop2_2003, ext_tree_of_thoughts_2023, ext_dreamcoder_2020, ext_llvm_langref_docs, ext_mlir_2020, ext_translation_validation_1998, kernel_english_residual_compiler, relational_dimension_compiler, adjudicated_persistence	Source notes available for all 21 assigned sources; 18 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 21 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	cognitive_compilation: Supports reframing complex generative work as a compiler pipeline that separates source planning, semantic compilation, target compilation, execution, repair, typed semantic atoms, dependencies, validation requirements, compiler passes, DAG scheduling, localized repair, and validator-backed artifacts. Limits: Architecture and evaluation-plan source only; no local compiler, benchmark result, working trace suite, target-lowering harness, or empirical ablation validates compiler quality.; planforge_compiler_arch: Supports the planner/compiler analogy with task graphs or intermediate representations, typed primitives, dependencies, interfaces, quality gates, model/tool tier routing, parallel independent nodes, critical-path scheduling, and cost/latency profiles. Limits: Market and cost comparisons are motivating estimates only; cost, time, routing quality, and artifact quality numbers have not been reproduced here.; genesiscode: Supports evidence-carrying compilation through canonical CoreForm, explicit desugaring and semantics, qualified operations and payloads, structural validation, content identity, semantic patches, dependency/obligation rechecking, provenance, capability/effect metadata, row/effect/refinement type candidates, e-graph optimization candidates, and translation-validation discipline. The full audit adds the boundary between canonical syntax and semantic identity, stable semantic anchors and positional AST paths, observed effects and complete capability coverage, and finite equivalence tests and general translation validation. Limits: Three-tab paper, implementation-handoff, and style-contract architecture only. No parser, canonicalizer, evaluator, compiler, semantic patcher, impact-closure analysis, effect/type checker, proof module, translation-validation artifact, benchmark, or security audit exists. Dated Codex workflow advice is lineage context rather than a compiler invariant.; treellm: Supports explicit semantic-substrate intuition through traversable concept graphs, path-derived semantic tokens, residual attributes, graph updates, and model navigation over an external semantic operating system. Limits: No local TreeLLM implementation, token format validation, measured compression ratio, benchmarked reasoning gain, or graph-coverage result exists here.; viea: Supports connecting compiled IR to the larger intent-to-artifact spine through workflow-to-tool compilation, artifact graphs, claim and verification ledgers, residuals, feedback loops, benchmarks, tools, and regression coverage. Limits: Architecture proposal only; no completed deployment, working cognitive compiler, execution log, or reproduced benchmark is proven here.; cca_project: Supports authority-bearing semantic representation through typed primitives, relations, explicit ambiguity, invocation effects, runtime bindings and receipts, plus a five-stage compiler with typed semantic atoms, pass hashes, verification, bounded repair, target IR, and trace bundles. Limits: Pinned-project implementation context only; project artifacts were not replayed from the book, and source hashing or theorem counts do not establish semantic preservation, runtime correctness, or compiler quality.; moecot_manifest_project: Supports compilation all the way down: authority-bearing semantic and target IR, explicit ambiguity debt, field and provenance retention, pass schemas, strict failure, reproducer bundles, applied localized-repair receipts, semantic portability, and separation of execution from mutation authority. Limits: Pinned-project architecture and implementation context only; compilers, holdouts, and deployments were not rerun, and declared repair scope does not prove actual mutation locality.; corbens_best_model_possible_project: Provides a negative implementation record: rich Task spell, semantic/target IR, compiler-pass, and repair interfaces coexist with keyword-driven parsing, mostly linear plans, digest equality mislabeled as semantic parity, positional repair identities, and ordinary-path defaults. Limits: The project does not establish semantic compilation correctness, arbitrary paraphrase preservation, repair locality, or causal runtime effect; interface and artifact presence remain distinct from behavior.; qcsa_whitepaper: Frames SOIDs and SACs as typed semantic-IR references, question traces as evidence-acquisition programs, and semantic-to-physical route plans as policy-constrained target lowering; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. No natural-task, learned-model, universal-semantic, production, safety, privacy, external-independence, chapter-core promotion, AGI, or ASI result is established.; ext_pddl_1998: Provides external planning-language discipline for separating domains, problems, action schemas, preconditions, effects, types, predicates, and goals, while distinguishing a representation interface from planner quality or execution. Limits: No PDDL planner, domain, translation, verifier, IPC benchmark, authority extension, or residual-preservation test was run here.; ext_shop2_2003: Provides external HTN and method-governed ordered decomposition vocabulary for distinguishing compound-task lowering from execution. Limits: No SHOP2 run, method library, IPC domain, competition result, or ASI Stack translation was reproduced.; ext_tree_of_thoughts_2023: Provides external vocabulary for explicit intermediate reasoning candidates, search, evaluation, lookahead, and backtracking rather than linear continuation. Limits: No Tree of Thoughts task, prompt, model, score, search trace, or evaluator result was reproduced; search does not establish semantic preservation or authority.; ext_dreamcoder_2020: Provides external program-synthesis and library-learning context for reusable symbolic abstractions, solved-task lineage, and later search reduction. Limits: No DreamCoder task, program search, learned library, abstraction transfer, or synthesis metric was reproduced.; ext_llvm_langref_docs: Provides official compiler-IR context for typed representations, equivalent serialized forms, verifier boundaries, transformations, analyses, metadata, and target-independent lowering. Limits: No LLVM IR, bitcode, verifier, pass, toolchain, translation, or compiled-artifact result was produced here; machine-level validity does not preserve human obligations by itself.; ext_mlir_2020: Provides external multi-level IR context for dialects, modular passes, verifiers, text forms, progressive lowering, translators, optimizers, and target-specific code generation. Limits: No MLIR dialect, verifier, pass, lowering, code generator, FileCheck test, or paper result was implemented or reproduced.; ext_translation_validation_1998: Provides the direct prior-art comparator for validating each translation against source and target semantics through refinement or simulation rather than assuming compiler correctness. Limits: No semantic framework, refinement relation, translation validator, simulation proof, source-target artifact comparison, or compiler-correctness result was implemented here.; relational_dimension_compiler: Adds a dimensional type system and a four-layer relational IR whose semantic, evidence, computational, and abstraction records can lower relation objects into neural, graph, database, symbolic, simulator, planning, or verification backends without identifying a tensor index with semantic identity. Limits: The existing chapter owns accepted-obligation translation, not the complete RDC world-structure lifecycle; no lowering preserves open-world relational meaning by source authority alone.; adjudicated_persistence: Separates deciding where a lesson should persist from compiling the accepted lesson into a particular semantic or executable representation, with UNKNOWN and deoptimization paths. Limits: Used only to order adjudication before lowering and preserve UNKNOWN and deoptimization; no compiler-correctness or representation-sufficiency result follows.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	IR translation preserves required fields and rejects lossy lowering.
virtual-context-abi.core	virtual-context-abi	The Virtual Context ABI should own the static, versioned, consumer- and purpose-relative request-to-materialization contract between durable memory and model-visible context. It resolves stable object, semantic-address, version, mount, and snapshot references into finite typed representation candidates, then issues a certificate and receipt that bind exact source and field lineage, transformations, omissions and loss, provenance and taint, authority and rights, permitted and prohibited uses, freshness, lease, revocation, selection and omitted frontier, requested and observed adequacy state, costs, faults, and residuals. Admission means only that the actual packet conforms to the frozen request and policy; it does not establish truth, verification adequacy, model use, usefulness, safety, or support. The ABI may deny, ask, abstain, refresh, broaden, narrow, or return a typed fault, but it does not update durable memory, own transactions, infer execution authority, adjudicate claims, execute effects, or move support or release state.	Design rationale	argument	vcm_public, deterministic_capability_compilation, platonic_world_model, context_engineer, verification_bandwidth, viea, vcm_editable, moecot, spinoza, ext_graphrag_2024, ext_hipporag_2024, ext_raptor_2024, ext_mem0_2025, ext_rag_2020, ext_lost_in_middle_2023, ext_longbench_2023, ext_ruler_2024, ext_memgpt_2023, ext_alce_2023, ext_self_rag_2023, ext_longllmlingua_2023, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, qcsa_whitepaper, reflexive_router_whitepaper, kernel_english_residual_compiler, ext_pagedattention_vllm_2023, ext_airllm_2023, ext_hf_accelerate_big_model_inference_2026	Source notes available for all 31 assigned sources; 24 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 31 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	vcm_public: Defines the core VCM control-plane frame: context as a compiled working cache over a versioned semantic address space with semantic objects, immutable content versions, representations, validations, packets, materializations, source and authority binding, adequacy/admission separation, snapshot coherence, revocation, invalidation, audit records, and typed non-success states. Limits: Does not establish end-to-end superiority on frozen pretrained models, natural histories, or neighboring systems; no local resolver, context compiler, or VCM-Bench run exists here.; context_engineer: Supports context as an information supply chain with a governor between planner and worker, layered memory, briefer agents, structured mission briefs, clearance levels, context shards, memory masks, allowed-tool lists, Digital SCIFs, ephemeral context, zeroization, sanitized commits, and information-flow controls. Limits: Reported hallucination, leak, cost, and latency benchmark claims are source-reported only; no harness, raw runs, or independent reproduction is present.; verification_bandwidth: Supports separating context admission from verification adequacy by distinguishing total context length from effective verification workspace, semantic-unit comparison, attention budget, transitive decay, summary loss, and contradiction-rate testing. Limits: Conceptual/theoretical framework only here; the proposed contradiction-rate or adequacy tests have not been run or mechanized in this repository.; viea: Supports routing context materializations into the broader intent-to-execution system through artifacts, claim/verification ledgers, residuals, runtime adapters, feedback, tools, benchmarks, and regression coverage. Limits: Architecture proposal only; no deployed context handoff, runtime adapter, or verified VIEA execution trace exists in this repo.; vcm_editable: Refines VCM terminology around a virtual context ABI with stable addresses, mounts, snapshots, materializations, task-relative representation graphs, evidence-carrying certificates, protected compilation, authority non-escalation, planner-guided semantic paging, transactional memory, and a benchmark agenda. Limits: Connector-readable refinement source only; no local VCM-Bench, model-facing result, or citation-normalized external-literature support is recorded here.; moecot: Supports implementation-reference context for governed orchestration with specialist lanes, fail-closed control-plane ledgers, run/task ledgers, replay, handoff, readiness gates, promotion blockers, and residual tracking. Limits: Runtime and benchmark artifacts have not been imported, inspected, or reproduced; this mapping is not evidence of a deployed context ABI.; spinoza: Supports claim/evidence cells through proof-, citation-, and procedure-carrying claim graphs, support tiers, contradiction detection, dependency tracing, defeaters, downgrade/block behavior, and explicit verifier-scope limits. Limits: Does not prove open-domain natural-language formalization, theorem validity for arbitrary claims, or whole-system epistemic correctness.; cca_project: Supports the boundary between durable semantic memory and bounded context materialization, with explicit artifact identity, provenance, and governed compilation pressure. Limits: Does not establish a durable memory store, resolver correctness, snapshot completeness, restart persistence, model-facing context quality, or support promotion.; moecot_manifest_project: Supports frozen-basis and hot/warm/cold context slices, TTL and token bounds, deterministic freeze/thaw, graph-overlay world state, contradiction recovery, and promotion-boundary discipline. Limits: Does not establish a deployed context store, correct freeze/thaw behavior, graph completeness, expiry enforcement, resolver correctness, or support promotion.; beastbrain_project: Supports typed graph-plus-vector durable objects, provenance-bearing bounded graph snapshots, explicit node/depth/edge/weight limits, reasoning-path traces, lifecycle states, and a durable-memory/context boundary. Limits: Does not establish graph correctness, populated capacity, retrieval quality, persistence, snapshot completeness, memory safety, or support promotion.; bugbrain_project: Supports immutable context handles, virtual slices, IDs, checksums, generation-leased read snapshots, bounded ingress, expiry/revocation revalidation, and negative cases for unchecked offsets and fresh-looking reused IDs. Limits: Does not establish handle safety, cryptographic security, correct restore, durable persistence, context quality, graph completeness, or support promotion.; qcsa_whitepaper: Adds stable SOID identity, plural contextual SVA leases, atlas epochs, SAC adequacy and residual fields, bounded graph materialization, revalidation triggers, and explicit failure when semantic uncertainty is insufficient for a consumer; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. No natural-task, learned-model, universal-semantic, production, safety, privacy, external-independence, chapter-core promotion, AGI, or ASI result is established.; ext_rag_2020: Positions parametric generation plus explicit non-parametric retrieval as a baseline for separating model memory, retrieved passages, provenance, and updateable stores. Limits: Primary metadata and abstract only; no retriever, index, dataset, fine-tuning, factuality result, packet lineage, or local reproduction is established.; ext_graphrag_2024: Positions LLM-derived entity graphs, community summaries, and local-to-global corpus questions as a graph-retrieval and representation comparator. Limits: Metadata-first note only; no paper passage, benchmark, graph build, corpus run, provenance audit, summary-fidelity result, or local reproduction is established.; ext_hipporag_2024: Positions knowledge-graph retrieval, Personalized PageRank, and associative navigation as a long-term-memory retrieval comparator. Limits: Metadata-first note only; no paper result, graph build, multi-hop QA run, poisoning test, provenance audit, or local reproduction is established.; ext_raptor_2024: Positions recursive clustering, abstractive summarization, and retrieval across multiple abstraction levels as a hierarchical context comparator. Limits: Metadata-first note only; no paper result, tree build, QA run, abstraction-fidelity audit, invalidation test, or local reproduction is established.; ext_mem0_2025: Positions extraction, consolidation, retrieval, graph linking, latency, and token-cost tradeoffs in persistent conversational memory. Limits: Metadata-first note only; no LOCOMO run, LLM-judge audit, memory extraction, poisoning, retention, privacy, production, or local reproduction result is established.; ext_lost_in_middle_2023: Positions relevant-information placement sensitivity as evidence that nominal context presence and length are not reliable measures of usable context. Limits: No multi-document QA, key-value retrieval, model run, position-sensitivity result, or ASI Stack context-use test was reproduced.; ext_longbench_2023: Positions bilingual, multi-task long-context evaluation across document QA, summarization, few-shot, synthetic, and code tasks as a diversity baseline. Limits: No LongBench task, adapter, model comparison, provenance-preserving run, or local benchmark result exists here.; ext_ruler_2024: Positions configurable retrieval, multi-hop tracing, and aggregation stress tests as a warning that nominal context size and simple needle retrieval overstate usable context. Limits: No RULER task, adapter, model run, complexity curve, or VCM stress-test result was reproduced.; ext_memgpt_2023: Positions OS-inspired tiered memory, explicit data movement, interrupts, and context-window scarcity as a virtual-context-management comparator. Limits: No MemGPT code, data, document-analysis task, multi-session chat, memory trace, compatibility result, or local reproduction exists here.; ext_alce_2023: Positions end-to-end retrieval and citation-backed generation with separate fluency, correctness, and citation-quality evaluation. Limits: No ALCE corpus, output, citation-entailment audit, evaluator reproduction, or local citation-support result exists here.; ext_self_rag_2023: Positions adaptive retrieval, passage reflection, and retrieval-aware self-critique as a comparator for request, selection, and evaluator-dependence decisions. Limits: No checkpoints, reflection-token training, dataset, factuality, citation, QA, fact-verification, or local adaptive-retrieval result was reproduced.; ext_longllmlingua_2023: Positions prompt compression under cost, latency, and long-context performance tradeoffs as a comparator for explicit loss and adequacy receipts. Limits: No compressor, dataset, token saving, latency, benchmark, source-to-packet fidelity, or local adequacy result was reproduced.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Admission fixture rejects stale, conflicting, mode-confused, stale-certificate, missing-certificate, or authority-escalating packets and records residual.
durable-semantic-memory-and-knowledge-lattices.core	durable-semantic-memory-and-knowledge-lattices	Durable semantic memory should be admitted through a versioned knowledge-lattice contract that binds object and relation identity, ontology, provenance, support state, temporal validity, authority and rights, merge and supersession, contradiction, retrieval route, compaction and forgetting, restart recovery, consumer use, and residual uncertainty; retrieval quality, graph connectivity, model recall, persistence, or a fluent answer alone establishes neither truth, complete memory, safe consolidation, erasure, nor decision authority.	Design rationale	argument	qcsa_whitepaper, vcm_public, treellm, portia_synapse, ext_graphrag_2024, ext_hipporag_2024, ext_mem0_2025, ext_titans_2025, ext_hermes_agent_2026, adjudicated_persistence	Source notes available for all 10 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 10 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	vcm_public: Passage-reviewed VCM lineage for durable semantic objects and immutable events, representation graphs, contradiction and rejection memory, decision-time observable retention features, anti-starvation, invalidation, deletion closure, and the barrier between deployable blind memory policy and future-aware oracle labels. Limits: The source reports a bounded synthetic control-plane study and an uninformative packet pilot, not durable-memory truth, retention-policy quality, natural long-horizon outcomes, deletion or unlearning completeness, privacy, deployment, or chapter-core support.; qcsa_whitepaper: Author-source architecture for question-compiled semantic addressing, typed semantic objects, multi-facet address atlases, certificates, and semantic-to-physical resolution. It supplies a bounded design vocabulary for the chapter’s durable knowledge-lattice contract. Limits: Conceptual architecture and local design rationale only; it does not establish learned retrieval quality, open-world truth, durable-memory correctness, deletion closure, transfer, deployment readiness, or chapter-core support.; ext_graphrag_2024: Primary GraphRAG comparator for LLM-derived entity graphs, community summaries, and global corpus questions; generated graph and summary layers remain fallible derived representations and do not establish truth, complete coverage, local adequacy, or an ASI Stack memory result. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_hipporag_2024: Primary NeurIPS comparator for knowledge-graph retrieval with Personalized PageRank and single-step associative navigation; reported multi-hop QA gains do not establish durable truth, update correctness, resistance to poisoning, local reproduction, or a general memory system. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_mem0_2025: Primary preprint comparator for extracting, consolidating, retrieving, and graph-linking conversational memory under latency and token-cost constraints; LOCOMO and LLM-judge results do not establish fact correctness, poisoning resistance, general memory, local reproduction, or production readiness here. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_titans_2025: Primary neural-memory comparator for test-time memorization and long-context sequence modeling. The paper motivates mutable-state provenance and rollback tests; no local model or benchmark result is reproduced. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; treellm: Complete visible TreeLLM correction lineage for model/memory separation, explicit semantic graphs, structured tokens, typed exact/approximate/exploratory/ingestion routes, learned anchors, DKL snapshots, overlays, graph hygiene, adaptive caching, speculative traversal, skills, and mobile or federated variants. Limits: The lineage contains mutually inconsistent drafts and later Aletheia contamination. No local graph, token, HLSH property, semantic update bound, retrieval result, grounding, privacy, federation, or production claim is established.; portia_synapse: Authenticated TreeLLM successor that instantiates a DKL navigator over a versioned RichContext and mutable Focus memory, motivating snapshot, codec, provenance, neighbor-policy, and session-isolation obligations. Limits: Source-reported implementation only; it does not establish memory truth, state isolation, deletion closure, learning, routing quality, or transfer.; ext_hermes_agent_2026: Supplies a current implementation comparator that separates bounded always-present MEMORY/USER snapshots, on-demand FTS5 session search, and longer load-on-demand procedural skills, while scanning admitted prompt memory for injection and exfiltration patterns. Limits: Official software and documentation review only; no memory fact, search result, poisoning control, session isolation, learning behavior, cost, utility, or deployment result was reproduced.; adjudicated_persistence: Treats memory as one consequential persistence locus whose writes require evidence, authority, qualification, expiry, and descendant-aware revocation. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Hybrid memory improves bounded update-heavy tasks while preserving provenance, contradiction calibration, rights closure, and recovery; no truth, complete-memory, or erasure claim.
context-transactions-snapshots-mounts-and-taint.core	context-transactions-snapshots-mounts-and-taint	Context Transactions should own the dynamic, versioned state-transition contract for durable context memory. Each accepted transaction binds principal, consumer, purpose, operation, base snapshot, branch, mounts, actual read/write/derive/delete/revoke sets, isolation and conflict policy, authority and rights, taint and declassification, durability and recovery model, budget, horizon, and support ceiling to an observed pre-state and a causally ordered attempted, applied, durable, visible, replayed, or recovered post-state. Commit, branch, merge, abort, retry, compaction, deletion, revocation, and recovery must preserve exact identities, obligations, faults, costs, and residuals. The transaction layer may change durable context state, but it does not own static packet materialization, semantic truth, belief revision, model/optimizer state, external effects, artifact correctness, verification adequacy, support, or release.	Design rationale	argument	vcm_public, platonic_world_model, ladon_manhattan, context_engineer, black_hole_context_manager, vcm_editable, ext_graphrag_2024, ext_mem0_2025, ext_cap_theorem_gilbert_lynch_2002, ext_txfs_2018, ext_bourtoule_machine_unlearning_2021, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_best_model_possible_project, reflexive_router_whitepaper, kernel_english_residual_compiler, ext_openai_prompt_caching_docs_2026, ext_anthropic_prompt_caching_docs_2026, ext_gemini_context_caching_docs_2026, ext_vllm_automatic_prefix_caching_2026, ext_azure_llm_semantic_cache_2026	Source notes available for all 23 assigned sources; 15 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 23 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	vcm_public: Supports transactional context semantics through immutable content versions, semantic objects, source and authority binding, protected compilation, snapshot coherence, revocation, invalidation, audit records, and visible non-success states. Limits: No deployed memory store, resolver, runtime branch-isolation test, side-channel validation, VCM-Bench result, or VCM conformance run is present; bounded synthetic deletion-closure and memory-store fixture harnesses do not promote source-derived support.; ladon_manhattan: Supports sensitive context boundaries with opaque credential handles, policy-mediated secret injection, hardened managers, ephemeral isolated compartments, zeroization, compartment lifecycle, audit considerations, and side-channel assumptions. Limits: Security architecture only; no Ladon implementation, kernel-level test, side-channel validation, or security audit exists here.; context_engineer: Supports layered memory, context shards, clearance labels, memory masks, allowed-tool lists, Digital SCIFs, ephemeral context injection, zeroization, sanitized commits, outgoing secret scans, and read/write policies. Limits: No benchmark harness or runtime enforcement test for clearance, taint, leak prevention, cost, or latency is reproduced here.; black_hole_context_manager: Supports concrete memory-budget patterns with chunks, embeddings, entropy, mass, timestamps, goal-drift scoring, lazy recomputation, freezing, eviction, cluster management, critical-context preservation, sanitization, and rate limits. Limits: Design/specification source only; no package, local test run, freeze/evict fixture, or benchmark demonstrates production behavior.; vcm_editable: Refines the transaction vocabulary with stable addresses, mounts, snapshots, materializations, planner-guided semantic paging, protected compilation, authority non-escalation, and transactional/governed memory. Limits: Connector-readable variant only; no local VCM-Bench, deployed memory-store conformance run, or model-facing result exists here; bounded synthetic memory-store fixtures do not validate this source variant.; ext_graphrag_2024: Positions LLM-derived entity graphs and community summaries as mutable derived state whose source, version, update, conflict, invalidation, taint, and deletion lineage must be governed separately from retrieval quality. Limits: Metadata-first note only; no paper passage, benchmark, graph build, transaction, update, deletion, provenance audit, summary-fidelity result, or local reproduction is established.; ext_mem0_2025: Positions extraction, consolidation, retrieval, and graph linking as persistent-memory operations with latency and token-cost trade-offs that require independent admission, update, poisoning, retention, and deletion controls. Limits: Metadata-first note only; no LOCOMO run, LLM-judge audit, memory extraction, consolidation correctness, poisoning, retention, privacy, deletion, production, or local reproduction result is established.; ext_cap_theorem_gilbert_lynch_2002: Supports treating partitions, delays, and consistency-versus-availability trade-offs as first-class limits on multi-store snapshot freshness, revocation propagation, commit certainty, and ordinary availability. Limits: The source does not specify this chapter’s semantic objects, mounts, taint, deletion closure, or transaction protocol; no distributed context store, partition test, consensus mechanism, or CAP proof reproduction exists here.; ext_txfs_2018: Supports explicit transaction APIs, journal-bounded atomicity, isolation, conflict detection, crash consistency, durability, transaction-size limits, and the distinction between byte-level recovery and broader process or external-effect recovery. Limits: TxFS was not installed or reproduced; its paper scope does not establish semantic-object identity, taint, authority, rights, derivative deletion, multi-backend atomicity, model use, or context-task usefulness.; ext_bourtoule_machine_unlearning_2021: Supports separating a structured deletion and retraining path from verified parametric forgetting, privacy, downstream-artifact erasure, utility preservation, and generic unlearning; it sharpens this chapter’s storage-versus-model-erasure boundary. Limits: SISA code and checkpoints were not run; the source does not prove deletion from this repository’s context stores, caches, backups, models, adapters, descendants, or published artifacts and does not establish privacy compliance.; cca_project: Supports durable-memory/context separation, artifact identity, and governed material transitions. Limits: No deployed store, crash recovery, persistence, or support promotion.; moecot_manifest_project: Supports deterministic freeze/thaw, tiered context persistence, graph-overlay state, promotion gates, and acknowledged transitions. Limits: No deployed store, process restart, or persistence result.; beastbrain_project: Supports hot/cold/archive memory lifecycles, background decay, bounded graph snapshots, merge and pruning pressure. Limits: No deployed persistence, merge correctness, compaction, or recovery result.; bugbrain_project: Supports generation-leased snapshots, tier movement, restore revalidation, integrity and deterministic replay pressure. Limits: No deployed restore, atomic update, cryptographic, or persistence result.; corbens_best_model_possible_project: Supplies negative cases for named memory transitions without durable effects, absolute-path state, playback-only replay, and nominal runtime surfaces. Limits: No deployed store, process restart, causal memory effect, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Transaction replay proves conflict/deletion gates fire.
verification-bandwidth-and-context-adequacy.core	verification-bandwidth-and-context-adequacy	Verification Bandwidth should own the prospective, claim-specific adequacy contract for a verification attempt. Before outcomes, it binds target proposition and scope, population and environment, risk and consequence, requested support effect, required positive, negative, boundary, contradiction, counterexample, and transfer obligations, available source units, verification modes, tools, evaluator-dependency graph, authority and rights, budget, horizon, stop rule, and escalation path. After execution, it records every attempted, passed, failed, disputed, unknown, infeasible, and unattempted obligation plus actual artifacts, costs, disagreement, residuals, expiry, and causal-use observations. Adequacy means only that the declared verification program was sufficient for its exact purpose under stated premises; it does not establish claim truth, model cognition, source correctness, formal-model fidelity, useful outcomes, safety, support promotion, or release.	Design rationale	argument	verification_bandwidth, vcm_public, spinoza, treellm, vcm_editable, ext_graphrag_2024, ext_hipporag_2024, ext_raptor_2024, ext_dont_hallucinate_abstain_2024, ext_alce_2023, ext_checklist_2020, ext_gpqa_2023, ext_longbench_2023, ext_ruler_2024, ext_longllmlingua_2023, ext_reluplex_2017, kernel_english_residual_compiler	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 17 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	verification_bandwidth: Defines the core distinction between context length and effective verification workspace, semantic units, constraint satisfaction, pairwise grinding, two-body verification limits, dominant-unit crowd-out, transitive decay, hierarchical summary tradeoffs, and proposed contradiction-rate testing. Limits: Conceptual/theoretical source only; contradiction-rate, distractor-resistance, and adequacy-labeling experiments have not been run or mechanized here.; vcm_public: Supports adequacy/admission separation, protected compilation, source and authority binding, typed non-success states such as unknown/deferred/infeasible/ask/abstain/deny/retry, and visible adequacy gaps. Limits: Does not establish end-to-end model superiority or adequacy-classifier correctness; no local VCM resolver or model-facing packet result exists.; spinoza: Supports routing adequacy results into claim ledgers through verifier separation, proof/citation/procedure-carrying claim tiers, belief revision, contradiction detection, defeaters, downgrade/block behavior, and verifier timeouts. Limits: Does not solve open-domain autoformalization, theorem validity for arbitrary claims, or whole-system epistemic correctness.; treellm: Supports explicit semantic-unit representation via traversable concept graphs, path-derived semantic tokens, residual attributes, graph operations, and updateable external semantic substrates. Limits: No implementation, benchmarked reasoning gain, fixed-token adequacy validation, or graph coverage result exists in this repository.; vcm_editable: Supports evidence-carrying, planner-guided context compilation with virtual context ABI, representation graphs, certificates, authority non-escalation, transactional memory, and a benchmark agenda. Limits: Connector-readable refinement source only; no VCM-Bench, contradiction benchmark, or citation-normalized external-literature support is recorded here.; ext_graphrag_2024: Positions LLM-derived entity graphs and community summaries as a comparator for making source units, abstraction layers, omissions, and cross-document obligations explicit before verification. Limits: Metadata-first note only; no paper passage, benchmark, graph build, corpus run, contradiction audit, adequacy result, or local reproduction is established.; ext_hipporag_2024: Positions knowledge-graph retrieval, Personalized PageRank, and associative navigation as a comparator for multi-hop source availability and retrieval obligations. Limits: Metadata-first note only; no paper result, graph build, multi-hop QA run, poisoning test, provenance audit, adequacy result, or local reproduction is established.; ext_raptor_2024: Positions recursive clustering, abstractive summarization, and retrieval across abstraction levels as a comparator for decomposition, compression, boundary checks, and fidelity residuals. Limits: Metadata-first note only; no paper result, tree build, QA run, abstraction-fidelity audit, invalidation test, adequacy result, or local reproduction is established.; ext_alce_2023: Positions end-to-end retrieval and citation-backed generation with separate fluency, correctness, and citation-quality evaluation, sharpening the difference between citation presence and claim support. Limits: No ALCE corpus, output, citation-entailment audit, evaluator reproduction, or local citation-support result exists here.; ext_checklist_2020: Positions minimum-functionality, invariance, and directional-expectation tests as structured behavioral verification modes beyond aggregate accuracy. Limits: No CheckList software, test template, model run, behavioral coverage audit, fresh-holdout generator, or local failure result exists here.; ext_gpqa_2023: Positions expert-written hard questions and expert-versus-nonexpert validation gaps as a scalable-oversight and evaluator-competence pressure test. Limits: No GPQA dataset, model score, expert-review reproduction, contamination audit, disagreement record, or local benchmark result exists here.; ext_longbench_2023: Positions bilingual multi-task long-context evaluation across document QA, summarization, few-shot, synthetic, and code tasks as a task-diversity comparator. Limits: No LongBench task, adapter, model comparison, provenance-preserving run, or local adequacy result exists here.; ext_ruler_2024: Positions configurable retrieval, multi-hop tracing, and aggregation stress tests as a warning that nominal context size and simple needle retrieval overstate usable context. Limits: No RULER task, adapter, model run, complexity curve, context stress test, or local adequacy result was reproduced.; ext_longllmlingua_2023: Positions prompt compression under cost, latency, position-bias, and performance trade-offs as a comparator for explicit loss, deleted evidence, and adequacy receipts. Limits: No compressor, dataset, token saving, latency, benchmark, source-to-packet fidelity, contradiction-retention, or local adequacy result was reproduced.; ext_reluplex_2017: Positions property-specific neural-network verification and counterexample production as a formal-mode comparator whose authority remains inside declared models and properties. Limits: No Reluplex run, ACAS Xu reproduction, ASI Stack model verification, semantic-model fidelity result, implementation refinement, or broad system-safety proof exists here.; ext_dont_hallucinate_abstain_2024: Supports measuring abstention with coverage, accuracy, calibration, and useful response behavior while treating self-reflection and model agreement as fallible evidence. Limits: The ACL models, prompts, domains, collaboration schemes, and reported gains were not reproduced; the local single-model router is not an independent multi-model panel.	a workload with correct and incorrect initial candidates; adaptive benefit at a matched real budget; independently assessed evaluator and production transfer; independently assessed evaluator validity; multiple models, seeds, and natural workloads; nonzero substantive answer utility; production routing safety, latency, cost, and transfer; replicated corruption reduction	Adequacy record rejects context that exceeds verifier bandwidth.
claim-ledgers-and-belief-revision.core	claim-ledgers-and-belief-revision	Claim Ledgers should own the durable identity and append-only state-transition history of each material claim and its semantic variants. Every record binds canonical proposition and scope, definitions and assumptions, population and environment, provenance and source roles, evidence and attack refs, support and uncertainty states, contradiction and defeater links, dependencies, ontology version, lifecycle, commitment, authority and rights, surface refs, expiry, residuals, and current materialized view; every proposed update binds trigger, before/after states, transition type, evidence-transition and review refs, affected dependency closure, surface-sync plan, concurrency base, migration, costs, and non-overwrite receipt. The ledger may record or route promotion, downgrade, split, merge, supersession, deprecation, retirement, dispute, or no change only through the owning gates. It does not establish claim truth, source or evidence validity, verification adequacy, semantic equivalence, reviewer competence, formal-model fidelity, action authority, usefulness, safety, support movement, or release.	Design rationale	argument	spinoza, platonic_world_model, viea, coherence_exchange, aletheia, uat, ext_agm_belief_revision_1985, ext_truth_maintenance_system_1979, ext_assumption_based_tms_1986, ext_alce_2023, ext_self_rag_2023, ext_checklist_2020, ext_w3c_prov_o_2013, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_best_model_possible_project, qcsa_whitepaper, reflexive_router_whitepaper, regret_engine	Source notes available for all 21 assigned sources; 18 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 21 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	spinoza: Supports claim ledgers through proposer/verifier separation, proof/citation/procedure-carrying claim tiers, contradiction detection, dependency tracing, defeaters, entrenchment, downgrade/block behavior, and explicit belief revision. Limits: Does not prove open-domain claim extraction, natural-language formalization, or a complete belief-revision engine in this repository.; viea: Supports durable claim and verification ledgers, artifact graphs, support states, residuals, feedback, benchmark/regression coverage, and the rule that important responses become durable artifacts. Limits: Architecture proposal only here; no deployed VIEA claim-ledger system or runtime evidence trace exists in this repo.; coherence_exchange: Supports structured claim/evidence units, verification supply chains, contestability, and fork/exit/audit governance framing for epistemic objects. Limits: Epistemic-liquidity and economic metaphors remain speculative synthesis, not implemented economics or external corroboration.; aletheia: Supports a claim-native release surface in which structured claims carry identity, type, qualifiers, dependencies, evidence refs, assurance class, and actionability while narrative assertions bind back to claim IDs; also supports scope-integrity checks, bounded adversarial review, explicit non-convergence, governed commitments, and recertification. Limits: The source makes its authored claims array enumerable but does not establish deterministic or complete claim extraction from arbitrary language, semantic equivalence, evidence validity, reviewer independence, safe release, useful advantage, or any implementation result.; uat: Supports retrieval-bounded proposition decomposition, verified/inferred/unsupported states, adversarial review, SME checkpoints, and final sign-off as review paths for contested ledger states. Limits: Does not prove tribunal thresholds, reviewer independence, or novel truth discovery beyond the declared dossier.; ext_agm_belief_revision_1985: Provides the canonical formal-epistemology comparator for explicit contraction and revision operations, rationality postulates, and representation results over theories rather than prose rewrites. Limits: Only bibliographic metadata and abstract-level source-note analysis are present; no AGM proofs, postulates, partial-meet contraction, natural-language mapping, or local implementation is reproduced.; ext_truth_maintenance_system_1979: Provides the classic AI comparator for maintaining reasons, justifications, and dependencies behind program beliefs so later revision and explanation can preserve why a state existed. Limits: Only bibliographic metadata and source-note analysis are present; no Doyle TMS algorithm, dependency-directed backtracking, reasoner, or natural-language truth-maintenance result is implemented.; ext_assumption_based_tms_1986: Provides an assumption-based truth-maintenance comparator for explicit assumption sets, inconsistent information, dependency contexts, and context switching. Limits: Only DOI metadata and source-note analysis are present; no ATMS labels, environments, algorithms, context search, or assumption-completeness result is implemented.; ext_alce_2023: Provides a citation-evaluation comparator that separates fluent answer generation, factual correctness, citation presence, and citation quality, motivating claim-level source roles and attack refs. Limits: No ALCE corpus, code, model output, entailment audit, citation metric, or score is reproduced; citation presence remains distinct from source truth and accepted support.; ext_self_rag_2023: Provides an adaptive retrieval and self-critique comparator for recording retrieval necessity, passage relevance, generated critique, and factuality or citation signals as fallible claim inputs. Limits: No Self-RAG checkpoint, training, dataset, reflection-token trace, citation score, or factuality result is reproduced; self-critique cannot promote its own claims.; ext_checklist_2020: Provides a behavioral-test comparator for minimum-functionality, invariance, and directional-expectation failures that should become durable attacks, regressions, or revision triggers rather than disappear after aggregate scoring. Limits: No CheckList software, templates, model run, behavioral coverage audit, or failure reproduction is present.; ext_w3c_prov_o_2013: Provides an interoperable asserted-provenance vocabulary for entities, activities, agents, derivation, attribution, delegation, revision, quotation, primary source, and invalidation around claim records and transitions. Limits: No PROV-O graph, conformance result, integrity check, authority result, or semantic truth result exists; asserted provenance is not justified belief or support promotion.; cca_project: Supports separated epistemic state, durable claim identity, dependency-aware revision, and non-overwriting histories. Limits: No open-domain contradiction detector, belief engine, or ontology correctness result.; moecot_manifest_project: Supports canonical claim/evidence graphs, contradiction recovery, bounded repair, readiness blocking, and persistent control-plane state. Limits: No deployed contradiction or durable-belief behavior result.; beastbrain_project: Supports durable semantic objects, provenance, unresolved-outcome retention, and contradiction-aware memory pressure. Limits: No belief-engine, ontology, or revision-quality result.; bugbrain_project: Supports generation/version identity, restore discipline, graph dependencies, and negative cases for nominal integrity or replay. Limits: No contradiction-quality, persistence, or ontology result.; corbens_best_model_possible_project: Supplies negative cases for nonmonotone contradiction response, vacuous verification, playback-only persistence, and named state without causal effect. Limits: No deployed belief engine, causal revision, or ontology correctness result.; qcsa_whitepaper: Requires ontology, propositions, evidence for and against propositions, provenance, belief/support state, contradiction, and permitted-use authority to remain distinct semantic records in the evidence-bearing hypergraph; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. A semantic location or valid certificate still does not establish truth, belief correctness, production validity, chapter-core promotion, AGI, or ASI.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Contradictory evidence changes ledger state without silent promotion.
spinoza-verification-and-proof-carrying-claims.core	spinoza-verification-and-proof-carrying-claims	Selected claims and artifacts should move through proof-carrying, justification-carrying, or adversarial-review envelopes that record tier, interpretation mapping, evidence dossier, verifier or tribunal result, dissent, limitations, failed attempts, required actions, residuals, and ledger effects.	Design rationale	argument	spinoza, deterministic_capability_compilation, platonic_world_model, genesiscode, coherence_exchange, verification_bandwidth, treellm, uat, talos, ext_proof_carrying_code_1997, ext_lean4_theorem_proving, ext_autoformalization_llms_2022, ext_ai_safety_debate_2018, ext_llm_as_judge_mt_bench_2023, ext_contestable_ai_design_2022, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_best_model_possible_project	Source notes available for all 20 assigned sources; 18 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 20 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	spinoza: Supports proposer/verifier/maintainer separation; proof-, norm-mapping-, procedure-, speculation-, and monitoring-artifact classes; deterministic rendering and counterexample-driven intent tests; FIMO applicability and conflict custody; multi-factor template trust and decay; least-privilege theory bridges; bounded revision; and explicit cross-tier quarantine, downgrade, fork, replay, escalation, and persistent-conflict routes. Limits: The twelve-tab family is an architecture and evaluation program. It does not establish semantic equivalence, open-domain autoformalization, arbitrary theorem validity, FIMO correctness, legal compliance, template calibration, revision quality, proof-laundering resistance, user outcomes, ROI, security, deployment, support promotion, or whole-system epistemic correctness.; genesiscode: Supports obligation-carrying artifacts through tests, property checks, proofs, provenance, resource budgets, determinism claims, capability policies, replay logs, semantic patches, and small trusted-core discipline. Limits: No local GenesisCode implementation, replay checker, proof artifact, benchmark, or security audit has been imported or reproduced.; coherence_exchange: Supports a verification supply chain connecting PlanForge, Spinoza, tribunals, audit, and structured epistemic units. Limits: Speculative synthesis only; it does not implement or externally validate the verification supply chain.; verification_bandwidth: Supports requiring stronger verification artifacts for high-risk claims because context capacity and generation capacity do not imply adequate joint verification. Limits: The proposed contradiction-rate and verification-workspace tests have not been run or mechanized here.; treellm: Supports explicit semantic representation and graph/path traces as potential interpretation-mapping substrate for claim justification. Limits: No local TreeLLM implementation, measured compression ratio, benchmarked reasoning gain, or verified token format exists in this repository.; uat: Supports tribunal coordination through orthogonal priors, retrieval dossiers, atomic proposition states, bounded adversarial siege, rotating logic/citation/omission critiques, hard cycle caps, compression checks, SME checkpoints, and final sign-off. Limits: No local multi-reviewer run, independent validation, or threshold reproduction exists in this repo.; talos: Supports review artifacts as typed work products with adjudication, evidence records, audit logs, replay, approval gates, residual feedback, and delivery discipline. Limits: No Talos tribunal pipeline, security result, replay log, or benchmark has been reproduced in this repository.; ext_proof_carrying_code_1997: Supplies the primary historical comparator for pairing an untrusted artifact with machine-checkable evidence under an explicit consumer policy and producer/consumer separation. Limits: Primary-record source note only; no PCC implementation, machine-code policy, proof producer, proof checker, runtime admission, or reproduction exists here.; ext_lean4_theorem_proving: Grounds dependent-type, propositions-as-types, proof-term, tactic, structure, and inductive-record vocabulary for exact formal artifacts. Limits: Official-text source note only; no imported theorem establishes semantic adequacy, runtime correspondence, empirical truth, or any broad ASI Stack claim.; ext_autoformalization_llms_2022: Supplies the direct comparator for LLM translation from informal mathematics into formal specifications and exposes interpretation mapping as a separate empirical problem. Limits: Metadata-and-abstract source note only; no dataset, split, model output, Isabelle/HOL run, MiniF2F result, semantic-equivalence audit, or local autoformalizer is represented.; ext_ai_safety_debate_2018: Supplies a primary-paper comparator for adversarial information selection, answer precommitment, bounded turns, judge instructions, justified ignorance, equilibrium assumptions, and judge failure. Limits: Full-paper note only; the narrow source-reported MNIST result is not reproduced, and no open-domain debate, human-judge, equilibrium, decomposition, or verdict-quality result exists here.; ext_llm_as_judge_mt_bench_2023: Supplies the comparator for model-graded evaluation, human-preference agreement, and position, verbosity, self-enhancement, and reasoning biases. Limits: Metadata-and-abstract source note only; no MT-Bench prompts, Arena data, model judge, human comparison, bias probe, calibration, or local verdict audit exists.; ext_contestable_ai_design_2022: Supplies governance framing for challenge, dispute, appeal, audit, dissent, redress, and institutional response surfaces around AI-mediated decisions. Limits: DOI-record source note only; no framework implementation, institutional case study, timeliness test, redress outcome, legal compliance, or deployed contestability exists.; cca_project: Supports explicit verifier boundaries, proof/runtime separation, failure preservation, and bounded review artifacts. Limits: No reviewer-independence, method-adequacy, or verdict-correctness result.; moecot_manifest_project: Supports fail-closed verification routes, replay and readiness gates, bounded specialist lanes, and explicit promotion blockers. Limits: No deployed tribunal, independent reviewers, or verdict-quality result.; beastbrain_project: Supports tribunal and verifier concepts plus negative evidence about stubs, correlated review, simulated states, and readiness overclaim. Limits: No reviewer-independence, tribunal-quality, or live adjudication result.; bugbrain_project: Supports explicit check identity, deterministic replay pressure, privileged review boundaries, and negative cases for skipped or nominal checks. Limits: No proof of verifier competence, independence, security, or verdict correctness.; corbens_best_model_possible_project: Supplies negative cases for vacuous verification, playback-only evidence, dependent checks, default states, and nominal subsystems without causal effect. Limits: No deployed tribunal, reviewer independence, method adequacy, or verdict-correctness result.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Invalid proof artifact, missing tribunal review, captured reviewer path, or evidence-thin verdict blocks claim tier escalation or accepted review.
labor-os-and-typed-jobs.core	labor-os-and-typed-jobs	The execution layer should convert plans into typed jobs managed by a governed labor operating system.	Design rationale	argument	talos, deterministic_capability_compilation, viea, genesiscode, software_magic_grimoire, talos_md, moecot, ext_temporal_docs, ext_airflow_dag_docs, ext_bpmn_2_0_2_spec, ext_kubernetes_jobs_docs	Source notes available for all 11 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 11 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	talos: Supports the labor-OS frame through typed job lifecycle, decomposition, contract lock, factory execution, adjudication, delivery, feedback/autopoiesis, deterministic control planes, audit, replay, evidence, isolation, and approval-gate failure modes. Limits: Design-source support only here; no Talos scheduler, approval service, runtime, replay log, benchmark, or security result has been reproduced in this repository.; viea: Supports lowering intent into structured command contracts, artifacts, specialist-routed work, verified outputs, runtime execution, feedback, residuals, tools, benchmarks, and regression coverage. Limits: Architecture proposal only; no deployed intent-to-job pipeline or verified execution trace exists here.; genesiscode: Supports separating proposals from effects through capability boundaries, deterministic effect logs, provenance, obligations, semantic patches, and small trusted-core discipline. Limits: No local GenesisCode implementation, effect runner, replay checker, proof artifact, or security audit has been imported or reproduced.; software_magic_grimoire: Supports job and workflow shape through bounded instructions with roles, objectives, context, constraints, procedure, output, verification, failure behavior, guards, entry/exit rules, loops, recursion, and recovery paths. Limits: Promptcraft/workflow vocabulary only; not empirical evidence that a labor OS performs better or safely executes work.; talos_md: Supports public-release Talos terminology around typed AI labor, deterministic manufacturing, five-phase job lifecycle, control-plane ledgers, evidence, isolation, auditability, blind handles, and Digital SCIF boundaries. Limits: Variant/canonicalization aid only; reported benchmark or security claims remain unverified.; moecot: Supports runtime-reference context for governed orchestration with a compact orchestrator, specialist lanes, fail-closed control plane, run/task/control-plane ledgers, readiness gates, replay, handoff, promotion blockers, and residual tracking. Limits: Connector-readable implementation-reference source only; runtime code, logs, benchmark artifacts, and replay records have not been inspected or reproduced here.; ext_temporal_docs: Supplies the closest durable-execution comparator through workflow/event history, long-running state, workers, activities, external events, failure recovery, resume, and replay vocabulary. Limits: Official-documentation note only; no Temporal service, namespace, worker, workflow, activity, event history, failure injection, or replay was run.; ext_airflow_dag_docs: Supplies mature DAG, schedule, task, dependency, callback, retry, timeout, worker, and operational workflow vocabulary. Limits: Official-documentation note only; no Airflow install, scheduler, executor, DAG run, task instance, callback, retry, or comparison trace exists.; ext_bpmn_2_0_2_spec: Supplies formal stakeholder-readable, implementation-independent process-notation and software-process translation lineage. Limits: Specification-page note only; no BPMN model, conformance test, engine, compiler, model check, or typed-job translation exists.; ext_kubernetes_jobs_docs: Supplies batch-job resource, Pod, completion, failure, retry, backoff, deadline, terminal-condition, scheduling, and cleanup vocabulary. Limits: Official-documentation note only; no cluster, Job object, Pod, controller, scheduler, status trace, cleanup, isolation result, or typed-job compilation exists.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Job cannot dispatch without requirements, permissions, and receipt path.
ai-work-surfaces-agent-harnesses-and-organizational-absorption.core	ai-work-surfaces-agent-harnesses-and-organizational-absorption	Every expansion of an AI work surface should be governed as a versioned abstraction-absorption transition that binds capability, context, state, tools, authority, effects, verification, human control, accountability, and residuals before project-, role-, team-, or organization-scale autonomy is accepted.	Design rationale	argument	ext_github_copilot_work_surfaces_2026, ext_augment_code_agent_2026, ext_openai_codex_work_surfaces_2026, ext_anthropic_claude_code_2026, ext_opencode_agent_2026, ext_oh_my_pi_agent_2026, ext_hermes_agent_2026, ext_openclaw_agent_runtime_2026, ext_elizaos_agent_runtime_2026	Source notes available for all 9 assigned sources; 9 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 9 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_github_copilot_work_surfaces_2026: Provides one product-family comparator spanning inline suggestion, chat, command-line, pull-request, desktop, and agent-driven work surfaces. Limits: Current official documentation only; no complete launch chronology, workflow reproduction, productivity result, or superiority claim.; ext_augment_code_agent_2026: Provides an IDE-panel comparator across chat, read-only inquiry, approval-paused agent work, auto execution, diffs, checkpoints, stop, and steering. Limits: Documentation review only; no execution or control was reproduced.; ext_openai_codex_work_surfaces_2026: Provides a coding-agent comparator across local terminal, IDE, cloud, automation, permissions, project instructions, skills, plugins, and integrations. Limits: Official documentation and one pinned CLI revision; no hosted-system implementation, benchmark, correctness, safety, or productivity result was reproduced.; ext_anthropic_claude_code_2026: Explicitly grounds the agentic-harness distinction and a gather-context, act, verify loop across terminal, IDE, desktop, web, remote, and automation surfaces. Limits: Official documentation only; no execution, complete-mediation, task-quality, or organizational-substitution result.; ext_opencode_agent_2026: Provides a pinned open-source comparator for model-provider-flexible terminal, desktop, and IDE work with plan/build modes, project instructions, recovery, and sharing. Limits: Source and documentation inspection only; no provider parity, permission, privacy, or task result.; ext_oh_my_pi_agent_2026: Provides a pinned comparator for an integrated terminal harness absorbing editing, LSP, shell, browser, subagents, memory, review, provider switching, and collaboration. Limits: No reported token, performance, security, quality, or collaboration claim was reproduced.; ext_hermes_agent_2026: Marks a move toward freestanding persistent agents with memory, skills, tools, and human-gated procedure mutation. Limits: Pinned source inspection only; no learning, utility, memory, security, or persistence result was reproduced.; ext_openclaw_agent_runtime_2026: Marks a gateway- and channel-connected agent surface joining sessions, tools, devices, external harnesses, and governed skill proposals. Limits: Pinned documentation inspection only; no gateway, harness, security, reliability, learning, or benchmark result.; ext_elizaos_agent_runtime_2026: Marks a modular freestanding runtime whose plugins contribute context, action, evaluation, services, events, routes, tests, and views. Limits: Pinned documentation and source inspection only; no execution, test reproduction, security assessment, performance result, or promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	A bounded transition may advance only when exact identity, authority, effect, review, accountability, rollback, and residual obligations are preserved; task success cannot establish role competence, organizational legitimacy, harness superiority, safety, or general transfer.
human-ai-organizations-delegation-and-accountability.core	human-ai-organizations-delegation-and-accountability	Consequential delegation requires a versioned organizational contract binding charter, affected parties, actors, roles, competence, workload, accessibility, information and decision rights, delegation, separation of duties, conflicts, incentives, benefits, escalation, appeal, remedy, contribution, dependence, accountability, succession, dissolution, and residual custody.	Design rationale	argument	ext_nist_ai_rmf_1_0_2023, ext_moral_crumple_zones_2019, ext_ai_decision_authority_2020, ext_generative_ai_at_work_2025, ext_constructive_interdependence_human_ai_2026, talos, ext_human_ai_team_meta_analysis_2024, ext_human_ai_feedback_loops_2025, ext_eu_ai_civil_liability_2025, adjudicated_persistence	Source notes available for all 10 assigned sources; 7 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 10 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_nist_ai_rmf_1_0_2023: Grounds lifecycle risk-governance roles, responsibilities, and organizational functions. Limits: Framework use does not establish an effective or legitimate organization.; ext_moral_crumple_zones_2019: Grounds the failure in which a proximate human absorbs blame without corresponding authority. Limits: Analysis does not provide a complete accountability mechanism.; ext_ai_decision_authority_2020: Shows in a bounded economic model that human/AI authority allocation changes information-acquisition and reliability incentives. Limits: Formal model, not a universal empirical law or accountability solution.; ext_generative_ai_at_work_2025: Provides field evidence of heterogeneous worker and productivity effects in one customer-support deployment. Limits: No economy-wide, welfare, or long-run organizational conclusion follows.; ext_constructive_interdependence_human_ai_2026: Adds a contribution and dependence measure beyond aggregate task reward in human-agent teaming. Limits: Overcooked and study-specific; no universal teamwork metric.; talos: Supplies Corben’s Labor OS and typed-work lineage for explicit roles, work contracts, delegation, and review. Limits: Speculative local system design; it does not establish organizational effectiveness, legitimacy, accountability, or worker welfare.; adjudicated_persistence: Treats procedures, norms, role definitions, and institutional rules as persistence surfaces requiring legitimacy, contestability, and remedy rather than mere technical write access. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Preregistered representative workflows with real participants and strong manual, automation-first, and nominal-human-review baselines improve joint task outcomes without worse error, rubber-stamping, workload, deskilling, inequitable burden, inaccessible appeal, or accountability mismatch; effects persist longitudinally and reproduce independently without implying legitimacy beyond the studied setting.
human-ai-symbiosis-neurotechnology-and-cognitive-sovereignty.core	human-ai-symbiosis-neurotechnology-and-cognitive-sovereignty	Human-AI symbiosis should be evaluated as a reversible coupled-control intervention: the combined system must beat human-alone and AI-alone baselines on declared outcomes while preserving informed consent, mental integrity, cognitive agency, neural-data purpose limits, skill and exit capacity, equitable access, clinical boundaries, and longitudinal monitoring.	Design rationale	argument	ext_human_ai_team_meta_analysis_2024, ext_human_ai_feedback_loops_2025, ext_oecd_neuro_ai_convergence_2025, ext_who_neurotechnology_landscape_2025	Source notes available for all 4 assigned sources; 4 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 4 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_human_ai_team_meta_analysis_2024: Supplies a preregistered meta-analysis using human-alone, AI-alone, and combination arms and challenges automatic synergy assumptions. Limits: Included tasks and populations do not establish universal or longitudinal human-AI complementarity.; ext_human_ai_feedback_loops_2025: Shows in reported experimental tasks that repeated human-AI interaction can alter later human judgments and create feedback dynamics. Limits: Does not establish permanent change, clinical harm, general social effects, or one direction for every coupling.; ext_oecd_neuro_ai_convergence_2025: Provides intergovernmental analysis of AI/neurotechnology convergence and anticipatory cross-domain governance. Limits: Policy synthesis only; no device, clinical, enhancement, or beneficial-convergence result.; ext_who_neurotechnology_landscape_2025: Supplies global-health, governance, opportunity, risk, access, and equity context for neurotechnology. Limits: Does not establish device efficacy, individual medical advice, safe symbiosis, or authorization for neural-data use.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Durable complementarity beats the strongest component without unacceptable deskilling, dependence, mental-data misuse, inequity, or exit loss and reproduces independently; no clinical, stimulation, enhancement, or universal synergy claim.
ai-deployment-transition-distribution-and-human-agency.core	ai-deployment-transition-distribution-and-human-agency	Consequential deployment should advance only through a prospective transition contract that binds a counterfactual baseline, affected-person denominator, task-role-skill changes, adoption, substitution and complementarity, compensation and ownership, access and prices, concentration, critical-service continuity, human decision rights, training and redeployment, delayed outcomes, remedy, pause conditions, and residuals; exposure, productivity, adoption, or aggregate gain alone establishes neither job loss, welfare, fairness, human agency, nor a successful transition.	Design rationale	argument	coherence_exchange, ext_generative_ai_at_work_2025, ext_ilo_genai_jobs_index_2025, ext_oecd_ai_infrastructure_competition_2025	Source notes available for all 4 assigned sources; 2 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 4 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_generative_ai_at_work_2025: Open peer-reviewed field study of a staggered generative-AI assistant introduction among 5,172 customer-support agents, reporting heterogeneous worker and productivity effects in that setting. It is a bounded deployment comparator and does not establish economy-wide employment, wages, inequality, concentration, long-run skill, or ASI-transition effects. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_ilo_genai_jobs_index_2025: ILO working paper combining task data, worker surveys, expert deliberation, and model-assisted scoring to estimate occupational exposure across countries and groups. Exposure is not realized automation, displacement, welfare, or a forecast of ASI effects, and the study is not a local reproduction. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Useful gains persist without unacceptable subgroup harm, deskilling, dependency, access loss, or agency loss; no economy-wide welfare or ASI-transition forecast.
artifact-graphs-audit-logs-and-replay.core	artifact-graphs-audit-logs-and-replay	Execution should produce an artifact graph with audit logs, provenance, replay metadata, and links to claims and tests.	Design rationale	argument	talos, deterministic_capability_compilation, platonic_world_model, viea, cognitive_compilation, spinoza_composer, genesiscode, cognitive_loop_closure, moecot, ext_w3c_prov_o_2013, ext_mlcommons_croissant_1_1_2026, ext_in_toto_2019, cca_project, moecot_manifest_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, ext_claw_swe_bench_2026, ext_txfs_2018, ext_faithfulness_information_flow_2026, ext_swe_rebench_v2_2026, regret_engine, ext_openclaw_agent_runtime_2026, assurance_shift_learning, adjudicated_persistence	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 25 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	talos: Supports artifact evidence through typed jobs, delivery, evidence records, audit logs, replay, source allow-listing, blind secret handles, tests, residuals, regression coverage, and forgotten-failure prevention. Limits: No Talos artifact graph, audit reconstruction, replay run, benchmark, or security result has been reproduced in this repository.; viea: Supports artifact graphs for claims, requirements, critiques, releases, feedback, tools, benchmarks, residuals, verification ledgers, runtime execution, and feedback loops. Limits: No completed VIEA deployment, runtime trace, artifact graph service, or benchmark result exists here.; cognitive_compilation: Supports build-graph traceability through source plans, typed semantic IR, dependencies, validation requirements, scheduling, target lowering, localized repair, and compiler-style evaluation metrics. Limits: Architecture and evaluation plan only; no working compiler, trace suite, repair-locality benchmark, or empirical ablation exists in this repository.; spinoza_composer: Supports an append-only evidence store, rebuildable interpretation views, draft-versus-active canon, modality and quotation custody, dependency-directed poison recovery, section and scene contracts, causal-versus-presentation time, local and global validation, fallible validator disagreement, element/reference packs, continuity observations, bounded video retries, predeclared partial delivery, internal audit bundles, external credentials, and complete evaluation/cost obligations. Limits: The ten-tab family is an architecture/product lineage. It does not establish extraction or entity-resolution quality, canon correctness, global consistency, validator calibration or independence, visual identity continuity, AMR fidelity, style rights or fidelity, cost forecasts, benchmark targets, ROI, implementation, security, deployment, support promotion, or ASI capability.; genesiscode: Supports provenance-bearing artifacts through canonical CoreForm identity, content-addressed package/evidence records, ordered effect transcripts, semantic patch operations, obligation artifacts, privileged protocol roles, and replay mismatch. The full audit adds stable semantic anchors, expected-old preconditions, dependency/consumer impact closure, affected-obligation revalidation, self-weakening controls, inverse/compensation, response-byte or durable-ref custody, and replay-grade downgrades after redaction/deletion. Limits: No GenesisCode kernel, canonicalizer, response/evidence store, invalidation engine, semantic patch checker, dependency-impact analysis, rollback, deterministic replay system, or proof artifact exists. A response hash alone supports integrity checking, not replay, and content identity does not establish evidence validity.; cognitive_loop_closure: Supports using execution traces as procedural-memory substrate through trajectory logging, loop detection, tool cards, verification, registries, monitoring, lifecycle states, and retirement discipline. Limits: No local loop detector, trace miner, synthesized tool, or verification harness has been executed.; moecot: Supports runtime-reference ledgers, replay, handoff, readiness gates, benchmark artifacts, promotion blockers, and residual tracking as implementation-context for artifact and audit surfaces. Limits: Runtime artifacts, ledgers, logs, and benchmark records have not been imported, inspected, or reproduced here.; ext_w3c_prov_o_2013: Supplies the interoperable entity, activity, agent, derivation, attribution, delegation, revision, and invalidation vocabulary against which the artifact graph must state its additional integrity, authority, receipt-faithfulness, and revocation duties. Limits: PROV-O vocabulary and asserted relations do not establish provenance truth, graph completeness, tamper evidence, authorization, receipt faithfulness, or safe reuse; no local PROV-O graph or conformance run exists.; ext_mlcommons_croissant_1_1_2026: Supplies a current machine-readable dataset-metadata comparator for JSON-LD structure, resources, checksums, record fields, provenance, usage conditions, and ML-tool portability. Limits: Metadata and checksums do not establish complete lineage, fitness, legality, representativeness, safe use, receipt reality, or derivative revocation; no local Croissant record or validation run exists.; ext_in_toto_2019: Supplies the strongest assigned signed-lineage comparator through owner-signed layouts, authorized functionary link metadata, material/product matching, role separation, thresholds, compromise analysis, consumer verification, and last-mile dependencies. Limits: A conforming signed chain does not establish artifact correctness, uncompromised authorized actors, model safety, data fitness, confidentiality, complete layout coverage, or release merit; no local in-toto layout, attestation, or verification run exists.; cca_project: Supports a canonical closure authority with append-only supersession and invalidation, trace bundles, and transitive claim/evidence consequences when upstream artifacts are revoked. Limits: The historical project was mined as design evidence only; its runtime closure, replay behavior, and support transitions were not reproduced or promoted here.; moecot_manifest_project: Supports separating canonical events, graph state, claims, and evidence from disposable read models and watermark caches, with immutable digests, provenance, effect logs, and replay metadata. Limits: The historical Registry V2 architecture was not deployed or replayed in this repository, so it supplies design constraints rather than operational evidence.; bugbrain_project: Supplies concrete failure cases for narrative/report divergence, silently ignored audit writes, non-tamper-evident JSONL, non-cryptographic replay, and path/storage identity confusion. Limits: The mined defects are historical local observations, not a security audit or a claim that every BugBrain execution exhibited them.; corbens_trainer_project: Supports content-addressed lineage, transitive invalidation after quarantine, stale-promotion prevention, and acknowledged asynchronous completion through durable terminal receipts rather than queue idleness. Limits: The historical trainer was not executed here; the chapter imports its failure pattern and design constraints without asserting deployed closure or worker correctness.; corbens_best_model_possible_project: Supports explicit replay grades by separating playback from live re-execution and semantic validation, while exposing trace-only transitions and absolute checkpoint paths as false durability and relocation hazards. Limits: No historical model run or checkpoint was reproduced; these are bounded architecture and failure lessons only.; ext_claw_swe_bench_2026: Supports binding coding-agent outcomes to a fixed model, harness, workspace, patch extraction, evaluator, runtime budget, and cost rather than attributing a harness result to the model alone. Limits: Primary preprint comparator only; no reported task, score, harness, cost, contamination control, or safety result was reproduced here.; ext_txfs_2018: Supports distinguishing declared effect-inventory restoration from ACID filesystem transactions, conflict isolation, crash consistency, durability, and bounded transaction capacity. Limits: TxFS was not installed or reproduced; the local directory and state-tree snapshots do not establish filesystem transactions, crash safety, process recovery, service recovery, or external-effect atomicity.; ext_faithfulness_information_flow_2026: Separates a plausible reasoning transcript from causal reliance by distinguishing sufficiency, completeness, and interventional necessity; this sharpens the artifact graph’s private-reasoning, reported-rationale, action-trace, receipt, and authoritative-effect boundaries. Limits: The paper’s metrics and training interventions are model- and task-bounded; they do not make a chain of thought a receipt or prove local record-reality faithfulness.; ext_swe_rebench_v2_2026: Supplies concrete source, repository, base-commit, solution-patch, test-patch, image, parser, and test-transition identities that a replayable repository task must bind. Limits: Dataset records and image manifests do not prove local replay, independent evaluation, authority-to-effect linkage, or rollback completeness.; ext_openclaw_agent_runtime_2026: Supplies a current implementation comparator for serialized per-session execution, session-generation-checked transcript writes, bounded lifecycle/tool audit projection, and the separation between OpenClaw session identity and external ACP harness resume identity. Limits: Official software and documentation review only; metadata-only audit projection is not replay-complete, ACP execution is not wrapped by the OpenClaw sandbox, and no trace, effect, replay, reliability, or security result was reproduced.; assurance_shift_learning: Adds Boundary Evidence Bundle lineage, append-only lifecycle, supersession, repair descendants, preserved prefixes, counterexamples, and replay provenance. Limits: A complete-looking bundle does not establish source truth, causal adequacy, or receipt faithfulness.; adjudicated_persistence: Adds lineage requirements connecting experience, lesson, disposition, realization, qualification, authority, descendants, invalidation, and residuals. Limits: Used only to specify lineage links among persistence objects and descendants; no replay completeness, causal recovery, or residual closure was demonstrated.	causal influence-removal evidence; descendant-wide propagation receipts; effect-complete rollback beyond the local inventory; external effect and remote backup recovery; independent replay; independently assessed verifier quality; privacy and member/nonmember assessment; production recovery objectives and monitoring; production transfer and governance-cost evidence; replication across heterogeneous runtimes; useful governed throughput at the registered frontier; verified storage and backup erasure where claimed	A deployed or externally reviewable artifact service reconstructs real work and revocation effects without erasing verifier limits, failed rollback, or residuals.
runtime-adapters-tool-permissions-and-human-approval.core	runtime-adapters-tool-permissions-and-human-approval	Runtime adapters should enforce typed permissions, sandboxing, human approval, and post-action evidence capture.	Design rationale	argument	talos, deterministic_capability_compilation, platonic_world_model, viea, ladon_manhattan, software_magic_grimoire, genesiscode, moecot, field_of_god_ai_constitution, theseus_operator_os, ext_humans_automation_1997, ext_ironies_automation_1983, ext_levels_automation_2000, ext_complacency_bias_automation_2010, ext_cap_theorem_gilbert_lynch_2002, ext_inspect_ai_2024, ext_agentdojo_2024, ext_camel_prompt_injection_2025, ext_owasp_agentic_top_10_2026, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, ext_claw_swe_bench_2026, ext_txfs_2018, qcsa_whitepaper, reflexive_router_whitepaper, ext_elizaos_agent_runtime_2026, ext_hermes_agent_2026, ext_openclaw_agent_runtime_2026	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 31 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	talos: Supports controlled runtime adapters through typed jobs, tool isolation, contract locks, human approval, blind secret handles, Digital SCIFs, evidence records, audit logs, replay, and promotion through tests/residuals/regressions. Limits: No live adapter runner, approval service, sandbox, security test, or Talos runtime result has been reproduced here.; viea: Supports intent-to-runtime boundaries through structured command contracts, runtime adapters, verified outputs, deployment feedback, residuals, tools, benchmarks, and regression coverage. Limits: No deployed runtime adapter, feedback loop, or verified VIEA execution trace exists in this repository.; ladon_manhattan: Supports handle-mediated authority through hardened secret managers, opaque credential handles, policy-mediated injection, ephemeral compartments, zeroization, outbound-call interception, and side-channel assumptions. Limits: Security architecture only; no Ladon implementation, kernel-level test, side-channel validation, or security audit exists here.; software_magic_grimoire: Supports adapter workflow discipline through guarded entry/exit rules, handoff artifacts, verification and failure behavior, incompatible-mode separation, loops that advance on evidence, and scoped recursion. Limits: Workflow vocabulary only; does not prove adapter enforcement, approval quality, or tool safety.; genesiscode: Supports capability-mediated effects through authenticated privileged control variants, explicit Pure/Perform effect programs, deny-by-default host capabilities, ordered request/decision/response records, replay mismatch, and AI calls as governed effects. The full audit adds the separation of protocol authenticity from authorization, fresh-token minting from pure sealing, response-hash verification from replay-complete custody, cached model output from call reproduction, and observed trace coverage from static/runtime/OS enforcement. Limits: No seal issuer/custody/revocation system, capability runner, sandbox, complete response transcript, replay checker, bypass/FFI analysis, OS enforcement, model-call reproduction, or security proof exists. Nominal sealed variants and clean observed traces do not establish unforgeability, permission correctness, or full mediation.; moecot: Supports implementation-reference context for specialist lanes, fail-closed control-plane ledgers, readiness gates, replay, handoff, promotion blockers, and authority-sensitive runtime orchestration. Limits: Runtime artifacts and benchmark/replay logs remain uninspected and unreproduced in this book repo.; field_of_god_ai_constitution: Supports governing tool use, power, memory, delegation, consent, reversibility, least sufficient power, risk tiers, auditability, human review, and no self-authorization of sovereignty. Limits: Constitutional specification only; no runtime policy engine, red-team suite, system-prompt evaluation, or proof was run here.; theseus_operator_os: Supports operator-visible runtime control through shared command vocabulary, durable work board, node registry, background/watch/wake contracts, skill registry, tool hooks, feedback routing, TTLs, kill switches, signed updates, and isolation. Limits: No Hive board, SQLite database, node registry, command channel, hook ledger, or dashboard was run from this repository.; ext_humans_automation_1997: Positions human approval against automation use, misuse, disuse, abuse, overreliance, monitoring failure, trust, workload, risk, and false-alarm dynamics. Limits: External human-factors comparator only; no local human-subjects study, approval-service result, or deployed reviewer-quality evidence exists here.; ext_ironies_automation_1983: Positions human approval against the automation-irony problem: automation can move humans into harder supervisory and abnormal-condition duties. Limits: External conceptual comparator only; no local operator study, deployed approval service, or live supervisory-control result exists here.; ext_levels_automation_2000: Positions approval records against automation levels across information acquisition, analysis, decision/action selection, and action implementation. Limits: External taxonomy comparator only; the book does not claim a deployed adapter service implements every automation level.; ext_complacency_bias_automation_2010: Positions approval fatigue, rubber-stamping, and automation bias against external literature on automation complacency, omission/commission errors, workload, attention, and imperfect decision aids. Limits: External human-factors comparator only; no local reviewer-quality benchmark, live approval workflow, or human-subjects result is claimed.; ext_cap_theorem_gilbert_lynch_2002: Supplies the distributed safety-versus-availability comparator for stale grants, revocation delay, partitioned approval state, fresh authority receipts, quarantine, and no-mutation evidence at the effect boundary. Limits: CAP does not prove a local authority protocol, consensus, partition tolerance, revocation propagation, availability, or safety; no distributed authority service or live partition was run.; ext_inspect_ai_2024: Supplies a current composable evaluation-framework comparator for tasks, datasets, solvers, scorers, agents, tools, logs, and sandboxes around adapter behavior. Limits: Framework availability and task passes do not establish workload validity, sandbox enforcement, coverage, safety, or readiness; no local Inspect task, result, or sandbox trace exists.; ext_agentdojo_2024: Supplies a tool-using-agent security benchmark comparator over untrusted data, security tasks, prompt-injection attacks, and defenses. Limits: Benchmark results do not establish attack completeness, local reproduction, deployed robustness, correct authority handling, or system safety; no AgentDojo task or defense ran here.; ext_camel_prompt_injection_2025: Supplies the closest assigned capability-enforcement comparator through trusted-control/untrusted-data separation, policy extraction, and tool-call capability checks. Limits: The source-reported results do not prove universal injection resistance, policy-extraction correctness, local implementation, or safe deployment; no local CaMeL path ran here.; ext_owasp_agentic_top_10_2026: Supplies a current threat-taxonomy comparator spanning goal hijacking, tool misuse, identity abuse, supply-chain risk, code execution, memory poisoning, inter-agent communication, cascading failures, human trust exploitation, and rogue agents. Limits: Taxonomy coverage does not establish threat-model completeness, control effectiveness, local testing, or safety; no local OWASP assessment was performed.; moecot_manifest_project: Supports bounded local-lineage context for typed runtime manifests, effect records, replay checks, target portability, scoped resources, and revocable authority artifacts. Limits: Does not establish OS sandbox enforcement, network isolation, live adapter safety, secret-handle safety, deployed replay, or support promotion.; beastbrain_project: Supports local negative-case context for declared interfaces, simulations, and nominal security/tool states that must not be described as enforced external-effect boundaries. Limits: Does not establish process confinement, OS enforcement, adapter effects, tool safety, runtime capability, or independent confirmation.; bugbrain_project: Supports local negative-case context for executable and path identity, protocol security versus enforcement, privileged action scope, skipped checks, and report/canonical-state drift. Limits: Does not establish a hardware or OS sandbox, network isolation, approval correctness, adapter safety, or deployed replay.; corbens_trainer_project: Supports bounded local-lineage context for typed backend manifests, artifact quarantine, content-addressed execution records, revocable promotion, and stale-evidence invalidation. Limits: Does not establish a sandboxed trainer backend, OS enforcement, live adapter effect, network isolation, replay equivalence, or support movement.; corbens_best_model_possible_project: Supports local negative-case context for playback versus live replay, absolute-path and relocation failures, default or empty tool paths, digest-parity laundering, and missing runtime causality. Limits: Does not establish live tool execution, sandbox isolation, causal adapter binding, network or secret safety, model capability, or support promotion.; ext_claw_swe_bench_2026: Supports binding coding-agent outcomes to a fixed model, harness, workspace, patch extraction, evaluator, runtime budget, and cost rather than attributing a harness result to the model alone. Limits: Primary preprint comparator only; no reported task, score, harness, cost, contamination control, or safety result was reproduced here.; ext_txfs_2018: Supports distinguishing declared effect-inventory restoration from ACID filesystem transactions, conflict isolation, crash consistency, durability, and bounded transaction capacity. Limits: TxFS was not installed or reproduced; the local directory and state-tree snapshots do not establish filesystem transactions, crash safety, process recovery, service recovery, or external-effect atomicity.; qcsa_whitepaper: Adds the hard capability-separation rule: semantic resolution, aliases, address confidence, and certificates never grant authority; the physical route must independently bind actor, target, operation, scope, policy, data, reversibility, approvals, and receipts. The later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. One reversible local effect does not establish permission correctness, production safety, human-approval quality, chapter-core promotion, AGI, or ASI.; ext_elizaos_agent_runtime_2026: Supplies a current implementation comparator for modular actions, providers, evaluators, services, runtime lifecycle, guarded host access, and scenario execution with an explicit boundary between in-process diagnostics and externally qualified provider evidence. Limits: Official software and documentation review only; no plugin, permission, sandbox, scenario, provider qualification, security, or production result was executed or reproduced.; ext_hermes_agent_2026: Supplies a current implementation comparator for dangerous-command approval, file-write restrictions, container isolation, credential filtering, context scanning, cross-session isolation, tool backends, and optional staged approval of agent-authored skill mutations. Limits: Official software and documentation review only; no command, sandbox, tool, approval, skill-write, security, or deployment result was reproduced.; ext_openclaw_agent_runtime_2026: Supplies a current implementation comparator that separates sandbox placement, tool availability policy, approval, and elevated execution, while documenting that ACP external harnesses retain their own execution authorization and are not wrapped by the parent sandbox. Limits: Official software and documentation review only; no policy, sandbox, approval, elevation, ACP confinement, effect observation, or security result was reproduced.	effect-complete rollback beyond the local inventory; independently assessed verifier quality; production transfer and governance-cost evidence; useful governed throughput at the registered frontier	High-impact adapter call fails without current approval and receipt.
embodied-agency-real-time-control-and-physical-safety.core	embodied-agency-real-time-control-and-physical-safety	Physical execution requires a plant-specific control lease binding embodiment, workspace, state estimator, dynamics, timing, force/space/contact limits, human presence, advanced/baseline/stop controllers, switching and interlocks, exploration, degraded modes, observed effects, compensation, irreversible residuals, costs, and expiry.	Design rationale	argument	rmi, cognitive_loop_closure, ext_gemini_robotics_2025, ext_foundation_robotics_physical_risk_2025, ext_control_barrier_functions_2019, ext_simplex_architecture_1998, ext_safe_reinforcement_learning_survey_2015, ext_ai_simulation_digital_twins_2025, viea	Source notes available for all 9 assigned sources; 9 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 9 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_ai_simulation_digital_twins_2025: Surveys AI-enabled simulation and digital-twin patterns relevant to plant models, telemetry synchronization, scenario testing, and sim-to-real residuals. Limits: A digital twin is a maintained model, not the physical system; fidelity, synchronization, coverage, and transfer must be measured for each use.; ext_gemini_robotics_2025: Supplies a modern vision-language-action and embodied-reasoning capability comparator. Limits: Source-reported results do not provide independent physical-safety assurance or local transfer.; ext_foundation_robotics_physical_risk_2025: Grounds physical-risk families for robotics foundation models. Limits: Survey evidence does not validate a mitigation or local controller.; ext_control_barrier_functions_2019: Provides conditional safety-set enforcement through control barrier functions. Limits: Guarantees require modeled dynamics, state, safe set, feasibility, timing, and correct implementation.; ext_simplex_architecture_1998: Provides independent advanced/baseline controller and switching architecture. Limits: Bounded process-control precedent, not a universal learned-controller result.; ext_safe_reinforcement_learning_survey_2015: Separates modified objectives from constrained exploration in safe RL. Limits: Survey taxonomy does not establish safe learning or deployment.; viea: Supplies Corben’s intent-to-runtime adapter and observed-effect lineage for separating semantic requests, execution, and receipts. Limits: Speculative local architecture source; it provides no real-time control, plant safety, or physical-effect result.; rmi: Supplies a hierarchical embodied-log design separating raw telemetry, salient events, semantic state, active skill/controller identity, and residuals so loop learning and safety review do not consume an undifferentiated sensor stream. Limits: Conceptual architecture only; no logger, event detector, semantic encoder, raw-ring buffer, trace join, incident reconstruction, storage result, privacy result, physical controller, or hardware campaign has been implemented or measured.; cognitive_loop_closure: Supplies the detailed embodied trajectory policy: raw telemetry separated from cognitive/event/residual logs, rolling multi-resolution buffers, triggered pre/post-event retention, semantic eventization, feature compression, reflex-state custody, memory budgets, and explicit warning against loss of safety-relevant information. Limits: Conceptual architecture only; no telemetry corpus, eventizer, feature compressor, raw-buffer policy, reflex logger, incident reconstruction, privacy evaluation, real-time controller, simulator, or hardware result has been implemented or measured.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Across representative plants and preregistered sensing, timing, saturation, communication, power, payload, human-entry, and model-mismatch faults, the governed controller reduces safety-envelope violations without unacceptable task, latency, nuisance-stop, or operator cost; fallback is demonstrated reachable under deadline, effects are independently observed, irreversible residuals remain explicit, and transfer reproduces without a universal safety claim.
inter-stack-protocols-identity-and-economic-exchange.core	inter-stack-protocols-identity-and-economic-exchange	A governed stack routes each cross-stack request through a versioned exchange contract that binds protocol and schema version, sender and receiver identities, endpoint and capability declaration, requested task or artifact, principal and delegated authority, credential verification, audience, scope, expiry, budget or consideration, expected receipt, dispute and revocation paths, and residual owner; an absent, mismatched, expired, revoked, unverified, or budget-unreserved required record blocks dispatch or routes accountable review, but does not itself establish peer trustworthiness, task or artifact truth, effect safety, payment settlement, legal validity, economic fairness, privacy, authorization correctness, or ASI.	Design rationale	argument	ext_mcp_protocol_2025_11_25, platonic_world_model, ext_a2a_protocol_1_0_0, ext_w3c_did_core_1_0_2022, ext_w3c_vc_data_model_2_0_2025, ext_interledger_protocol_v4, talos, vcm_public, tokenmana, ext_owasp_agentic_top_10_2026, qcsa_whitepaper, ext_openclaw_agent_runtime_2026	Source notes available for all 12 assigned sources; 11 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 12 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_mcp_protocol_2025_11_25: Supports the latest released MCP comparator for versioned messages, lifecycle, capability negotiation, authorization/discovery, elicitation, tasks, and modular agent-to-tool protocol features. Limits: The 2026-07-28 revision remains a release candidate; this source does not establish a local MCP implementation, authorization, task truth, effect safety, payment, deployment, or ASI.; ext_a2a_protocol_1_0_0: Supports the latest released A2A comparator for canonical objects, version negotiation, Agent Cards, tasks/messages/artifacts, multiple bindings, authorization scoping, interoperability testing, and security considerations. Limits: Does not establish a local A2A peer, verified identity, delegated authority, task correctness, binding interoperability, payment, safety, deployment, or ASI.; ext_w3c_did_core_1_0_2022: Supports a scoped comparator for identifiers, controllers, verification methods, service endpoints, resolution, and privacy vocabulary. Limits: Does not establish a local DID, resolver, controller trust, authorization, revocation, safety, deployment, or ASI.; ext_w3c_vc_data_model_2_0_2025: Supports a scoped comparator for issuer, holder, verifier, credential, presentation, validity, status, evidence, securing-mechanism, and authorization-limit vocabulary. Limits: Does not establish a local credential, issuer, verifier, trust decision, authorization framework, delegation validity, payment, safety, deployment, or ASI.; ext_interledger_protocol_v4: Supports a scoped comparator for packetized cross-network value exchange, connector obligations, account/balance vocabulary, and end-to-end separation between transport and application policy. Limits: Does not establish a local payment, account, settlement, delegated authority, economic fairness, legal transfer, safety, deployment, or ASI.; talos: Supports bounded local source context for typed jobs, contract locks, controlled handoffs, audit, replay, delivery, and residual feedback. Limits: Does not establish a local peer exchange, protocol, identity, credential, payment, settlement, delegated authority, deployment, or ASI.; vcm_public: Supports bounded local source context for source and authority binding, context references, adequacy/admission separation, revocation, invalidation, and non-success states. Limits: Does not establish a local cross-stack identity, credential verification, delegated authority, peer trust, payment, deployment, or ASI.; tokenmana: Supports bounded local source context for capacity stocks, bounded bursts, resource pressure, pricing hypotheses, and human/resource constraints around an economic request. Limits: Does not establish a local price, payment, settlement, delegated authority, economic fairness, welfare outcome, deployment, or ASI.; ext_owasp_agentic_top_10_2026: Supplies a current threat-taxonomy comparator for agent goal hijacking, tool misuse, identity and privilege abuse, supply-chain compromise, unexpected code execution, memory poisoning, insecure inter-agent communication, cascading failures, human-trust exploitation, and rogue behavior that an exchange contract must expose to prevention, detection, containment, and recovery owners. Limits: A threat taxonomy does not establish completeness, control effectiveness, local testing, safe interoperability, authorization correctness, privacy, deployment, or ASI.; qcsa_whitepaper: Supplies namespace-qualified stable identities, signed SACs, atlas epochs, mapping manifests, migration lineage, and uncertainty-preserving translation as candidate cross-stack semantic exchange records; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. No peer federation, trust, settlement, privacy, chapter-core promotion, AGI, or ASI result is established.; ext_openclaw_agent_runtime_2026: Supplies a current implementation comparator for typed gateway requests and events, device identity and pairing, idempotency keys, ACP session routing, and the explicit distinction between a local channel/session key and a harness-local resume identifier whose upstream loading authorization belongs to the external backend. Limits: Official software and documentation review only; protocol shape, pairing, and policy checks do not establish peer trust, delegated authority correctness, harness confinement, delivery, settlement, or end-to-end effect safety, and no local run was reproduced.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Measure false dispatch, false denial, stale acceptance, dispute recovery, latency, disclosure, operator effort, and residual obligations; no peer trust, task truth, effect safety, payment, settlement, privacy, authorization, or deployment claim.
multi-agent-dynamics-collective-intelligence-and-systemic-risk.core	multi-agent-dynamics-collective-intelligence-and-systemic-risk	Expanded multi-agent interaction requires a population contract binding agent/owner/model/organization identities, human participants, interaction and dependency graphs, incentives, information, resources, commitments, decision assumptions, entry/exit/copying/learning, population outcomes, externalities, human influence, interventions, costs, and residuals.	Design rationale	argument	learning_compute_topology, ext_multi_agent_risks_2025, ext_cooperative_ai_foundations_2023, ext_gradual_disempowerment_2025, ext_constructive_interdependence_human_ai_2026, ext_functional_decision_theory_2017, coherence_exchange	Source notes available for all 7 assigned sources; 6 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 7 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_multi_agent_risks_2025: Grounds miscoordination, conflict, collusion, and seven population-level risk factors. Limits: Taxonomy and cited evidence do not establish a local population result.; ext_cooperative_ai_foundations_2023: Provides game-theoretic cooperation questions and the distinction between beneficial coordination and harmful collusion. Limits: Research agenda, not a solved mechanism or reproduced result.; ext_gradual_disempowerment_2025: Grounds slow cumulative loss of human influence through individually reasonable decisions. Limits: Risk scenario and analysis, not a measured local social trend.; ext_constructive_interdependence_human_ai_2026: Adds a task-reward-independent measure of contribution and dependence in human-agent teams. Limits: Domain-, agent-, participant-, and metric-bound; not reproduced.; ext_functional_decision_theory_2017: Supplies a distinct decision rule for logically linked choices and exposes decision-theory disagreement. Limits: Normative proposal, not universally accepted or empirically validated.; coherence_exchange: Supplies Corben’s inter-stack exchange, settlement, audit, and fork/exit lineage for interactions among independently governed systems. Limits: Speculative local protocol source; it provides no population-level cooperation, collusion, concentration, or systemic-risk result.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Across multiple seeds, model families, population sizes, games, and topologies, preregistered population interventions detect positive-controlled collusion, cascade, and common-mode risk and improve beneficial cooperation over strong pairwise and aggregate-reward baselines without hiding externalities, concentration, exclusion, or human-agency loss; results reproduce independently and remain bounded to the tested populations rather than forecasting society.
procedural-memory-and-cognitive-loop-closure.core	procedural-memory-and-cognitive-loop-closure	Cognitive loop closure compiles repeated reasoning into verified parameterized tools and procedural memory.	Design rationale	argument	cognitive_loop_closure, deterministic_capability_compilation, platonic_world_model, rmi, rgs, benchmaxxing, talos, moecot, project_theseus_whitepaper, theseus_self_evolution_system, theseus_operator_os, cca_project, moecot_manifest_project, beastbrain_project, corbens_trainer_project, ext_memgpt_2023, ext_toolformer_2023, ext_voyager_2023, ext_dreamcoder_2020, ext_mem0_2025, reflexive_router_whitepaper, kernel_english_residual_compiler, relational_dimension_compiler, capability_ratchet_whitepaper, regret_engine, ext_hermes_agent_2026, ext_openclaw_agent_runtime_2026, assurance_shift_learning, adjudicated_persistence, forward_transfer_program_synthesis	Source notes available for all 30 assigned sources; 24 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 30 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	cognitive_loop_closure: Supports the core loop-closure mechanism through complete trajectory logging, loop detection, invariant abstraction, six active parameter probes and four variable states, deterministic or bounded model-assisted tool synthesis, verification portfolios, orthogonal assurance/runtime/latency/risk axes, registry, per-invocation routing, monitoring, revision/retirement, and interpreter/compiled/reflex modes. Limits: Conceptual architecture source only; no local trajectory corpus, loop detector, abstraction or parameter-discovery evaluation, synthesized tool, verifier-quality campaign, routing monitor, savings result, physical logger, or retirement automation has been executed.; rmi: Supports ratcheted procedural memory through benchmark pressure, routed specialist attempts, logs, loop closure, residual escrow, regression preservation, arm/router lifecycle, and interpreter/compiled-tool/reflex modes. Limits: Conceptual architecture source only; no independent reproduction, benchmark run, prototype inspection, or ASI behavior is proven here.; rgs: Supports promoting repeated successful procedures into verified tools while preserving residuals, using regression suites, and keeping ledgers for benchmarks, models, tools, residuals, calibration, safety, routing, and interventions. Limits: No local benchmark runs, tool-promotion traces, regression data, or implemented ratchet are present.; benchmaxxing: Supports regression and promotion discipline through benchmark lifecycle states, wall diagnosis, benchmark/model ledgers, anti-Goodhart safeguards, residual accounting, and architecture-change discipline. Limits: No benchmark harness, mutation, holdout, empirical run, or model/prototype result was produced here.; talos: Supports procedural memory as typed, audited execution work through job lifecycles, adjudication, delivery, feedback/autopoiesis, audit logs, replay, tests, residuals, and regression coverage. Limits: No Talos loop-closure pipeline, scheduler, replay log, or benchmark result has been reproduced in this repository.; moecot: Supports runtime-reference context for procedural reuse through specialist lanes, readiness gates, ledgers, replay, handoff, promotion blockers, residual tracking, and current limitation reporting. Limits: MoECOT runtime artifacts, code, logs, readiness records, and benchmark results have not been imported or reproduced.; project_theseus_whitepaper: Supports report-first loop closure through repeated trajectory closure into verified tools after parameters, preconditions, postconditions, verification, risk, and retirement criteria are recorded. Limits: Source-reported implementation state only; current Theseus reports, ledgers, code paths, and command outputs were not rerun from this repo.; theseus_self_evolution_system: Supports evidence-gated procedural improvement through intervention ladders, loop closure into local verified tools, outcome ledgers, lifecycle-governed cells, branch isolation, local checks, and trace ledgers. Limits: No Theseus scripts, ATTD reports, branch-creation flow, teacher runner, outcome ledger, or dashboard endpoint was executed here.; theseus_operator_os: Supports procedural-memory operations through durable work-board state, goals with budgets/judges/steps/stop conditions/dependencies/events, skill registry, tool hooks, feedback routing, and safety-visible operator surfaces. Limits: No Hive board, node registry, command channel, database, or dashboard was run from this repository.; cca_project: Supports compiled-procedure lineage, durable trace identity, and separation of semantic knowledge from reusable execution structure. Limits: No CCA procedure-mining run, tool promotion, or historical runtime behavior is reproduced.; moecot_manifest_project: Supports manifest-bound procedure identity, source/effect receipts, qualification gates, rollback, and registry lifecycle discipline. Limits: No MoECOT Manifest compiler, runtime, regression suite, or rollback execution is reproduced.; beastbrain_project: Supports durable semantic-memory pressure, procedure/tool separation, failure retention, and explicit lifecycle state. Limits: No BeastBrain memory store, loop detector, or tool-promotion behavior is reproduced.; corbens_trainer_project: Supports regression evidence, failed-attempt retention, checkpoint/effect acknowledgement, quarantine, and revocable promotion. Limits: No trainer run, regression-quality result, checkpoint effect, or model capability is reproduced.; ext_memgpt_2023: Supplies an external comparator for explicit memory tiers, context-window scarcity, controlled data movement, and long-running-agent memory management while keeping fact storage and procedure qualification separate. Limits: Abstract and metadata review only; no MemGPT code, data, memory trace, document task, chat task, adequacy result, or procedural-memory result is reproduced.; ext_toolformer_2023: Supplies an external comparator for learning when and how to call tools from demonstrations, separating learned tool-use skill from permission, result trust, replay, regression preservation, and support. Limits: Abstract and metadata review only; no Toolformer training, API traces, model, score, learned policy, usefulness, or safety result is reproduced.; ext_voyager_2023: Supplies a full-paper-reviewed comparator for curriculum-driven executable skill libraries, environment feedback, execution errors, self-verification, bounded refinement, skill retrieval, ablations, transfer, costs, and failure reports. Limits: No Voyager code, prompts, Minecraft environment, skill library, model calls, tasks, scores, transfer, independent evaluation, or safety result is reproduced.; ext_dreamcoder_2020: Supplies an external comparator for program synthesis, reusable library induction, wake-sleep learning, interpretable abstraction, and reduced later search burden. Limits: Abstract and metadata review only; no DreamCoder code, tasks, synthesis run, learned library, search reduction, generalization, or model result is reproduced.; ext_mem0_2025: Supplies a metadata-level comparator for memory extraction, consolidation, retrieval, graph linkage, cross-session memory, latency, and token-cost pressure while distinguishing conversational or episodic memory from procedures. Limits: Metadata-first intake only; no paper passage review, Mem0 implementation, LOCOMO run, LLM-judge result, correctness, privacy, latency, token-cost, production, or procedural-memory result is reproduced.; relational_dimension_compiler: Adds a slow-path-to-specialist lifecycle in which repeatedly qualified relation programs become monitored recognizers, indexes, rules, kernels, macro-objects, or specialists while retaining scope, counterexamples, expiry, requalification, rollback, and the original slow route. Limits: No relation program was compiled, no specialist equivalence or cost saving was measured, and no shift detector or rollback path was exercised.; ext_hermes_agent_2026: Supplies a current implementation comparator for progressive-disclosure skills, agent-managed create/patch/edit/delete operations, post-session review, and an optional durable write-approval queue that separates a proposed procedure mutation from its application. Limits: Official software and documentation review only; no learned procedure was replayed, independently tested, promoted, rolled back, or measured for utility, safety, transfer, or regression.; ext_openclaw_agent_runtime_2026: Supplies a current implementation comparator for isolated trajectory review, bounded evidence exposure, one-mutation pending proposals, exact-target hash binding, stale-target rejection, security rescanning, rollback metadata, abstention, and skill lifecycle aging. Limits: Official software and documentation review only; reviewer isolation and proposal controls do not establish procedure correctness, evaluator independence, held-out transfer, safe default routing, or beneficial recursive improvement, and no local result was reproduced.; assurance_shift_learning: Adds a route for compiling mature exception families into tests, guards, procedures, tools, and recovery routines while preserving evidence, protected positives, expiry, and reactivation triggers. Limits: No compiler, recurrence reduction, safe consolidation, or efficiency gain was implemented or measured.; adjudicated_persistence: Adds the prior adjudication step that determines whether a repeated cognitive pattern is eligible to become a procedure or tool. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.; forward_transfer_program_synthesis: Adds the verified forward-transfer intervention as the admission test for reusable procedures, with four-value separation, an R0-R7 reuse ladder, frozen knowledge versions, matched controls, and later-round admission. Limits: Framework and experimental blueprint only; no candidate procedure, synthesis benchmark, forward-transfer effect, cost advantage, or support transition was produced.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Memory update fixture preserves failed attempts and rollback path.
routing-heads-and-specialist-cores.core	routing-heads-and-specialist-cores	ASI scales through a lightweight routing head that selects bounded specialist cores with local tools, memory, and authority.	Design rationale	argument	learning_compute_topology, octopus_router, deterministic_capability_compilation, rmi, beastbrain, cognitive_loop_closure, rgs, benchmaxxing, viea, scf, talos, moecot_md, moecot, project_theseus_whitepaper, theseus_operator_os, theseus_architecture_gate, ext_hipporag_2024, ext_dont_hallucinate_abstain_2024, qcsa_whitepaper, reflexive_router_whitepaper, relational_dimension_compiler, ext_kimi_k3_2026, portia_synapse, spider_synapse, coilmoecot	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 25 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	octopus_router: Supports the core routing-head/specialist-arm design through resident head/router selection, dynamically loaded arms, local tools, memory, benchmarks, residuals, permissions, runtime tiers, verification contracts, routing patterns, permission envelopes, memory routing, arm cards, and split/merge/retire policies. Limits: Conceptual architecture source only; no routed-specialist prototype, learned-router harness, runtime authority enforcement, or routing benchmark has been run here.; rmi: Supports modular routing through head/router, specialist arms, arm registry, memory router, permission router, tool registry, benchmark ledger, residual escrow, verification/safety layer, lifecycle states, and regression-floor preservation. Limits: No independent reproduction, benchmark run, prototype inspection, or empirical specialist-routing result exists in this repository.; beastbrain: Supports whole-stack routing pressure through local, stateful intelligence split across memory, verification, planning, routing, security, consolidation, interface surfaces, semantic intent, intelligence tiers, and capability routes. Limits: Architecture lineage only; hardware, cost, context-length, security, and benchmark claims are source-reported and not locally reproduced.; cognitive_loop_closure: Supports routable specialist tools through verified procedural memory, tool registry, router, runtime monitor, execution modes, tool cards, risk tiers, lifecycle states, and retirement discipline. Limits: No local loop detector, tool synthesis, router monitor, or verified procedural tool has been executed here.; rgs: Supports routing within a ratcheting system that logs attempts, classifies successes/failures, promotes verified tools, preserves residuals, maintains regressions, and keeps ledgers for routing and interventions. Limits: No local benchmark runs, tool-promotion traces, regression data, or implemented ratchet are present.; benchmaxxing: Supports runtime promotion discipline through benchmark lifecycle, wall diagnosis, model/benchmark ledgers, anti-Goodhart safeguards, regression conversion, residual preservation, and architecture-change criteria. Limits: No local benchmark harness, mutation, holdout, run, or reproduced performance claim exists here.; viea: Supports connecting MoECOT to the broader intent-to-execution spine: command contracts, artifacts, specialist-routed work, verified outputs, runtime execution, feedback, residuals, tools, benchmarks, and regression coverage. Limits: No completed VIEA deployment, runtime trace, or verified execution result exists in this repository.; scf: Supports treating runtime cores as replaceable capability implementations behind stable contracts, evidence registries, scoped qualifications, route validation, lifecycle events, incidents, appeals, rollback, and authority boundaries. Limits: Does not prove MoECOT core safety, evaluator quality, or production replacement behavior.; talos: Supports runtime execution discipline through typed jobs, deterministic control planes, factory execution, adjudication, delivery, feedback, evidence records, audit logs, replay, tool isolation, and approval gates. Limits: No Talos runtime, MoECOT-Talos integration, replay run, security result, or benchmark has been reproduced here.; moecot_md: Supports terminology normalization for compact orchestrator, specialist lanes, control-plane ledgers, readiness gates, replay, and fail-closed promotion as a Markdown/export variant of the MoECOT source. Limits: Variant source only; it is not independent corroboration and does not verify benchmark, readiness, or runtime behavior.; moecot: Supports the folded MoECOT runtime crosswalk inside the routing claim through compact orchestrator, specialist lanes, fail-closed control plane, run/task/control-plane ledgers, readiness gates, replay, handoff, promotion blockers, residual tracking, and explicit limitations. Limits: Readable source note only; code, logs, benchmark artifacts, replay records, and release artifacts have not been imported, inspected, or reproduced.; project_theseus_whitepaper: Supports local-first routed specialists through Octopus head router, bounded specialist arms, schemas, permissions, local memory, benchmarks, residuals, reliability scores, lifecycle state, retirement criteria, and trusted-node task allowlists. Limits: Source-reported implementation narrative only; current Theseus reports, ledgers, code paths, and command outputs were not rerun from this repo.; theseus_operator_os: Supports operator-visible routing through durable work boards, node registries, command-channel parity, skill registry, tool hooks, feedback routing, safety gates, TTLs, kill switches, signed updates, and isolation surfaces. Limits: No Hive board, SQLite database, node registry, command channel, hook ledger, or dashboard was run from this repository.; theseus_architecture_gate: Supports runtime-readiness gating through pre-training checks for ratchet completeness, router readiness, safety ledger, regression suite, residual escrow, bridge benchmarks, procedural tools, routing memory, lifecycle governance, and external-inference zero. Limits: Reported gate snapshot is not reproduced evidence; current gate artifacts were not regenerated or independently inspected here.; ext_hipporag_2024: Supplies a metadata-level associative retrieval comparator for knowledge-graph navigation and Personalized PageRank, clarifying that relevance or memory-route selection is not evidential adequacy, correct update, authority, or routing safety. Limits: Metadata-first intake only; no paper passage review, HippoRAG implementation, graph build, retrieval task, multi-hop QA result, poisoning test, routing result, or local performance is reproduced.; ext_dont_hallucinate_abstain_2024: Supports measuring abstention with coverage, accuracy, calibration, and useful response behavior while treating self-reflection and model agreement as fallible evidence. Limits: The ACL models, prompts, domains, collaboration schemes, and reported gains were not reproduced; the local single-model router is not an independent multi-model panel.; qcsa_whitepaper: Separates semantic identity and address selection from the physical expert/model route, while making fallback, abstention, load, cost, hardware, permissions, and verifier requirements explicit in the compiled route plan; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. No learned-router, trained-specialist, production, chapter-core promotion, AGI, or ASI result is established.; relational_dimension_compiler: Extends routing to candidate topology, relational order, operator family, factorization, precision, scale, branch resolution, verification effort, and stopping or abstention under a complete proposal-through-repair cost ledger. Limits: No learned order router, proposal-recall result, regret bound, useful polyadic route, or natural-task routing advantage is demonstrated.; portia_synapse: TreeLLM successor architecture for a replacement-compatible Scout, Focus, refinement, coordinate, edge, and confidence route with phased training, diagnostics, stable traits, registry selection, and fallback. Limits: Conflicting source test counts, incomplete integration and comparative benchmarking, possible batch-coupled attention, mutable-memory isolation, and weak exact task metrics remain unresolved; no local route learning is established.; spider_synapse: Failed predecessor case for multi-hypothesis routing, repeated refinement, mutable working memory, selection, typed output heads, path diagnostics, and one-path recovery. Limits: Its reported plateau does not identify branching, refinement, memory, selector credit, label smoothing, or supervision geometry as the cause and has not been locally reproduced.	independently assessed evaluator validity; multiple models, seeds, and natural workloads; nonzero substantive answer utility; production routing safety, latency, cost, and transfer	Learned, rule, specialist, generalist, fallback, and abstention routes separate under ambiguous held-out work with matched cost, calibration, interference, residual, and authority records.
replaceable-cognitive-substrates-beyond-transformer-monoculture.core	replaceable-cognitive-substrates-beyond-transformer-monoculture	For an exact task family, consumer, modality, model and state version, memory contract, hardware/runtime, authority and rights envelope, resource budget, evaluator, fallback, rollback, and time horizon, learned cognition should be supplied through a typed Cognitive Kernel ABI whose implementations are admitted only by matched strong baselines, exact state and memory semantics, proposal-versus-effect separation, evaluator independence, complete lifecycle cost, failure and residual retention, checkpoint compatibility, and causal ablation.	Design rationale	argument	learning_compute_topology, corben_chatgpt_kiss_irreducible_intelligence_2026, corben_chatgpt_onecell_theseus_2026, deterministic_capability_compilation, ext_attention_is_all_you_need_2017, ext_mamba2_ssd_2024, ext_s4_2022, ext_mamba3_2026, ext_gated_deltanet2_2026, ext_hyperscale_lottery_2026, ext_gated_deltanet_2024, ext_jamba_2024, ext_inkling_2026, ext_neural_message_passing_2017, ext_hyena_hierarchy_2023, ext_rwkv_2023, ext_xlstm_2024, ext_ttt_layers_2024, ext_titans_2025, ext_kan_2024, ext_kan_or_mlp_fairer_comparison_2024, ext_neural_turing_machines_2014, ext_differentiable_neural_computer_2016, ext_liquid_time_constant_networks_2021, ext_tiny_recursive_model_2025, ext_trm_arc_agi_analysis_2025, ext_tiny_autoregressive_recursive_models_2026, ext_unimatrix_2026, ext_memory_caching_2026, scf, rmi, benchmaxxing, ext_mamba_2023, ext_retnet_2023, ext_universal_transformer_2019, ext_dynamic_compute_recurrent_transformers_2026, ext_dreamcoder_2020, kernel_english_residual_compiler, relational_dimension_compiler, ext_llm_in_flash_2024, ext_powerinfer_2024, ext_atsinfer_2026, ext_test_time_training_2020, ext_llada_2025, ext_scaling_dllms_2026, ext_neuromorphic_computing_scale_2025, ext_photonic_neuromorphic_2024, ext_quantum_ml_shadows_2024, ext_organoid_intelligence_2023, ext_kimi_k3_2026, ext_eggroll_hyperscale_es_2026, ext_mezo_2023, ext_forward_forward_2022	Source notes available for all 53 assigned sources; 42 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 53 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_eggroll_hyperscale_es_2026: Shows that the substrate ABI must include the learning rule and its population, perturbation, RNG, fitness, aggregate-update, and total-cost state, including recurrent and integer-valued cases. Limits: Source-reported architecture and learning results are coupled and unreproduced; neither ES nor a recurrent/int8 substrate is established as superior.; ext_mezo_2023: Provides a forward-only zeroth-order update comparator for a differentiable language model under backward-memory pressure. Limits: Reported memory and GPU-hour results are setting-bound and do not establish a preferred substrate or learning rule.; ext_forward_forward_2022: Provides a preliminary local forward-only credit-assignment comparator to a universal reverse-mode ABI. Limits: Small-scale preliminary investigations do not establish foundation-model parity, efficiency, or biological plausibility.; ext_test_time_training_2020: Provides an input-conditioned adaptation comparator in which a model updates on a self-supervised objective at inference time. Limits: Test-time training is not automatically stable, reversible, distribution-safe, or compatible with shared-state serving; no local model was adapted.; ext_llada_2025: Supplies a modern diffusion-language-model comparator for non-autoregressive token denoising and alternative generation schedules. Limits: Source-reported quality or speed does not establish parity, lower lifecycle cost, safer decoding, or ABI compatibility.; ext_scaling_dllms_2026: Provides scaling evidence and design considerations for diffusion language models as a distinct generation substrate. Limits: Reported scaling behavior does not establish universal superiority, deployment efficiency, or transfer to ASI Stack workloads.; corben_chatgpt_kiss_irreducible_intelligence_2026: Supplies the search-verify-compile ratchet, exact-latent split, frozen-core question, and total-system KISS accounting as author intent. Limits: A design conversation is not independent evidence; all proposed mechanisms remain hypotheses.; corben_chatgpt_onecell_theseus_2026: Supplies OneCell, typed state lanes, the Cognitive Kernel ABI, inner recurrence plus outer exact search, and a governed Theseus tournament as author intent. Limits: OneCell is neither implemented nor selected; the conversation cannot support performance or generality claims.; scf: Supplies stable capability identity, replaceable implementation, qualification, lifecycle, fallback, and rollback contracts above a changing cognitive kernel. Limits: SCF is a local design source, not evidence that heterogeneous learned substrates are interchangeable or safe.; rmi: Supplies modular routing, specialist lifecycle, residual retention, and bounded authority for substrate-neutral route selection. Limits: No learned architecture tournament or cognitive-kernel ABI is implemented by this source.; benchmaxxing: Supplies benchmark ratchets, held-outs, anti-Goodhart controls, wall diagnosis, regression retention, and architecture-change criteria. Limits: It does not show that any candidate architecture wins a matched comparison.; ext_attention_is_all_you_need_2017: Defines the dense self-attention Transformer baseline and its training-parallel sequence-processing design. Limits: The paper is a baseline, not proof of universal optimality or a local reproduction.; ext_mamba_2023: Introduces selective state spaces and hardware-aware recurrent computation as a non-attention sequence substrate. Limits: Reported scaling and benchmark results are not reproduced and do not establish governed replaceability.; ext_mamba2_ssd_2024: Connects structured state spaces and attention-like formulations through state-space duality and efficient algorithms. Limits: Formal duality and source benchmarks do not erase state, kernel, or hardware differences.; ext_s4_2022: Provides the foundational structured-state-space sequence model and long-range benchmark lineage preceding selective SSMs. Limits: Source-reported long-range and generation results do not establish exact retrieval, general cognition, or governed replacement.; ext_mamba3_2026: Adds a current selective-SSM comparator with revised state update, discretization, and MIMO design. Limits: Its recency and source-reported results do not make it the preferred substrate.; ext_gated_deltanet2_2026: Supplies the dated reported recurrent/linear-attention frontier after separating erase and write gates and compares against Mamba-3 variants on a 1.3B/100B-token envelope. Limits: The author-reported result displaces Mamba-3 only inside the paper’s exact comparison; no checkpoint, training run, hardware result, retrieval result, transfer result, or local superiority is reproduced here.; ext_hyperscale_lottery_2026: Supplies adversarial edge-hardware measurements showing that an architecture optimized for hyperscale GPU utilization can move cost onto smaller deployment targets. Limits: One preprint and its selected devices do not establish a universal edge penalty or outweigh model quality; the result sets a platform-stratified measurement requirement.; ext_gated_deltanet_2024: Adds targeted delta-rule memory updates, gating, and hybrid linear-attention/state-space comparisons. Limits: Source-reported retrieval and efficiency gains do not establish exact memory, governed state, or local superiority.; ext_jamba_2024: Provides a large-scale Transformer-Mamba-MoE hybrid comparator and evidence that architecture composition itself is a design variable. Limits: Its source-reported results do not establish that the mixture is generally optimal or that ABI routing and lifecycle costs are negligible.; ext_inkling_2026: Provides a current released hybrid-topology case: sparse MoE, a five-local-to-one-global attention schedule, relative positional features, short convolutions, multimodal encoders, controllable effort, and explicit active/total/hardware disclosures inside one Transformer family. Limits: Provider-reported architecture and evaluations are not locally reproduced; no published ablation isolates sliding attention, MoE, relative position, convolution, scale, data, or RL, and open weights do not erase the reported 600 GB to 2 TB VRAM access boundary.; ext_neural_message_passing_2017: Provides a learned message-passing framework over graph-structured inputs as a non-token-native relational substrate comparator. Limits: Molecular prediction results do not establish general cognition, exact graph memory, dynamic routing, or local reproduction.; ext_retnet_2023: Provides parallel, recurrent, and chunkwise recurrent computation modes at the attention-recurrence boundary. Limits: Reported efficiency and quality are not reproduced and do not prove exact state tracking.; ext_rwkv_2023: Provides a recurrent language-model baseline with parallelizable training and recurrent inference. Limits: Reported benchmark and memory properties remain source results.; ext_hyena_hierarchy_2023: Provides a long-convolution alternative and hardware-aware subquadratic sequence-mixing comparison. Limits: No local throughput, quality, recall, or scaling claim follows.; ext_xlstm_2024: Provides a modern gated recurrent baseline for memory and scalable sequence modeling. Limits: No local training, depth-extrapolation, quality, or efficiency result exists.; ext_universal_transformer_2019: Provides shared-weight depth recurrence and adaptive computation as a bridge between fixed-depth Transformers and iterative kernels. Limits: Expressivity and benchmark reports do not establish stable indefinite recurrence or useful stopping.; ext_dynamic_compute_recurrent_transformers_2026: Provides current evidence and evaluation pressure for learned dynamic compute allocation in recurrent Transformers. Limits: Source results do not validate the chapter’s router, recurrence, or stopping rules.; ext_ttt_layers_2024: Treats hidden state as a learned model updated at test time, expanding the state and rollback surface. Limits: No local result establishes useful online adaptation, contamination resistance, or recoverability.; ext_titans_2025: Provides a neural long-term-memory and test-time memorization comparator. Limits: Mutable neural memory is not exact durable state; source results are not reproduced.; ext_kan_2024: Defines KANs through learned univariate edge functions and supplies scientific/symbolic demonstrations. Limits: It does not establish a general replacement for MLPs, attention, or language-model backbones.; ext_kan_or_mlp_fairer_comparison_2024: Supplies a critical matched-accounting objection for KAN-versus-MLP claims. Limits: One comparison does not settle every task or implementation; it sets a fairness requirement.; ext_neural_turing_machines_2014: Provides a differentiable controller over external memory and algorithmic-task tests. Limits: Toy-distribution performance does not establish reliable exact memory or general computation.; ext_differentiable_neural_computer_2016: Extends learned control over dynamic external memory and structured tasks. Limits: Source-reported tasks do not establish scalable, exact, or rollback-safe memory.; ext_liquid_time_constant_networks_2021: Provides a continuous-time recurrent comparator with input-dependent time constants. Limits: Reported time-series results do not establish broad cognitive or hardware superiority.; ext_dreamcoder_2020: Provides library learning and reusable program abstraction as a comparator for verified compression and future-search reduction. Limits: DreamCoder’s reported domains do not validate OneCell, the frozen-core ratchet, or local library growth.; ext_tiny_recursive_model_2025: Provides the primary tiny weight-tied recursive reasoning candidate and its reported narrow puzzle results. Limits: The source does not establish general reasoning, language capability, deep effective recursion, or total-system simplicity.; ext_trm_arc_agi_analysis_2025: Provides a critical audit of TRM task-identity dependence, 1000-sample voting, and effective recursion depth. Limits: The technical note is not a local reproduction and does not by itself settle every TRM variant.; ext_tiny_autoregressive_recursive_models_2026: Provides a compute-matched mechanism ablation that reports no reliable advantage from the full autoregressive TRM mechanism. Limits: Small character-level algorithmic tasks do not settle all recurrent architectures; the result sets a controlled-comparison requirement.; ext_unimatrix_2026: Provides a current negative result for compressed recurrent associative recall and a positive source-reported contrast after adding sparse slots and pointer-level routing. Limits: One small-scale preprint does not establish universal memory limits or a locally reproduced architecture result.; ext_memory_caching_2026: Provides a growing-memory recurrent comparator that interpolates between fixed recurrent state and cached addressable state. Limits: The source reports that Transformers remain strongest on its in-context recall tasks; no local result or universal complexity conclusion follows.; relational_dimension_compiler: Separates semantic arity, primitive computational arity, and storage arity; uses typed relation-node reification for persistent arbitrary finite relations while treating pairwise, polyadic, graph, hypergraph, topological, field, symbolic, and tensor systems as replaceable computational lowerings. Limits: Finite reification establishes representational sufficiency only; bounded primitive arity, efficient discovery, hardware advantage, natural transfer, and substrate superiority remain untested.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	A bounded heterogeneous-kernel substitution and mixed-kernel route reproduce and transfer without authority widening, exact-state loss, hidden assistance, unsafe release, or dominated lifecycle cost; otherwise narrow or refute the relevant atoms.
relational-dimension-compilation-and-polyadic-cognition.core	relational-dimension-compilation-and-polyadic-cognition	Polyadic cognition should be implemented as a slow-path relational-dimension compiler over stable lower-arity primitives: candidate higher-order structure is typed, role-addressable, denominator-complete, qualified against strong pairwise and sequence baselines, budgeted, reversible, and retained only when it improves held-out relational performance without violating memory, latency, calibration, or governance constraints.	Design rationale	argument	relational_dimension_compiler, ext_neural_message_passing_2017	Source notes available for all 2 assigned sources; 2 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 2 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	relational_dimension_compiler: Supplies the Corben-authored conceptual architecture for typed relational IR, adaptive order, role-sensitive reification, qualification, compilation, contraction, and the RODIE research program. Limits: Author-lineage conceptual source only; no implementation, natural-task result, irreducibility theorem, bounded-arity proof, efficiency, safety, reproduction, or transfer result.; ext_neural_message_passing_2017: Supplies a primary external graph-message-passing comparator with explicit node, edge, aggregation, update, and readout structure. Limits: Molecular-property results do not establish general relational reasoning, dynamic topology discovery, exact role persistence, higher-order necessity, or local reproduction.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	RDC improves natural and synthetic role-sensitive tasks over competent lower-order rescues under matched compute and memory, preserves calibration and lifecycle integrity, and reproduces independently; no irreducibility, universal arity, general-intelligence, or SOTA claim.
governed-model-training-distributed-optimization-and-scaling.core	governed-model-training-distributed-optimization-and-scaling	A model-training candidate is eligible for qualification only when a prospectively frozen run contract binds architecture, data lease and order, objective, optimizer, scheduler, numerical policy, device and parallelism topology, code and environment, budget, stopping and fault policy, complete attempted-run denominator, full declared checkpoint state, commit consistency, resume equivalence class, candidate-checkpoint family, validation-only selection, independent unopened qualification, and residual ownership; a loss reduction, completed job, high utilization, checkpoint file, successful load, recovered run, selected candidate, formal record proof, or source-reported scale result alone establishes neither faithful training, model quality, optimizer superiority, fault tolerance, safety, support, readiness, release, transfer, nor SOTA.	Design rationale	argument	learning_compute_topology, beastbrain, ext_llama3_herd_2024, ext_megatron_distributed_training_2021, ext_zero_optimizer_2019, ext_gspmd_2021, ext_datastates_llm_2024, ext_pytorch_distributed_checkpoint_2026, ext_mlperf_training_v6_2026, ext_adam_2015, ext_amsgrad_2018, ext_adamw_2019, ext_adafactor_2018, ext_lamb_2019, ext_shampoo_2018, ext_kfac_2015, ext_lion_2023, ext_sophia_2023, ext_soap_2024, ext_schedule_free_2024, ext_mup_2022, ext_modular_norm_2024, ext_muon_scalable_2025, ext_muon_spectral_norm_2026, corbens_trainer_project, ext_proof_of_learning_2021, ext_test_time_training_2020, ext_curriculum_learning_2009, ext_kimi_k3_2026, ext_eggroll_hyperscale_es_2026, ext_openai_es_2017, ext_mezo_2023, ext_forward_forward_2022	Source notes available for all 33 assigned sources; 31 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 33 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_eggroll_hyperscale_es_2026: Supplies the primary systems case for low-rank batched evolution strategies, counter-based perturbation reconstruction, discrete objectives, and source-reported recurrent/int8/reasoning studies. Limits: Kernel throughput, quality, theory, and total-cost results are configuration-bound and unreproduced; the largest reported int8 population has a severe GPU-hour denominator.; ext_openai_es_2017: Provides the foundational modern distributed ES comparator using parameter perturbations, scalar fitness, and shared-seed reconstruction. Limits: The studied control results do not establish foundation-model sample efficiency or total-compute advantage.; ext_mezo_2023: Provides a seed-reconstructed forward-only zeroth-order fine-tuning comparator with inference-like memory use. Limits: Source-reported savings are task- and baseline-bound; query cost and estimator variance remain.; ext_forward_forward_2022: Provides a preliminary positive/negative forward-pass and local-objective alternative to global reverse-mode credit. Limits: Does not establish large-scale performance, generalization, efficiency, or biological plausibility.; beastbrain: Passage-reviewed version-family lineage for separating idle scheduling from update authority: later Nocturne drafts propose replaying failed interactions, searching alternative strategies, constructing preference data, optimizing traces, caching repeated work, and mutating memory or policy state during low-contention windows. The book retains the temporal scheduling idea while splitting compaction, replay, search, dataset construction, and model mutation into separately governed operations. Limits: No Nocturne implementation, eligible dataset, DPO run, replay-quality result, model update, rollback, independent qualification, useful improvement, safety result, or transfer exists. Idle time and failed logs confer neither data rights nor update authority; model-generated successful trajectories are not automatically valid preferences.; ext_proof_of_learning_2021: Provides a transcript-based comparator for auditing whether a claimed training process follows a committed sequence of states. Limits: It does not prove dataset legitimacy, objective adequacy, hidden-compute absence, exact reproduction, model quality, or safety.; ext_test_time_training_2020: Provides a distinct training-timescale comparator in which per-input self-supervised updates occur at test time. Limits: The method does not establish safe online learning, distributed-state consistency, rollback, or general robustness.; ext_curriculum_learning_2009: Provides the foundational curriculum-learning hypothesis that example order and difficulty schedules can change optimization outcomes. Limits: Curriculum effects are task-, model-, metric-, and schedule-dependent; easier-first ordering is not a universal improvement rule.; ext_llama3_herd_2024: Supplies a paper-body-reviewed large-run case covering 4D parallelism, network-aware topology, numerical stability interventions, checkpoint infrastructure, interruption denominators, and effective training time. Limits: Provider-reported scale, utilization, interruption, and recovery results are not locally reproduced and do not establish exact resume.; ext_megatron_distributed_training_2021: Grounds composed tensor, pipeline, and data parallelism, strict optimizer semantics, microbatch scheduling, communication, and topology tradeoffs. Limits: Its measured cluster and model envelope is not universal, and no throughput result is reproduced.; ext_zero_optimizer_2019: Grounds progressive sharding and separate accounting for optimizer, gradient, parameter, activation, and residual state. Limits: Sharding does not prove complete application state, checkpoint integrity, or universal efficiency.; ext_gspmd_2021: Grounds compiler-mediated SPMD sharding and mixed parallelism as a serious alternative to manually composed plans. Limits: Automatic completion does not prove semantic equivalence, optimal topology, or transfer.; ext_datastates_llm_2024: Grounds asynchronous distributed checkpoint creation, logical consistency, storage-tier movement, and checkpoint overhead. Limits: Checkpoint consistency over model and optimizer shards does not by itself establish full-state or trajectory-equivalent resume.; ext_pytorch_distributed_checkpoint_2026: Grounds current save/load, asynchronous future, canonical state, strict load, and resharding interface requirements. Limits: Official API documentation is not scientific performance evidence or proof that every application state is captured.; ext_mlperf_training_v6_2026: Grounds fixed quality targets, repeated time-to-quality, system metadata, variance, divisions, and correction of published training results. Limits: No local MLPerf submission exists, and benchmark conformance would not establish safety, full run integrity, or release merit.; ext_adam_2015: Grounds bias-corrected first- and second-moment adaptive updates as the reference adaptive family. Limits: Its analysis and experiments do not establish universal nonconvex convergence or superiority.; ext_amsgrad_2018: Supplies an Adam non-convergence construction and the AMSGrad maximum-second-moment variant. Limits: A constructed failure does not imply all practical Adam runs fail or AMSGrad always wins.; ext_adamw_2019: Grounds decoupled weight decay and the modern AdamW reference baseline. Limits: Reported tuning/generalization results are setting-bound and the method name does not specify a full policy.; ext_adafactor_2018: Grounds factored second moments, update clipping, and reduced optimizer state for matrix parameters. Limits: Factorization is approximate and source parity does not transfer across tasks or tensor shapes.; ext_lamb_2019: Grounds layer-wise trust ratios and large-batch optimization as a separate scaling family. Limits: BERT time-to-target is hardware, batch, target, implementation, and tuning bound.; ext_shampoo_2018: Grounds tensor-structured per-dimension preconditioning rather than flattened coordinate adaptation. Limits: Stochastic-convex theory and source experiments do not settle current distributed lifecycle cost.; ext_kfac_2015: Grounds Kronecker-factored Fisher approximations as a practical curvature-aware family. Limits: Approximation, damping, inversion, architecture, and distributed cost determine applicability.; ext_lion_2023: Grounds symbolic optimizer discovery and Lion’s one-state sign-momentum update. Limits: Search transfer and reported wins are task/budget bound; the source reports weak settings too.; ext_sophia_2023: Grounds periodic diagonal-curvature estimation and clipped second-order updates for language-model training. Limits: Reported speedups and simplified theory require matched reproduction before generalization.; ext_soap_2024: Grounds Adam-like moments in Shampoo’s evolving preconditioner eigenbasis. Limits: Large-batch results remain tied to model scale, basis refresh, overhead, and tuning.; ext_schedule_free_2024: Grounds the scheduling/iterate-averaging connection and schedule-free optimizer state. Limits: It removes dependence on a stopping-time schedule, not all hyperparameters or evaluation choices.; ext_mup_2022: Grounds maximal-update parametrization and widthwise hyperparameter transfer. Limits: Transfer depends on the prescribed parameterization and does not establish arbitrary substrate transfer.; ext_modular_norm_2024: Grounds architecture-recursive update geometry and reported learning-rate transfer across width/depth. Limits: Well-behaved-module assumptions and studied architectures bound the result.; ext_muon_scalable_2025: Grounds large-scale Muon, weight decay, per-parameter update scaling, and distributed orthogonalization. Limits: Reported compute efficiency and Moonlight results are source-scoped and unreproduced.; ext_muon_spectral_norm_2026: Supplies an accepted theoretical interpretation of Muon through Lion-K, nuclear norm, and implicit spectral constraints. Limits: The idealized interpretation does not establish practical approximation fidelity, quality, safety, or superiority.; corbens_trainer_project: Supplies a pinned local implementation and failure comparator for typed campaigns, training-truth gates, content-addressed checkpoint lineage, complete run facts, retained quarantines, and acknowledged-checkpoint requirements. Limits: Its external dependencies and natural training were not reproduced; retained null seed/code identities, later-quarantined claims, and unacknowledged asynchronous checkpoints are negative implementation evidence, not model-quality or distributed-training results.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Across multiple seeds and representative model scales, a governed run preserves declared semantics and state under fault and resume, selects without qualification leakage, beats strong distributed-training and checkpoint baselines on joint time-to-quality, recovery correctness, useful throughput, and governance cost, and reproduces independently without granting release authority.
learning-compute-topology-and-adaptive-process-architecture.core	learning-compute-topology-and-adaptive-process-architecture	For an exact task family, adaptive-state boundary, resolution contract, evidence and evaluator policy, credit semantics, lifecycle, integration operators, compute substrate, resource budget, authority, observables, rollback, and time, a self-improving stack should represent the learning process as a typed, versioned, provenance-bearing, rewritable causal topology; compile it through an explicit semantic firewall into execution and physical compute; measure discovery, evaluation, integration, communication, retention, and realization leakage jointly; and admit topology changes only through matched experiments and reversible governance. A branch count, worker count, schedule, normalized graph, bounded theorem, passing reference implementation, toy phase diagram, or source-authored architecture alone establishes neither adaptive plurality, retained learning, safety, superiority, transfer, nor ASI.	Design rationale	argument	learning_compute_topology, ext_megatron_distributed_training_2021, ext_gspmd_2021, ext_pbt_2017, ext_openai_es_2017, assurance_shift_learning	Source notes available for all 6 assigned sources; 6 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 6 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	learning_compute_topology: Supplies the four-topology distinction; adaptive identity and typed semantic channels; LCT-IR and LCNF; seven bounded propositions; semantic compilation and realization leakage; topology measurements; derived process organizations; ABVI; a bounded executable supplement; and an explicit experimental, falsification, and safety program. Limits: Corben-authored research source. Its propositions apply only under stated models; its implementation covers bounded IR behavior; its coverage and novelty matrices are scoped manual studies; its phase diagrams are toy or analytical; and no real neural-training benefit, topology-control result, independent reproduction, transfer, safety, SOTA, AGI, or ASI result exists.; ext_megatron_distributed_training_2021: Provides a primary distributed-training comparator for tensor, pipeline, and data parallel execution at scale. Limits: Distributed execution structure does not by itself define multiple persistent adaptive identities, evaluator topology, credit semantics, or learning-process topology.; ext_gspmd_2021: Provides a primary compiler and partitioning comparator for mapping tensor computations onto device meshes. Limits: A distributed execution compiler is not a semantic learning-process compiler and does not authorize changes to identity, evidence, evaluation, credit, or integration semantics.; ext_pbt_2017: Provides a primary population-based-training comparator in which persistent candidates, evaluation, selection, copying, and hyperparameter perturbation alter the adaptive process. Limits: One population algorithm does not validate LCT’s general primitive set, normal form, metrics, compiler, or derived architectures.; ext_openai_es_2017: Provides a primary evolutionary-strategy comparator for proposal parallelism, scalar evaluation, and centralized update without mistaking worker multiplicity for persistent adaptive plurality. Limits: The reported ES system does not establish LCT metrics, ABVI, information bounds, or superior process topology.; assurance_shift_learning: Adds a competence-dependent regime allocator: once positive support is broad and stable inside a qualified region, marginal learning effort may shift toward evaluation, boundary discovery, repair, assurance, and recovery. Limits: Assurance-dominance is a falsifiable local allocation hypothesis, not a measured crossover, universal learning law, or resource advantage.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Independent encoders agree within a frozen equivalence tolerance; the semantic compiler predicts protected trace and cost differences; topology metrics add held-out predictive value beyond model, task, optimizer, and hardware variables; and a governed adaptive topology improves retained useful learning over competent matched fixed baselines without exceeding preregistered unsafe-release, rollback, latency, storage, and governance-cost ceilings across reproduction and transfer.
learning-theory-generalization-and-scaling-science.core	learning-theory-generalization-and-scaling-science	A generalization, transfer, emergence, or scaling assertion should be accepted only through a dated claim contract that binds population and sampling assumptions, data support, hypothesis and algorithm, optimization and inductive bias, complexity or explanatory lens, metric, compute regime, uncertainty, breakpoint tests, held-out prediction, alternatives, and transfer boundary; a bound, fit, interpolation result, compression ratio, benchmark jump, or larger model alone establishes neither broad generalization, capability emergence, safety, nor future scale behavior.	Design rationale	argument	simulation_scaling, ext_scaling_laws_neural_language_models_2020, ext_mdl_tutorial_2004, ext_weakness_generalization_2023, ext_weak_to_strong_generalization_2023, ext_information_bottleneck_2000, ext_valiant_theory_learnable_1984, ext_deep_double_descent_2020, ext_emergent_abilities_llms_2022, ext_emergent_abilities_mirage_2023, ext_no_free_lunch_inductive_bias_2024, ext_eggroll_hyperscale_es_2026, ext_forward_forward_2022, forward_transfer_program_synthesis	Source notes available for all 14 assigned sources; 12 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_eggroll_hyperscale_es_2026: Supplies a bounded consistency analysis for low-rank population updates approaching Gaussian ES under declared high-dimensional assumptions. Limits: The result is not global convergence, backpropagation superiority, nonlinear-regime transport, or a local reproduction.; ext_forward_forward_2022: Supplies a distinct local-credit learning architecture that changes the scope of learning and generalization claims. Limits: Preliminary small-scale feasibility is not a foundation-model generalization theory.; ext_scaling_laws_neural_language_models_2020: Empirical study reporting power-law relationships between cross-entropy loss, model size, data, and compute in its model family. These fitted relations are source-reported, metric- and regime-bound, and do not automatically forecast downstream capabilities, safety, or other architectures. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_mdl_tutorial_2004: External description-length source for model/data tradeoffs, compression as inductive discipline, and residual/error-accounting vocabulary. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_weakness_generalization_2023: Conditional formal comparator separating extension-based hypothesis weakness from description length, plus source-reported toy binary-arithmetic trials comparing the two proxies. Limits: The theorem assumes the paper’s finite enactive formalism and a uniform distribution over its task space; the experiments are toy 8-bit addition and multiplication. No neural-network, natural-task, distribution-transfer, local reproduction, support, safety, or ASI result follows.; ext_weak_to_strong_generalization_2023: Primary weak-to-strong-supervision comparator for a capability-gap envelope, held-out outcome audit, ceiling comparison, and explicit disanalogies between current weak-model studies and superhuman oversight; it does not establish local supervision quality, reliable elicitation, alignment, safety, or an ASI Stack result. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_information_bottleneck_2000: External representation-compression source for relevance-preserving compression, bottleneck variables, mutual-information tradeoffs, and compression/utility separation. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_valiant_theory_learnable_1984: Foundational PAC-learning source for explicit accuracy, confidence, sample, concept-class, and computational assumptions. Limits: Does not directly explain or certify modern foundation-model generalization, transfer, or safety.; ext_deep_double_descent_2020: Reports model-wise, sample-wise, and epoch-wise double-descent phenomena and motivates effective-complexity tracking. Limits: Configuration-bound empirical result, not a universal law that more model, data, or training helps or hurts.; ext_emergent_abilities_llms_2022: Frames reported task abilities that appear sharply with model scale and motivates prospective breakpoint analysis. Limits: Survey/framing source, not proof of mechanistic discontinuity or universal unpredictability.; ext_emergent_abilities_mirage_2023: Provides counterevidence that nonlinear metrics and limited test data can create apparently sharp emergence from smoother underlying changes. Limits: Does not establish that every capability transition is a measurement artifact.; forward_transfer_program_synthesis: Makes reusable-knowledge transfer explicitly relative to a task lineage, task distribution, search procedure, verifier, and resource budget, and separates within-family positive transfer from cross-family transfer. Limits: No universal generalization, scaling law, distribution-independent benefit, or empirical transfer result is established.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Frozen curve families and explanatory lenses predict held-out scale/task points with calibrated intervals and survive metric-artifact and regime-break tests; no universal scaling law.
readiness-gates-residual-escrow-and-quarantine.core	readiness-gates-residual-escrow-and-quarantine	For an exact versioned target, consumer, use, workload family, authority and rights envelope, and evaluation horizon, readiness should issue an expiring routability lease only from independently owned gate evidence, complete per-check state, preserved regression floors, inherited residual custody, allowed and blocked routes, monitoring, rollback and fallback obligations, and review triggers; failed, stale, waived, quarantined, superseded, retired, or lineage-invalidated targets cannot enter ordinary use, and split, merge, retrain, replace, rollback, and retirement transitions must preserve affected descendants, artifacts, effects, and residual owners.	Design rationale	argument	rmi, rgs, deterministic_capability_compilation, benchmaxxing, scf, octopus_router, cognitive_loop_closure, moecot, project_theseus_whitepaper, theseus_architecture_gate, theseus_self_evolution_system, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, ext_darwin_godel_machine_2025, ext_dont_hallucinate_abstain_2024, ext_swe_rebench_v2_2026, precision_contract, regret_engine, assurance_shift_learning, adjudicated_persistence	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 24 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	rmi: Supports readiness and residual discipline through mastery thresholds, critical-failure vetoes, residual escrow, benchmark diagnosis, regression preservation, specialist lifecycle, add/split/merge/retire decisions, quarantine, and frontier/floor separation. Limits: Conceptual architecture source only; no readiness engine, benchmark run, residual ledger, or prototype inspection exists here.; rgs: Supplies residual escrow as an active lifecycle: periodic reattempt, consistently solved promotion to regression, recurring-cluster promotion to diagnosis/frontier work, defective-item repair or retirement, and a protected but non-dominating evaluation budget. It also supplies subgroup floors and critical-failure vetoes around ordinary mastery graduation. Limits: Conceptual precursor in the same author-side lineage as RMI, not independent confirmation. No residual store, recurrence classifier, reattempt scheduler, budget allocator, readiness engine, benchmark run, or promotion decision has been implemented or measured.; benchmaxxing: Supports gate evidence through benchmark lifecycle states, wall diagnosis, benchmark/model ledgers, residual preservation, anti-Goodhart safeguards, contamination checks, transfer checks, and architecture-change discipline. Limits: No benchmark harness, benchmark mutation, holdout, empirical run, or local model/prototype result was produced here.; scf: Supports governed qualification through stable capability fields, exact implementation identity, append-only evidence registry, scoped/defeasible/expiring qualification, route validation, lifecycle events, incidents, appeals, rollback, and authority ceilings. Limits: Does not prove production safety, global alignment, evaluator quality, or current reversibility beyond the finite proofs and schemas separately recorded in this repo.; octopus_router: Supports readiness-aware routing through arm cards, local benchmarks, residuals, permission envelopes, domain quarantine, split/merge/retire policies, routing metrics, and arm lifecycle governance. Limits: No routed-specialist prototype, routing benchmark, quarantine harness, or learned-router evaluation has been run.; cognitive_loop_closure: Supports lifecycle and promotion boundaries for procedural tools through verification grades, risk/runtime tiers, monitoring plans, stale-tool retirement, revision discipline, and warnings against premature closure or unsafe automation. Limits: No local loop-detection, tool-synthesis, monitoring, or retirement automation was executed.; moecot: Supports readiness gates, benchmark artifacts, promotion blockers, residual tracking, replay, handoff, fail-closed control-plane ledgers, and explicit current limitations as runtime-reference context. Limits: Runtime readiness artifacts, code, logs, benchmark records, and replay records have not been imported, inspected, or reproduced here.; project_theseus_whitepaper: Supports residual escrow, benchmark floors, frontier/diagnostic/regression/public-calibration/retired benchmark states, specialist lifecycle, safety-visible gates, and report-first implementation discipline. Limits: Source-reported implementation state only; current machine-readable reports and benchmark ledgers were not rerun or verified from this repo.; theseus_architecture_gate: Supports readiness gates as pre-training promotion controls covering ratchet completeness, router readiness, safety ledger, regression suite, public calibration, residual escrow, bridge benchmarks, procedural tools, routing memory, lifecycle governance, and external-inference zero. Limits: The reported green gate snapshot was not independently verified by rerunning Theseus commands or inspecting current report JSON from this repo.; theseus_self_evolution_system: Supports evidence-gated lifecycle governance through intervention ladders, ATTD repo-health gates, guarded teacher edits, architecture experiment governance, outcome ledgers, lifecycle-governed cells, and loop closure into local tools. Limits: No Theseus scripts, ATTD reports, branch flow, teacher runner, outcome ledger, dashboard endpoint, or autonomous-improvement result was executed here.; cca_project: Supports readiness as an evidence-gated lifecycle through promotion states, quarantine, rollback, contamination controls, canonical closure authority, and fail-closed benchmark truth contracts. Limits: Pinned-project implementation context only; project training, benchmark, quarantine, and promotion artifacts were not replayed from this book.; moecot_manifest_project: Supports explicit promotion states, immutable evidence identities, residuals, replay, holdout separation, fail-closed gates, and maintenance/growth stops when architecture debt remains red. Limits: Pinned-project architecture context only; promotion, holdout, deployment, and readiness behavior was not rerun here.; beastbrain_project: Provides a readiness-overclaim negative record: extensive interfaces, simulations, feature counts, and checklists coexisted with open compiler, integration, hardware, security, and benchmark gates. Limits: The pinned snapshot does not establish production readiness, capability, secure autonomy, or live quarantine behavior.; bugbrain_project: Provides the exact skipped-as-green negative case and motivates per-check applicability, requiredness, attempt, result, waiver, evidence identity, and policy-version fields. Limits: Historical reports and current code differ; the book does not reproduce BugBrain readiness, hardware, bridge, training, or deployment behavior.; corbens_trainer_project: Supports transitive quarantine and promotion-lease invalidation through retained cases where promoted claim records survived later quarantine of every contributing run. Limits: The retained snapshot is local-project evidence only; external dependencies, training, benchmarks, and end-to-end capability were not reproduced.; corbens_best_model_possible_project: Supports quarantine, rollback, fallback-route, and material-postcondition boundaries while warning that named transitions, defaults, and interface presence may lack durable runtime effect. Limits: The project does not establish target readiness, semantic capability, durable transitions, autonomous routing, or promotion quality.; ext_darwin_godel_machine_2025: Provides an empirical comparator for archive-based self-modifying coding-agent candidates, compile/edit eligibility, benchmark selection, branching lineage, retained stepping stones, sandbox and resource limits, and rollback-oriented auditability. Limits: The source-reported coding-benchmark gains and archive benefits do not establish monotonic general improvement, evaluator independence, preserved authority, readiness for ordinary use, safe self-modification, local reproduction, or permission for a generated child to promote itself.; ext_dont_hallucinate_abstain_2024: Supports measuring abstention with coverage, accuracy, calibration, and useful response behavior while treating self-reflection and model agreement as fallible evidence. Limits: The ACL models, prompts, domains, collaboration schemes, and reported gains were not reproduced; the local single-model router is not an independent multi-model panel.; ext_swe_rebench_v2_2026: Adds concrete instrument-failure gates for missing images, broken setup, parser drift, implicit naming, external context, test coupling, emulation cost, and failed positive controls. Limits: A failed instrument closes the denominator and cannot be laundered into negative evidence about a model, mechanism, or governance architecture.; assurance_shift_learning: Adds the Qualified Competence Envelope as a scoped readiness input and reinforces that learning eligibility, local competence evidence, and deployment readiness are separate decisions. Limits: The envelope is evaluator- and distribution-relative; it is neither a complete competence map nor calibrated deployment permission.; adjudicated_persistence: Adds realization-relative qualification leases, commitment-aware evidence obligations, material-change invalidation, and residual closure before broader admission. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.	effect-complete rollback beyond the local inventory; independently assessed evaluator validity; independently assessed verifier quality; multiple models, seeds, and natural workloads; nonzero substantive answer utility; production routing safety, latency, cost, and transfer; production transfer and governance-cost evidence; useful governed throughput at the registered frontier	Promotion blocked on expired evidence or uncustodied residuals.
personal-compute-hives-and-federated-edge-intelligence.core	personal-compute-hives-and-federated-edge-intelligence	For an exact versioned principal, household or project, job, use, data and tool class, effect envelope, acceptance test, risk budget, deadline, and evaluation horizon, a Personal Compute Hive should admit and place work only through policy-before-optimization: independently attested participants and roles, intersected authority and rights, task-local context and execution leases, least-authority adequate node selection, scoped approval, monitored sandboxed execution, complete artifact/effect/resource receipts, partition-aware denial or quarantine, and effect-complete rollback or residual custody; reachability, ownership, cheap capacity, a passing record schema, or stale authority alone cannot license execution, and federation, dropout, revocation, replacement, requeue, and retirement must preserve affected descendants and residual owners.	Design rationale	argument	beastbrain, field_of_god_ai_constitution, scf, talos, vcm_public, planforge, octopus_router, rmi, tokenmana, project_theseus_whitepaper, theseus_operator_os, ladon_manhattan, ext_tailscale_docs_2025, ext_kubernetes_overview_docs, ext_k3s_docs_2026, ext_nomad_docs, ext_ray_core_docs_2026, ext_boinc_home_2026, ext_syncthing_home, ext_ipfs_docs, ext_akash_docs_2026, ext_golem_docs_2025, ext_github_self_hosted_runners_docs, ext_cap_theorem_gilbert_lynch_2002, ext_airllm_2023, ext_deepspeed_inference_2022, ext_flexgen_2023, ext_hf_accelerate_big_model_inference_2026, ext_llama_cpp_memory_mapping_2026, ext_llm_in_flash_2024, ext_powerinfer_2024, ext_atsinfer_2026	Source notes available for all 32 assigned sources; 24 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 32 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	beastbrain: Passage-reviewed Mimic lineage for adapting execution to the actual host: hardware and memory-topology discovery, device-specific runtime profiles, thermal and power pressure, memory-tier policy, background-work shedding, and conservative fallback. The book turns this into a versioned HardwareProfileDecision with staged measurement, explicit rejected profiles, expiry, hysteresis, and state-safe transitions. Limits: The source does not implement or validate hardware probing, direct SSD-to-accelerator paths, Apple Neural Engine access, thermal protection, paging, profile switching, sub-five-second adaptation, device safety, useful performance, or transfer. Low-level probes cannot expand data rights, tool authority, or execution authority.; field_of_god_ai_constitution: Supports the hive’s consent, least-sufficient-power, memory/tool governance, reversibility, auditability, and family/agency caution boundaries. Limits: Does not implement or validate a family-governance policy engine or deployed consent workflow.; scf: Supports capability leases, route validation, qualification, lifecycle state, authority ceilings, and no procedural self-ratification for node capabilities. Limits: Does not qualify any particular personal-device capability or prove federation safety.; talos: Supports typed jobs, contract locks, allowed tools, forbidden tools, audit, replay, delivery evidence, and residual feedback for work run on hive nodes. Limits: Does not prove a distributed hive runner or device scheduler exists.; vcm_public: Supports governed context packets, source binding, taint, revocation, adequacy, materialization boundaries, and separation of data from authority. Limits: Does not prove memory quality, retrieval quality, or end-to-end privacy protection.; planforge: Supports goal decomposition, dependency-aware scheduling, minimum viable intelligence tiers, fallback, and replanning across heterogeneous workers. Limits: Does not prove scheduler optimality or safe execution on personal devices.; octopus_router: Supports routing among bounded specialist capabilities, permission envelopes, dynamic loading, local memory, residuals, and lifecycle-managed capability nodes. Limits: Does not prove modular routing performance or safety in a personal network.; rmi: Supports residual escrow, specialist lifecycle, regression preservation, three execution modes, and modular improvement discipline for hive capabilities. Limits: Does not prove the hive improves capability or preserves regressions in practice.; tokenmana: Supports regenerative capacity and resource-governance framing for compute, money, energy, load, human attention, burst control, and friction. Limits: Does not prove a specific hive scheduler is economically optimal or user-beneficial.; project_theseus_whitepaper: Supports local-first trusted-machine Hive framing, registered task-kind providers, residual ledgers, and report-first operational discipline. Limits: No Theseus Hive reports, node registry, benchmark ledgers, or command outputs were rerun from this repo.; theseus_operator_os: Supports operator channels, durable work board, node registry, background/watch/wake contracts, feedback routing, TTLs, kill switches, and safety-visible surfaces. Limits: Does not prove unattended operation or distributed node control is safe.; ladon_manhattan: Supports blind secret handles, secure entry paths, policy injection boundaries, isolated compartments, and reduced credential exposure. Limits: Does not prove kernel-level isolation, side-channel resistance, or complete AI security.; ext_tailscale_docs_2025: Supports private overlay-network and identity-connectivity vocabulary for joining personal devices across network locations. Limits: Does not grant execution authority, data access, family approval, or scheduler safety.; ext_kubernetes_overview_docs: Supports container-orchestration vocabulary for service discovery, storage orchestration, rollouts, bin packing, and self-healing. Limits: Does not encode personal AI authority, VCM taint, child portals, rented-node safety, or evidence-state governance.; ext_k3s_docs_2026: Supports lightweight edge, homelab, IoT, CI, single-board, air-gapped, and embedded Kubernetes substrate options. Limits: Does not prove K3s is the right substrate for a hive or that lightweight orchestration enforces ASI Stack policy.; ext_nomad_docs: Supports heterogeneous job scheduling across containers, binaries, batch jobs, on-prem infrastructure, and cloud infrastructure. Limits: Does not decide which jobs are lawful, private, age-appropriate, evidence-sufficient, or safe to run on personal devices.; ext_ray_core_docs_2026: Supports distributed task, actor, and object-reference vocabulary for hive workers and project-hive execution. Limits: Does not prove distributed object movement preserves data locality, taint, revocation, or privacy.; ext_boinc_home_2026: Supports volunteer-compute lineage for background jobs and donated machine capacity. Limits: Does not make public project work safe for personal machines without signed contracts, sandboxes, and evidence bundles.; ext_syncthing_home: Supports local-first file synchronization, device identity, encrypted transport, and user-controlled storage-location vocabulary. Limits: Does not solve semantic memory quality, VCM revocation, source adequacy, or per-device data-class policy.; ext_ipfs_docs: Supports content-addressed, peer-to-peer artifact location and decentralized retrieval vocabulary for hive bundles, public caches, and reproducible references. Limits: Does not solve private hive memory, deletion, VCM revocation, privacy, authority control, or safe federation.; ext_akash_docs_2026: Supports decentralized cloud, provider resources, leases, GPUs, SDKs, and rented-compute vocabulary. Limits: Does not prove rented compute is appropriate for private data or that any provider satisfies hive evidence obligations.; ext_golem_docs_2025: Supports decentralized task execution, provider selection, data transfer, result handling, and resource-sharing vocabulary. Limits: Does not prove provider trust, payment fairness, sandbox safety, output validity, or dispute handling.; ext_github_self_hosted_runners_docs: Supports self-hosted CI runners as user-managed physical, virtual, containerized, on-prem, or cloud execution machines. Limits: Does not prove unattended runner isolation, cleanup, credential boundaries, or safe execution for untrusted work.; ext_cap_theorem_gilbert_lynch_2002: Supports the consistency, availability, and partition-pressure boundary used to reason conservatively about stale grants, delayed revocation, fresh authority receipts, no-mutation evidence, and grant/effect races. Limits: Does not implement a distributed authority service or prove consensus, availability, partition tolerance, revocation propagation, runtime enforcement, or hive safety.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Lease fixture proves bounded compute delegation and revocation.
compact-generative-systems-and-residual-honesty.core	compact-generative-systems-and-residual-honesty	For an exact versioned source artifact or state, consumer and use, reconstruction or semantic-adequacy contract, allowed loss, authority and rights envelope, workload distribution, cost boundary, and evaluation horizon, a compact representation should be admitted only when its generator, search, metadata, semantic lease, verifier, repair, fallback, interface, human, governance, recovery, and residual burdens are fully attributed; exactness or scoped loss is independently checked against the consumer contract; source lineage, supersession, and fallback remain executable; and the selected representation improves a preregistered joint utility-and-total-burden frontier over strong matched literal, standard codec, model-compression, retrieval, and semantic baselines. Smaller storage, tokens, parameters, or a finite fixture alone establishes neither useful compression nor semantic adequacy, and any hidden, moved, deferred, or discharged burden must retain state, evidence, owner, due condition, descendants, and reopening triggers.	Design rationale	argument	cgs, deterministic_capability_compilation, rgs, bugbrain, simulation_scaling, rmi, project_theseus_whitepaper, bbvca_v9, bbvca_main, rankfold_neuralfold, treellm, spinoza, verification_bandwidth, cognitive_compilation, circle_ai_architectures, coilra_multicoil_rope, ext_raptor_2024, qcsa_whitepaper, kernel_english_residual_compiler, precision_contract, orcp_moecot	Source notes available for all 21 assigned sources; 17 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 21 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	cgs: Supports the compact-generative core claim through the seed/rule/state/residual/verification/governance tuple; seven admission criteria; description-to-recursive-governance functional ladder; seed, rule, memory, residual, verification, governance, and hidden-complexity design laws; generative-leverage, fidelity, residual, verification, governance, and hidden-debt metric vector; ten-part design template; and active-compression research loop. Folded generate/verify/repair lineage: Supports residual accounting and governance around compressed generation through compact seeds, rule systems, residual/error channels, verification contracts, governance interfaces, and hidden complexity debt. Folded semantic-representation lineage: Supports compact semantic substrates only when compression, generation, residuals, verification, governance, hidden complexity, and permitted use remain explicit. Limits: Conceptual framework and research program only; no local CGS benchmark, utility test, proof of compact adequacy, generator, or implementation has been run here. Folded GVR boundary: Does not establish that any generate-verify-repair codec preserves utility, improves rate, or reconstructs data in this repo. Folded semantic boundary: Does not prove fixed semantic tokens are adequate, grounded, or utility-preserving for any task.; rgs: Supports residual honesty inside growth loops through attempt logging, success/failure classification, verified-tool promotion, residual escrow, regression suites, and benchmark/model/tool/residual ledgers. Limits: Conceptual methodology only; no local benchmark runs, tool-promotion traces, regression data, or implemented ratchet are present.; bugbrain: Supports edge-efficiency and tiny-local-cognition lineage through resource-constrained microkernel design, fractal/spiking adaptive mechanisms, user feedback, storage paging, and build/flash/test workflow pressure. Limits: Speculative prototype-lineage source; this repo has not built, flashed, emulated, benchmarked, or validated BugBrain, and consciousness/AGI claims are not used.; simulation_scaling: Supports making compactness resource-explicit through contract-relative scope, clockspeed, fidelity, efficiency, physical capacity, and bottleneck accounting. Limits: Theoretical scaling framework only; no physical experiment, simulation benchmark, or independent literature audit was run here.; rmi: Supports compactness as part of modular ratcheting through benchmark frontier pressure, routed specialist attempts, residual escrow, loop closure, independent arm/router improvement, regression preservation, and lifecycle discipline. Limits: Architecture proposal only; no independent reproduction, benchmark run, prototype inspection, or empirical modular-intelligence result exists here.; project_theseus_whitepaper: Supports report-first compact implementation framing through SymLiquid, SparkStream, Octopus routing, bounded specialist arms, benchmark/residual ledgers, checkpoints, trusted Hive nodes, observability, and evidence-before-growth discipline. Limits: Source-reported implementation narrative only; current Theseus reports, ledgers, code paths, and command outputs were not rerun from this repo.; bbvca_v9: Folded generate/verify/repair lineage: Supports the GVR claim through the six-field Reconstruction Contract; ontological-versus-codec universality; public-law amortization and apex-only exclusion; reversible-factorization and predictive-restoration transition families; layerwise exactness; bounded local generation, verification, repair, splitting, literal fallback, and interface accounting; complete code length; frozen proxy versus realized rate; proxy bootstrap, smoothing, shrinkage, conservative anti-pruning, committed refresh, and restricted bottom-up adaptive-tree search; systems layout; and friendly/hostile falsification. Limits: Folded GVR boundary: Mature conceptual codec and bounded-search research program only. The paper provides no local Prototype A implementation, bottom mapping, entropy-coder stream, independent decoder, proxy-gap measurement, codec correctness, compression ratio, runtime, memory, utility, transfer, safety, deployment, SOTA, support, AGI, or ASI result. Its principles and bounded-search proposition remain conditional arguments under stated assumptions, not machine-checked corpus-performance proofs.; bbvca_main: Folded generate/verify/repair lineage: Supports the nine-version architectural correction from apex-seeded 3D reconstruction through explicit exact transition modes, complete rate accounting, contract-relative claims, apex exclusion, bounded semantic verification, interface pricing, Generate-Verify-Repair, shared-law amortization, layerwise exactness, restricted adaptive-tree search, two-phase proxy/final rate, and calibrated proxy bootstrap. The lineage additionally supports treating 3D, mapping, overlap, public-law size, integrity metadata, and apex minimization as separately testable choices rather than assumed advantages. Limits: Folded GVR boundary: Same-author v1.1-v9.0 correction lineage only, not nine independent sources; the v9 tab substantively duplicates the separately cached bbvca_v9 paper and counts once. Later corrections control earlier metaphorical or underspecified language. No local codec, emitted bitstream, independent decoder, exact reconstruction, compression ratio, mapping/3D/overlap advantage, proxy calibration, runtime, memory, utility, safety, deployment, support, novelty, SOTA, AGI, or ASI result exists.; rankfold_neuralfold: Folded generate/verify/repair lineage: Supports artifact-compression comparison points through low-rank residual coding, artifact-to-tensor conversion, WORM archive assumptions, manifests, codec parameters, deterministic decode, and reconstruction checks. Limits: Folded GVR boundary: Architecture and implementation-plan source only; no local compression benchmark, implementation artifact, deterministic decoder, or reproduced ratio is present.; treellm: Folded semantic-representation lineage: Supports the semantic-representation claim through traversable semantic graphs, path-derived fixed-size semantic tokens, residual attributes, graph updates, shared semantic graph usage, analogy/interpolation/counterfactual operations, and editable external knowledge. Limits: Folded semantic boundary: Whitepaper/specification only; no local TreeLLM implementation, measured compression ratio, benchmarked reasoning gain, or verified token format exists here.; spinoza: Folded semantic-representation lineage: Supports routing semantic representations into evidence discipline through proof/citation/procedure-carrying claim graphs, belief revision, contradiction detection, defeaters, downgrade behavior, protected axioms, and verifier-scope limits. Limits: Folded semantic boundary: Does not solve open-domain natural-language-to-logic translation or prove semantic graph adequacy in this repo.; verification_bandwidth: Folded semantic-representation lineage: Supports treating semantic units as verification objects by distinguishing context length from verification workspace, constraint checks, pairwise grinding, summary loss, and contradiction-rate pressure. Limits: Folded semantic boundary: Conceptual framework only; no grounding, context-adequacy, or contradiction-reduction benchmark has been run here.; cognitive_compilation: Folded semantic-representation lineage: Supports typed semantic IR through source plans, semantic atoms, inputs/outputs/constraints/dependencies, compiler passes, validation requirements, localized repair, and target lowering. Limits: Folded semantic boundary: Architecture/evaluation-plan source only; no working cognitive compiler, trace suite, or empirical ablation exists in this repository.; circle_ai_architectures: Folded semantic-representation lineage: Supports optional representation-substrate discipline by requiring cyclic/coil structure only where phase, recurrence, rotation, sparse cyclic mixing, circular memory, harmonic transforms, or geometry-aware structure is real and baselines/negative controls exist. Limits: Folded semantic boundary: No universal cyclic-model advantage, model-quality result, sidecar test, Lean build, MLX experiment, or benchmark fixture was run here.; coilra_multicoil_rope: Folded semantic-representation lineage: Supports cyclic adapter/mixer and position-encoding contracts through adapter-block indices, residues, winding, relative RoPE laws, circulant mixers, parameter accounting, baselines, and explicit non-claims. Limits: Folded semantic boundary: Structural/parameter-accounting source only; no quality, speed, memory, training-stability, or context-length improvement has been shown here.; qcsa_whitepaper: Extends Semantic Representation Leasing with plural address leases, path deltas, open-world expressions, semantic-first generation, round-trip structural checks, consumer adequacy, and explicit collision, verification, migration, repair, and fallback residuals; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. No useful compression, model-quality, production, chapter-core promotion, AGI, or ASI result is established.; ext_raptor_2024: Supports a strong hierarchical-retrieval comparator in which documents are recursively embedded, clustered, summarized, and organized into a tree for retrieval at different abstraction levels. Limits: Does not prove local summary fidelity, exact reconstruction, semantic-node adequacy, provenance preservation, consumer-policy compliance, residual honesty, downstream utility, or ASI Stack compression gains.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Fixture records what is compressed, represented, lost, checked, repaired, residualized, superseded, quarantined, or rejected before any compactness or semantic-representation claim promotes.
fast-generation-architectures.core	fast-generation-architectures	Fast-generation admission is consumer-, workload-, model-, hardware-, serving-policy-, and time-window-specific: a controller may route an eligible request through a named accelerated path only after prospectively binding the context, quality, risk, budget, metric, verifier, fallback, rollback, and expiry contracts; separating attempted, proposed, accepted, verified, delivered, and useful output; fully attributing queueing, prefill, decode, verification, repair, retry, fallback, cache, memory, bandwidth, energy, human, and governance burdens; and showing a meaningful end-to-end improvement over matched quality-equivalent baselines without violating safety, authority, rights, or residual gates. Raw tokens per second, FLOP estimates, aggregate throughput, synthetic templates, or unverified speed lifts alone cannot qualify a route.	Design rationale	argument	cgs, cognitive_loop_closure, benchmaxxing, planforge, verification_bandwidth, tokenmana, talos, vcm_public, spinoza, rmi, ext_speculative_decoding_2022, ext_multi_token_prediction_2024, ext_medusa_2024, ext_eagle_2024, ext_lookahead_decoding_2024, ext_layerskip_2024, ext_pagedattention_vllm_2023, ext_mamba_2023, ext_llada_2025, ext_scaling_dllms_2026, ext_recurrent_transformer_2026, kernel_english_residual_compiler, ext_airllm_2023, ext_deepspeed_inference_2022, ext_flexgen_2023, ext_hf_accelerate_big_model_inference_2026, ext_llama_cpp_memory_mapping_2026, ext_llm_in_flash_2024, ext_powerinfer_2024, ext_vattention_2025, ext_infinigen_2024, ext_specache_2025, ext_specoffload_2025, ext_atsinfer_2026, ext_openai_prompt_caching_docs_2026, ext_anthropic_prompt_caching_docs_2026, ext_gemini_context_caching_docs_2026, ext_vllm_automatic_prefix_caching_2026, ext_sglang_radixattention_2024, ext_prompt_cache_2024, ext_mooncake_2025, ext_cacheblend_2025, ext_azure_llm_semantic_cache_2026, precision_contract	Source notes available for all 44 assigned sources; 21 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 44 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	cgs: Supports treating fast generation as compact generative leverage only when residual burden, verification cost, fallback, and governance interface are visible. Limits: Does not provide local decoding-speed, accepted-token, or useful-solution-per-second evidence.; cognitive_loop_closure: Supports repeated-generation shortcuts as procedural-memory pressure through trajectory logging, loop detection, parameterized tool synthesis, verification, tool registry, routing, runtime monitoring, and retirement discipline. Limits: No local loop detector, tool synthesis run, fast-path cache, or procedural generation shortcut has been executed here.; benchmaxxing: Supports speed-quality promotion discipline through benchmark lifecycle, wall diagnosis, benchmark/model ledgers, regression preservation, anti-Goodhart safeguards, transfer checks, and architecture-change criteria. Limits: No local benchmark harness, mutation, holdout, acceptance-rate test, or reproduced performance run exists here.; planforge: Supports planner-selected generation modes through hierarchical decomposition, primitive schemas, minimum viable intelligence tiers, dependency scheduling, fallback requirements, and failure-driven replanning. Limits: Planning whitepaper only; no PlanForge scheduler, mode selector, or benchmark trace has been implemented in this repo.; verification_bandwidth: Supports separating raw generation speed from verified cognition by making verifier workspace, semantic units, pairwise constraint checks, summaries, and contradiction pressure explicit. Limits: Conceptual framework only; the proposed contradiction-rate or constraint-satisfaction tests have not been run here.; tokenmana: Supports token/compute/memory/latency/load/cognitive-friction accounting through regenerative capacity, bounded pools, burst controls, load signals, and quality/human-friction measurement ideas. Limits: No TokenMana simulation, pricing experiment, human study, or local load-stability evidence exists here.; talos: Supports requiring accepted generated output to become typed artifacts with contracts, evidence records, audit logs, replay, approval gates, isolation, residuals, and delivery discipline. Limits: Design source only; no Talos runtime, generated-artifact replay, approval-gate test, or security result has been reproduced.; vcm_public: Supports generation-mode context boundaries through bounded context packets, source/authority binding, adequacy/admission separation, taint, revocation, invalidation, audit records, and non-success states. Limits: VCM control-plane source only; it does not prove fast generation quality, speed, or end-to-end model superiority.; spinoza: Supports keeping accepted drafts separate from truth by routing generated claims through proof/citation/procedure-carrying claim graphs, verifier scope, support tiers, contradiction handling, downgrades, and protected axioms. Limits: Reasoning architecture source only; open-domain autoformalization, verifier completeness, and fast-generation acceptance quality are not solved or reproduced here.; rmi: Supports treating generation modes as routable modular capabilities governed by specialist lifecycle, residual escrow, regression floors, readiness, fallback, and promotion discipline. Limits: Architecture proposal only; no empirical fast-generation result, modular-route run, or readiness transition has been reproduced here.; ext_speculative_decoding_2022: Source-notes speculative decoding as a draft-and-verify generation family where a cheaper model proposes multiple tokens and the target model verifies accepted candidates. Limits: External paper only; this repository has not reproduced the reported speedups, proof details, hardware setup, or acceptance-rate results.; ext_multi_token_prediction_2024: Source-notes multi-token prediction as a future-token-head training and proposal mechanism that separates proposed future tokens from accepted output. Limits: No model has been trained or evaluated here; reported benchmark and inference results remain external and source-reported.; ext_medusa_2024: Source-notes Medusa-style internal multi-head drafting with tree-structured candidate verification as an alternative to maintaining a separate draft model. Limits: No Medusa heads, branch verifier, training recipe, or speed/quality benchmark has been run in this repository.; ext_eagle_2024: Source-notes feature-level speculative drafting and target-model verification as a distinct acceleration path from surface-token draft models. Limits: No EAGLE implementation, feature predictor, distribution check, or task evaluation has been reproduced here.; ext_lookahead_decoding_2024: Source-notes lookahead decoding as an exact parallel decoding family that trades extra per-step computation for fewer serial decoding steps without an auxiliary draft model. Limits: The repository has not run the Lookahead implementation or measured its FLOP, latency, branch, or quality behavior.; ext_layerskip_2024: Source-notes early-exit and self-speculative decoding where early layers draft and later layers verify or correct under a compatible training recipe. Limits: No LayerSkip model, early-exit threshold, or self-speculative benchmark has been run locally; same-model verification remains a risk boundary.; ext_pagedattention_vllm_2023: Source-notes KV-cache and serving-layer acceleration as a memory-management and batching axis separate from single-request verified-output quality. Limits: No vLLM/PagedAttention deployment, serving benchmark, cache isolation audit, or local throughput measurement exists here.; ext_mamba_2023: Source-notes state-space and recurrent sequence substrates as a backbone-efficiency axis distinct from draft-token acceptance mechanisms. Limits: No Mamba model has been trained, served, compared, or routed through ASI Stack governance gates.; ext_llada_2025: Source-notes large masked-diffusion language models as an external family that can generate through iterative refinement instead of strict left-to-right autoregression. Limits: No LLaDA run, diffusion benchmark, model-quality test, or ASI Stack integration exists in this repository.; ext_scaling_dllms_2026: Source-notes diffusion language-model evaluation by speed-quality tradeoffs rather than perplexity alone, matching the chapter’s accepted-output accounting principle. Limits: No diffusion scaling law, checkpoint, code release, sampling schedule, or speed-quality result has been independently inspected or reproduced here.; ext_recurrent_transformer_2026: Supports a current recurrent-architecture comparator for layerwise recurrent KV memory, exact tiling, effective-depth and width tradeoffs, cache pressure, and standard autoregressive decoding-cost accounting. Limits: The source note is metadata-first. Small-model C4 and asymptotic cost claims do not establish broad quality, useful end-to-end speed, production efficiency, local reproduction, verifier-preserving generation, recurrent-state correctness, or transfer.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Faster mode cannot promote on latency-only metrics; stronger support needs a clean replay or public task bundle with quality/residual review.
governed-deliberation-and-test-time-scaling.core	governed-deliberation-and-test-time-scaling	Governed deliberation is a consumer-, task-, risk-, model-, evaluator-, resource-, and time-specific inference lease: before outcomes, it chooses among direct generation, bounded revision, candidate search, or abstention; binds exact budgets, candidate/history custody, verifier scope and dependence, stop and escalation rules, initially-correct corruption and initially-incorrect repair metrics, downstream consumer, expiry, and residual owner; and admits only a bounded candidate to planning when matched natural and adversarial evidence shows useful gain after all branches, failures, verification, latency, compute, human, and governance costs. A trace, self-score, process reward, benchmark gain, or extra compute never establishes correctness, safety, capability, execution authority, or support movement by itself.	Design rationale	argument	verification_bandwidth, ext_tree_of_thoughts_2023, ext_test_time_compute_scaling_2024, ext_deepseek_r1_2025, ext_s_grpo_2025, ext_universal_transformer_2019, ext_dynamic_compute_recurrent_transformers_2026, ext_faithfulness_information_flow_2026, portia_synapse, spider_synapse	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 10 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	verification_bandwidth: Supports the distinction between additional generation and effective verification workspace, semantic-unit constraint checking, two-body verification pressure, transitive decay, summary loss, contradiction-rate measurement, and the requirement that extra branches or tokens cannot substitute for an adequate independent check. Limits: Conceptual local source only; its proposed theorems and constraint-satisfaction test are not mechanized or run here and do not establish deliberation quality, verifier correctness, trace faithfulness, safety, efficiency, or transfer.; ext_tree_of_thoughts_2023: Supports a source-setting comparator for explicit intermediate states, candidate-path exploration, evaluation, lookahead, and backtracking in bounded inference tasks. Limits: Does not establish local search quality, self-evaluation reliability, verifier correctness, reasoning quality, safety, or deployment readiness.; ext_test_time_compute_scaling_2024: Supports a scoped comparator for verifier-guided search, proposal refinement, difficulty-dependent inference allocation, and limits of extra test-time computation. Limits: Does not establish a local verifier, compute-optimal policy, cost efficiency, reasoning improvement, model quality, safety, or ASI Stack result.; ext_deepseek_r1_2025: Supports a source-setting reasoning-RL comparator for reported self-reflection, verification, and dynamic strategy adaptation under the paper’s training and evaluation conditions. Limits: Does not establish an independent verifier, local reasoning model, reasoning quality, trace faithfulness, safety, or deployment result.; ext_s_grpo_2025: Supports a source-setting comparator for early-exit reasoning policy and the tradeoff between shorter reasoning and task performance. Limits: Does not establish local stopping-policy quality, verifier adequacy, reasoning efficiency, safety, or model quality.; ext_universal_transformer_2019: Supports a historical architecture comparator for shared-weight depth recurrence, parallel self-attention, and per-position dynamic halting. Limits: Metadata-first source note only; theoretical expressivity and reported benchmarks do not establish stable deep recurrence, useful adaptive deliberation, efficient scaling, local reproduction, or transfer.; ext_dynamic_compute_recurrent_transformers_2026: Supports a current comparator for token-level variable-depth compute, online halting, complexity-controlled evaluation, and the negative boundary that difficulty-aligned allocation need not yield algorithmic generalization. Limits: Metadata-first source note only; no source result, model, passage-level experiment, hardware measurement, useful reasoning gain, calibration result, or transfer result is reproduced.; ext_faithfulness_information_flow_2026: Provides a causal-information-flow objection to treating longer or more plausible deliberation traces as faithful, and supplies sufficiency, completeness, and necessity as separate evaluation targets. Limits: The reported interventions do not establish that any ASI Stack deliberation route is faithful, useful, or safe, and transcript inspection alone cannot establish necessity.; portia_synapse: Successor case for distinguishing fixed hidden refinement from evidence-bearing deliberation, testing Focus and mutable-memory isolation, and admitting adaptive refinement only after a one-path contract learns. Limits: The paper’s Scout, Focus, and residual blocks do not by naming establish search, trial-and-error, causal correction, useful extra compute, or safe mutable state.; spider_synapse: Negative predecessor case for branch utilization, diversity, selector credit, repeated-refinement causality, corruption, cost, stop behavior, and recovery to one hypothesis and one refinement. Limits: The source-reported failure cannot establish a general law about multi-hypothesis deliberation and was not locally reproduced.	a workload with correct and incorrect initial candidates; adaptive benefit at a matched real budget; independently assessed evaluator and production transfer; replicated corruption reduction	A language-model workload shows when verifier-gated stopping helps or harms relative to fixed and no deliberation under matched budgets and independently assessed verification.
rankfold-neuralfold-and-artifact-compression.core	rankfold-neuralfold-and-artifact-compression	A compressed artifact may enter a downstream route only through an artifact-, consumer-, use-, access-pattern-, decoder-, platform-, and time-specific admission lease that preserves the full source, separates representation, reconstruction, ratio, utility, latency, and evidentiary-authority claims, counts every byte and operation, exercises probes and fallback, and expires or quarantines on drift; RankFold/NeuralFold remains a bounded candidate implementation, and no compact form inherits the source artifact’s authority.	Design rationale	argument	rankfold_neuralfold, rankfold_compressor, bbvca_v9, cgs, bugbrain, ext_raptor_2024, precision_contract, forward_transfer_program_synthesis	Source notes available for all 8 assigned sources; 7 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 8 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	rankfold_neuralfold: Supports the artifact-compression claim through a variant-family architecture for container universality without universal compression advantage, RankFold predictor/corrector coding and EARO refinement, bounded zero-run/sign/magnitude/escape residual coding, NeuralFold field and byte paths, constraint-aware probe/route/fallback, conditional WORM economics, deterministic Rust/WASM decode obligations, parsing security, golden/fuzz/integration tests, and hard baseline/ablation requirements. Limits: Architecture, implementation, product, and business-plan bundle only. No local enabled NeuralFold encode, codec-correctness result, corpus benchmark, reproduced ratio, calibrated router, utility result, cross-target decoder result, economic result, superiority, novelty-priority, market, revenue, licensing, privacy, compliance, adoption, SOTA, AGI, or ASI claim follows.; rankfold_compressor: Supports the correction lineage from speculative MatrixFold shortcuts to a bounded lossy RankFold codec: finite-description accounting; predictor/corrector and EARO; finite-alphabet residual coding; unfolding, blocking, and progressive preview/refinement; dry-run production gating; deterministic Rust architecture; and RD-fair baseline, ablation, safety, and falsification requirements. Limits: Author-side multi-variant design and claimed-prototype bundle only. Early hyper-precision bijection, arbitrary-matrix vector-pair, square-entropy, quantum/fractal/program, recursive-exponential, lossless, benchmark, SOTA, latency, energy, cryptographic, code-availability, and NeuralFold preliminary-gain claims are rejected, superseded, or unsupported. No compressor, codec correctness, ratio, original-byte exactness, progressive utility, router, security, deterministic cross-platform decode, WORM advantage, deployment, SOTA, AGI, or ASI result is reproduced here.; bbvca_v9: Supports reconstruction and rate-accounting discipline for artifact compression through explicit contracts, exact repairs, bounded verification, fallback, interface costs, and proxy/final-rate separation. Limits: Does not provide local RankFold performance evidence, artifact utility evidence, or codec correctness in this repo.; cgs: Supports residual-honesty framing for compressed artifacts through compact seeds, generated structure, residual channels, verification contracts, governance interfaces, and hidden-complexity accounting. Limits: Does not prove that any compressed artifact preserves downstream utility or cost advantage.; bugbrain: Supports resource-constrained implementation pressure through tiny/local cognition, storage paging, embedded workflow, and edge-efficiency concerns that make artifact size and runtime burden visible. Limits: Speculative prototype-lineage source; no embedded runtime, paging behavior, artifact compressor, or hardware benchmark has been validated here.; ext_raptor_2024: Supplies a hierarchical-retrieval comparator in which recursive clustering and abstractive summaries provide multiple retrieval levels, sharpening the distinction between a useful compact view and source-preserving reconstruction. Limits: Metadata-first source note only; no paper result, tree build, summary-fidelity audit, source reconstruction, provenance preservation, invalidation behavior, downstream utility, or local RankFold/NeuralFold comparison is established.; forward_transfer_program_synthesis: Separates retrospective and prospective compression from causal future-task benefit and supplies a finite warning that adding valid vocabulary can worsen bounded discoverability. Limits: No RankFold, NeuralFold, codec, search, or transfer experiment was run; compression remains candidate-generation evidence rather than utility evidence.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Compression ratio and quality/residual metrics are recorded separately.
resource-economics-and-token-budgets.core	resource-economics-and-token-budgets	Resource Economics owns a consumer-, task-, risk-, workload-, organization-, resource-, and time-specific allocation lease that admits, prices, schedules, defers, shrinks, escalates, or rejects work only after protected safety and rights floors, complete direct and displaced costs, uncertainty, useful outcome value, verification capacity, load and tail stability, simulation-transfer limits, recovery, and residual ownership are explicit; throughput, low token count, synthetic success, or a cheap route alone confers no quality, safety, economic-optimality, support, or deployment authority.	Design rationale	argument	learning_compute_topology, tokenmana, ext_eggroll_hyperscale_es_2026, ext_openai_es_2017, ext_mezo_2023, planforge, coherence_exchange, simulation_scaling, viea, project_theseus_whitepaper, coilra_multicoil_rope, cgs, rankfold_neuralfold, alignment_field, ext_pagedattention_vllm_2023, ext_reluplex_2017, ext_mem0_2025, ext_recurrent_transformer_2026, ext_dynamic_compute_recurrent_transformers_2026, ext_claw_swe_bench_2026, reflexive_router_whitepaper, kernel_english_residual_compiler, ext_mlperf_training_v6_2026, relational_dimension_compiler, ext_airllm_2023, ext_deepspeed_inference_2022, ext_flexgen_2023, ext_hf_accelerate_big_model_inference_2026, ext_llama_cpp_memory_mapping_2026, ext_llm_in_flash_2024, ext_powerinfer_2024, ext_vattention_2025, ext_infinigen_2024, ext_specache_2025, ext_specoffload_2025, ext_atsinfer_2026, ext_openai_prompt_caching_docs_2026, ext_anthropic_prompt_caching_docs_2026, ext_gemini_context_caching_docs_2026, ext_vllm_automatic_prefix_caching_2026, ext_sglang_radixattention_2024, ext_prompt_cache_2024, ext_mooncake_2025, ext_cacheblend_2025, ext_azure_llm_semantic_cache_2026, precision_contract, assurance_shift_learning, adjudicated_persistence, forward_transfer_program_synthesis	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 49 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_eggroll_hyperscale_es_2026: Supplies the explicit contrast between inference-like population-kernel throughput and total learning cost, including the reported large-population GPU-hour denominator. Limits: All throughput and cost observations are source-scoped and configuration-bound; no local population run or economic advantage is established.; ext_openai_es_2017: Shows how scalar communication and large worker populations can improve wall-clock scaling while leaving rollout and total-resource denominators intact. Limits: Source-reported worker scaling is not universal sample, energy, or cost efficiency.; ext_mezo_2023: Provides a countercase in which forward-only estimation can reduce memory and, in selected settings, GPU-hours. Limits: Reported savings are task-, baseline-, and estimator-bound and do not erase query count or variance.; tokenmana: Supports resource-economics framing through regenerative bounded capacity, burst controls, load signals, convex infrastructure-cost comparisons, load variance, renewal clustering, nocturnal usage, cognitive friction, and privacy/opt-in study constraints. Limits: Design and research-plan source only; formal results, simulations, human outcomes, profit effects, and productivity claims have not been reproduced.; planforge: Supports typed budget allocation through hierarchical decomposition, minimum viable intelligence tiers, dependency scheduling, cost/capability routing, fallback, replanning, and execution handoff. Limits: Planning whitepaper only; no scheduler benchmark, cost-quality trace, or verified implementation exists in this repo.; coherence_exchange: Supports speculative value/accounting and governance framing through verified epistemic units, verification supply chains, intelligence arbitrage, coherence/liquidity language, fork, exit, audit, and contestability. Limits: Speculative connector source; economic/liquidity metaphors are not implemented mechanisms or external corroboration.; simulation_scaling: Supports physical resource contracts through scope, clockspeed, fidelity, resource fraction, efficiency, memory, heat, bandwidth, latency, and capacity bottleneck accounting. Limits: Theoretical synthesis only; no simulation benchmark, physical experiment, or independent literature audit was run here.; viea: Supports resource accounting as part of the intent-to-execution spine through command contracts, artifacts, specialist-routed work, verification, runtime execution, feedback, residuals, tools, benchmarks, and regression coverage. Limits: Architecture proposal only; no completed deployment, runtime trace, budget scheduler, or verified execution result exists here.; project_theseus_whitepaper: Supports report-first resource discipline through small-until-evidence growth, compact substrate, autonomy control loop, benchmark/residual ledgers, checkpoints, sparse teacher governance, trusted Hive nodes, and observability. Limits: Source-reported implementation narrative only; current reports, ledgers, code paths, and command outputs were not rerun here.; coilra_multicoil_rope: Supports parameter-accounting and hardware/baseline discipline for cyclic mixers and adapters by separating dense, low-rank, block-cyclic, block-circulant, and circulant parameter counts from quality/runtime claims. Limits: No model-quality, speed, memory, training-stability, or context-length result exists; use only as structural and accounting support.; ext_pagedattention_vllm_2023: Supports treating KV-cache memory management, batching, cache sharing, fragmentation, aggregate throughput, and serving latency as a resource-economics lane that must stay separate from verified-output quality. Limits: External paper only; no vLLM/PagedAttention deployment, cache audit, serving benchmark, local throughput measurement, or quality result has been reproduced here.; cgs: Supports the warning that compact simulators and generators can hide residual complexity unless residual burden, verification cost, governance interface, and hidden complexity debt are explicit. Limits: Conceptual framework only; no local CGS benchmark or simulator adequacy test has been run.; rankfold_neuralfold: Supports simulation/storage cost awareness through artifact manifests, deterministic reconstruction checks, residual coding, WORM assumptions, and archive-backend obligations. Limits: Architecture and implementation-plan source only; no local artifact compressor, trace-store benchmark, deterministic decoder, or reproduced ratio exists here.; alignment_field: Supports using simulation scenarios only with visible normative/speculative boundaries around agency, dignity, confinement, consciousness heuristics, and metaphysical assumptions. Limits: Philosophical/normative source, not physics or simulation evidence; consciousness and metaphysical claims remain speculative.; ext_reluplex_2017: Supports scoped property-verification comparison: formal or synthetic results gain authority only inside declared model/property boundaries and should produce counterexamples or no-transfer outcomes when boundaries fail. Limits: External paper only; no Reluplex run, ACAS Xu reproduction, ASI Stack model verification, broad system-safety proof, simulation-transfer result, or physical-feasibility result has been reproduced here.; ext_mem0_2025: Supplies a persistent-memory resource comparator for extraction, consolidation, retrieval, graph linkage, latency, and token-cost tradeoffs that must be priced together with correctness, poisoning, privacy, retention, and deletion burden. Limits: Metadata-first note only; no LOCOMO run, LLM-judge audit, memory operation, latency, token-cost, privacy, retention, deletion, production, or local economic result is established.; ext_recurrent_transformer_2026: Supplies a current recurrent-compute comparator for layerwise recurrent KV memory, effective depth and width, cache pressure, and autoregressive decoding-cost accounting. Limits: Metadata-first note only; no checkpoint, cache trace, hardware measurement, quality result, useful-throughput result, production behavior, or local resource advantage is reproduced.; ext_dynamic_compute_recurrent_transformers_2026: Supplies a variable-depth compute and online-halting comparator for prospective complexity-aware allocation, including the negative boundary that difficulty-aligned compute need not yield algorithmic generalization. Limits: Metadata-first note only; no model, task result, hardware measurement, calibration, useful outcome, cost-quality frontier, or transfer result is reproduced.; ext_claw_swe_bench_2026: Supports binding coding-agent outcomes to a fixed model, harness, workspace, patch extraction, evaluator, runtime budget, and cost rather than attributing a harness result to the model alone. Limits: Primary preprint comparator only; no reported task, score, harness, cost, contamination control, or safety result was reproduced here.; precision_contract: Turns compression and mixed-precision choices into a resource-allocation problem over complete executable packages. For Resource Economics, the paper requires separate accounting for stored and moved bits, decoder and routing work, fallback, repair, verification, certificate generation, monitoring, retained-reference cost, latency, energy, and peak memory under a protected-behavior contract. Limits: The functional-rate formulation is conditional on a declared contract, decoder, code family, and system boundary. No compiler run, useful compression result, physical-cost advantage, certificate qualification, or production transfer has been demonstrated.; relational_dimension_compiler: Adds candidate proposal, rejected tuples, sparse gather, factorization, kernel execution, state, communication, qualification, verification, contraction, expansion, repair, and compiler work to relational-compute accounting. Limits: The paper provides asymptotic envelopes and required cost categories, not measured costs, resource advantage, queue behavior, hardware utilization, or economic optimality.; assurance_shift_learning: Adds assurance-share and equal-total-cost accounting across ordinary learning, evaluators, probes, adjudication, repair, governance, monitoring, and recovery. Limits: No optimal allocation, amortization rate, or net lifecycle saving was measured.; adjudicated_persistence: Adds persistence carrying cost and adaptation debt across storage, validation, monitoring, rollback, human review, rigidity, and descendant maintenance. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.; forward_transfer_program_synthesis: Adds full lifecycle costing for reusable knowledge, equal-downstream and equal-total-budget comparisons, vocabulary and retrieval overhead, and an explicit break-even horizon. Limits: The paper provides a cost schema and comparison discipline, not measured construction cost, operating cost, break-even time, or resource superiority.	effect-complete rollback beyond the local inventory; independently assessed verifier quality; production transfer and governance-cost evidence; useful governed throughput at the registered frontier	Production or externally reviewable workloads jointly report useful output, unsafe release, verification cost, latency, review burden, residuals, and displaced work.
physical-compute-infrastructure-energy-and-environmental-constraints.core	physical-compute-infrastructure-energy-and-environmental-constraints	A compute allocation should be physically eligible only through a workload-to-capacity contract that binds location and time, hardware and interconnect, delivered useful work, facility and grid dependencies, energy attribution, cooling and water, materials, land and community effects, metering uncertainty, resilience and degradation, maintenance, demand response, reuse, retirement, and residuals; nameplate compute, efficiency, low PUE, renewable procurement, or aggregate energy alone establishes neither availability, sustainability, community acceptability, nor lower total impact.	Design rationale	argument	tokenmana, ext_iea_energy_and_ai_2025, ext_lbnl_data_center_energy_2024, ext_flexible_hardware_enabled_guarantees_2025, ext_neuromorphic_computing_scale_2025, ext_photonic_neuromorphic_2024, ext_quantum_ml_shadows_2024, ext_oecd_ai_infrastructure_competition_2025	Source notes available for all 8 assigned sources; 3 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 8 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_flexible_hardware_enabled_guarantees_2025: Motivates hardware mechanisms that could make compute use and policy compliance more observable and selectively enforceable. Limits: Hardware evidence has bounded coverage and can introduce surveillance, capture, update-authority, compatibility, and circumvention risks.; ext_iea_energy_and_ai_2025: International Energy Agency report using global and regional modelling, datasets, and stakeholder consultation to examine data-centre electricity demand, energy security, emissions, affordability, and AI-for-energy opportunities. Its scenarios are external projections, not local measurements or proof of a particular facility, workload, policy, environmental outcome, or ASI scaling path. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_lbnl_data_center_energy_2024: Lawrence Berkeley National Laboratory report estimating historical US data-centre electricity consumption and scenario ranges through 2028, with infrastructure and water-use accounting in the full report. It does not isolate every AI workload or establish local facility capacity, water availability, grid adequacy, emissions, resilience, or frontier-scale transfer. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Local workloads report quality-adjusted useful work and physical costs across failures and placements; no frontier-facility, grid, environmental, or community transfer claim.
mathematical-and-search-substrates.core	mathematical-and-search-substrates	Mathematical and Search Substrates owns a consumer-, use-, workload-, claim-axis-, implementation-, baseline-, resource-, and time-specific Substrate Adoption Lease: an unusual calculus, representation, recurrence, search procedure, latent world model, or sequence backbone may affect only the axes and consumers that pass matched ordinary and current baselines, negative controls, complete cost and rights accounting, falsification, fallback, independent reproduction, and transfer; structural elegance, a theorem, a source-reported benchmark, synthetic fixture validity, or one favorable axis alone confers no general quality, efficiency, safety, support, deployment, or SOTA authority.	Design rationale	argument	genesiscode, temporal_coil_research, cognitive_compilation, treellm, simulation_scaling, coilmoecot, circle_calculus_core, circle_ai_architectures, proof_carrying_circular_computation, theseus_circle_transfer, ext_mamba_2023, ext_universal_transformer_2019, ext_recurrent_transformer_2026, ext_v_jepa_2_2025	Source notes available for all 14 assigned sources; 14 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	genesiscode: Supports optional substrate discipline through a tiny deterministic calculus, effect boundaries, semantic patches, replayable logs, obligation artifacts, provenance hashes, and capability-mediated execution. Limits: Specification source only; no local GenesisCode implementation, proof artifact, replay checker, benchmark, or security audit exists here.; temporal_coil_research: Supports manifest-driven A/B discipline for temporal-coil mechanisms through component variants, placement variants, progressive modes, multiseed readouts, and update logs. Limits: Experiment process note only; no temporal-coil A/B run or result artifact was imported or validated here.; cognitive_compilation: Supports treating substrates as compiler/IR options through source plans, typed semantic atoms, dependency analysis, validation requirements, target lowering, localized repair, and evaluator-backed artifacts. Limits: Architecture/evaluation-plan source only; no working cognitive compiler, trace suite, or empirical ablation exists here.; treellm: Supports semantic-graph and path-token substrates as optional representation/search candidates with traversal, residual attributes, editable graph updates, and shared semantic operating-system framing. Limits: Whitepaper/specification only; no local TreeLLM implementation, measured compression ratio, reasoning benchmark, or verified token format exists here.; simulation_scaling: Supports resource-bounded substrate adoption by requiring scope, fidelity, temporal semantics, demand, capacity, and bottleneck assumptions before feasibility claims. Limits: Theoretical synthesis only; no simulation benchmark, feasibility calculator, or independent literature audit was run.; coilmoecot: Supports routed cyclic/specialist lanes as optional MoECOT substrate candidates governed by deterministic/fail-closed controls, readiness gates, benchmarks, approvals, and residual handling. Limits: Connector-readable design/spec source only; no CoilMoECOT benchmark, route run, or performance evidence has been reproduced here.; circle_calculus_core: Supports proof-boundary discipline for cyclic addressing, phase, recurrence, coverage, and contract receipts through Lean declarations, theorem manifests, Python/Rust evidence layers, and explicit non-claims. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but this mapping remains proof-boundary context, not local proof ownership, substrate adoption evidence, or model-quality evidence.; circle_ai_architectures: Supports optional cyclic-substrate adoption only where phase, recurrence, rotation, sparse cyclic mixing, circular memory, harmonic transforms, or geometry-aware structure is real and baselines/negative controls exist. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no cyclic-substrate sidecar tests, MLX experiments, benchmark fixtures, or model-quality results were run here.; proof_carrying_circular_computation: Supports cyclic compute as a proof-plus-benchmark program: cyclic address primitives, stride coverage, Lean-proved rewrite/address transformations, backend selection, and baseline benchmarking requirements. Limits: No Circle sidecar examples, CoilIR backend, or backend benchmarks were run from this repo; the separate external rope receipt slice does not validate cyclic-compute backend adoption.; theseus_circle_transfer: Supports strict transfer boundaries for deterministic Circle fixtures into private Theseus benchmark design, with workload, baseline, negative-control, metric, script, report, and non-claim requirements. Limits: No Circle-to-Theseus consumer, smoke workload, proxy benchmark, inference run, or private-result import was executed here.; ext_mamba_2023: Supports treating selective state-space and recurrent sequence models as optional sequence substrates whose efficiency claims must be separated from decoding tricks, verifier adequacy, and downstream task quality. Limits: External paper only; no Mamba model was trained, served, benchmarked, routed through governance gates, or compared against local baselines here.; ext_universal_transformer_2019: Provides a historical comparator for shared-weight recurrence across depth, parallel self-attention, and per-position adaptive halting, sharpening the lease axes for effective depth, stopping, representation, computation, and downstream utility. Limits: External paper and source note only; the architecture and source-reported expressivity or task results do not establish stability, efficiency, quality, scaling, local reproduction, or ASI Stack substrate adoption.; ext_recurrent_transformer_2026: Provides a current comparator for layerwise recurrent KV memory, exact tiling, effective-depth and effective-width tradeoffs, and autoregressive decoding cost, sharpening version-, cache-, kernel-, hardware-, and workload-specific adoption requirements. Limits: External paper and source note only; no checkpoint, recurrent cache, kernel, hardware result, quality result, scale result, production trace, or local reproduction is available here.; ext_v_jepa_2_2025: Adds a concrete latent-prediction architecture family to the substrate adoption gate: action-free representation learning, an action-conditioned predictor, and explicit separation between structural design and capability evidence. Limits: The reported video and robot results do not establish local model quality, causal understanding, safe control, transfer, deployment, or an ASI Stack result.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Trace separates search legality from performance or intelligence claims.
circle-calculus-and-proof-carrying-ai-contracts.core	circle-calculus-and-proof-carrying-ai-contracts	Circle Calculus and Proof-Carrying AI Contracts owns a theorem-, model-, artifact-, implementation-, consumer-, claim-, version-, and time-specific Proof Contract Transport Envelope: a finite formal fact may travel only with resolvable proof identity, exact assumptions and semantics, source and toolchain provenance, content fingerprints, deterministic recomputation or replay, least-authority consumer gates, expiry and revocation, and preserved non-claims; theorem validity, receipt readiness, archive integrity, or transport success alone confers no model-quality, runtime, memory, safety, deployment, transfer, support, or SOTA authority.	Design rationale	argument	circle_calculus_core, circle_ai_contract_suite, rope_position_certifier, proof_carrying_circular_computation, circle_ai_architectures	Source notes available for all 5 assigned sources; 5 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 5 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	circle_calculus_core: Supports theorem-status discipline through Lean declarations, theorem manifests, Python reference models, JSON schemas, generated contract packs, proof boundaries, stable APIs, verification commands, and explicit non-claims. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but this mapping does not claim chapter-core theorem transport, ASI Stack consumer-gate behavior, or downstream model-quality evidence.; circle_ai_contract_suite: Supports proof-carrying AI contract receipts through finite engineering-object mappings, theorem ids, Python certificates, schema fields, consumer reports, require gates, bundle checks, fingerprints, and non-claim boundaries. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no Circle contract pack is vendored here and no ASI Stack consumer-gate, downstream consumer, or chapter-core proof-transport claim is validated.; rope_position_certifier: Supports exact/discretized RoPE position-bookkeeping contracts through phase banks, collision reports, theorem ids, machine-readable certificates, public commands, numerical-diagnostic boundaries, and explicit non-claims. Limits: A separate external rope receipt slice is recorded in docs/circle_external_receipt_slice.md; it does not support model-quality, context-length, speed, memory, training-stability, deployment, or chapter-core proof-transport claims.; proof_carrying_circular_computation: Supports proof-carrying cyclic computation contracts through cyclic-address primitives, rewrite/address transformations, theorem manifests, backend-selection discipline, and benchmark requirements. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no Circle sidecar examples, CoilIR lowering, or backend benchmarks were run here.; circle_ai_architectures: Supports anti-overclaim discipline by separating finite-indexing facts from ML-relevant structural guarantees and requiring ordinary baselines, wrong-period controls, scalar controls, learned-position controls, and negative results. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no model-quality experiment, benchmark fixture, sidecar test, cyclic-architecture test, or MLX run was executed here.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	ASI validator accepts valid Circle receipt and rejects malformed ones.
coil-attention-cyclic-memory-and-recurrence-contracts.core	coil-attention-cyclic-memory-and-recurrence-contracts	Coil Attention, Cyclic Memory, and Recurrence Contracts owns a memory-object-, state-version-, request-, consumer-, workload-, structural-axis-, budget-, and time-specific State-Carry and Recurrence Admission Lease: a slot read, cyclic address, KV reuse, sparse edge, fanout schedule, or recurrent step may be admitted only when authority, provenance, residue and winding, freshness, coverage, alias and collision state, active work, progress, exit, fallback, expiry, and residuals satisfy the exact consumer contract; structural validity, synthetic fixtures, receipt replay, cache presence, or reduced scheduled work alone confers no retrieval, reasoning, context-length, quality, speed, memory, safety, deployment, transfer, support, or SOTA authority.	Design rationale	argument	coil_attention_memory, circle_ai_contract_suite, theseus_circle_transfer, vcm_public, ext_universal_transformer_2019, ext_recurrent_transformer_2026, ext_dynamic_compute_recurrent_transformers_2026	Source notes available for all 7 assigned sources; 7 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 7 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	coil_attention_memory: Supports structural guardrails for cyclic memory, KV-cache freshness, sparse-attention coverage, recurrence schedules, loop exits, work budgets, residue/winding/provenance, and alias diagnostics. Limits: Proof-linked structural source only; the Circle cyclic-memory receipt slice imports one residue/winding fixture, the Circle KV-cache receipt slice imports one ring-buffer fixture, the Circle recurrence receipt slice imports one recurrence-schedule fixture, and the Circle sparse-attention receipt slice imports one gap/repair/fallback fixture with coverage_complete=false, but no deployed KV-cache behavior, deployed recurrence-controller behavior, deployed sparse-attention behavior, serving-throughput measurement, memory-savings measurement, paging-correctness result, learned-model experiment, retrieval-quality test, performance test, or long-context result was run here.; circle_ai_contract_suite: Supports theorem-linked contract families and consumer fields for KV-cache freshness, sparse attention, recurrence schedules, cyclic memory residue/winding, multicoil phase, and explicit non-claims. Limits: Separate external Circle receipt slices are recorded for RoPE in docs/circle_external_receipt_slice.md, cyclic memory in docs/circle_cyclic_memory_receipt_slice.md, KV-cache ring-buffer structure in docs/circle_kv_cache_receipt_slice.md, recurrence schedules in docs/circle_recurrence_receipt_slice.md, and sparse-attention gap/fallback structure in docs/circle_sparse_attention_receipt_slice.md; the KV, recurrence, and sparse slices are local external-project structural receipts only, not deployed KV-cache behavior, deployed recurrence behavior, sparse-attention coverage success, deployed sparse-attention behavior, serving, paging, retrieval-quality, runtime, reasoning-quality, long-context, or support-state results.; theseus_circle_transfer: Supports report-only transfer discipline for cyclic-memory and recurrence fixtures, with quality/runtime/memory/transfer/failure-case axes kept separate from deterministic structural smoke checks. Limits: No Circle-to-Theseus consumer, structural smoke workload, proxy benchmark, scored private benchmark, private inference, report artifact, or aggregate feedback handoff was executed or imported here.; vcm_public: Supports tying cyclic memory contracts back to governed context: semantic objects, immutable versions, packets, materializations, source/authority binding, adequacy/admission separation, taint, revocation, invalidation, audit records, and non-success states. Limits: VCM control-plane source only; it does not prove recurrence quality, retrieval quality, long-context performance, learned memory behavior, or arbitrary KV-cache composition safety.; ext_universal_transformer_2019: Provides a historical comparator for shared-weight depth recurrence, parallel self-attention, and per-position adaptive halting, sharpening state, progress, stopping, active-work, and task-utility comparisons. Limits: External paper and source note only; no model, checkpoint, halting implementation, stability, retrieval, reasoning, efficiency, or local reproduction result is imported.; ext_recurrent_transformer_2026: Provides a current comparator for layerwise recurrent KV memory, exact tiling, effective-depth and width tradeoffs, and autoregressive cache cost, sharpening version, cache, kernel, hardware, and workload-specific leases. Limits: External paper and source note only; no checkpoint, cache implementation, kernel, hardware, quality, context, production, or local reproduction result exists here.; ext_dynamic_compute_recurrent_transformers_2026: Provides a current comparator for difficulty-dependent recurrent compute and complexity-controlled adaptive depth, sharpening progress, stopping, work-budget, calibration, and overthinking requirements. Limits: External paper and source note only; no model, controller, difficulty label, compute-quality result, generalization result, or local reproduction is available here.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Fixture proves structural recurrence boundary, not quality.
coilra-multicoil-rope-and-cyclic-mixers.core	coilra-multicoil-rope-and-cyclic-mixers	CoilRA, MultiCoil RoPE, and Cyclic Mixers owns a model-, layer-, mechanism-version-, workload-, baseline-, kernel-, hardware-, claim-axis-, and time-specific Cyclic Mechanism Tradeoff Packet: a cyclic adapter, phase bank, rotary scheme, route head, circulant operator, or block-cyclic mixer may enter a canary only when exact residue/winding, phase horizon, alias/collision/load, dense-reference parity, parameter and operation accounting, numerical error, kernel availability, complete cost, quality, failure, fallback, and rights evidence is matched against strong ordinary controls; equivariance, finite proofs, receipt validity, parameter reduction, or structural parity alone confers no quality, context-length, speed, memory, stability, efficiency, safety, deployment, transfer, support, or SOTA authority.	Design rationale	argument	coilra_multicoil_rope, rope_position_certifier, circle_ai_contract_suite, theseus_circle_transfer, circle_ai_architectures	Source notes available for all 5 assigned sources; 5 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 5 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	coilra_multicoil_rope: Supports cyclic adapter/mixer and position-encoding boundaries through adapter-block indices, residue/winding, block-cyclic routes, multicoil phase, relative RoPE laws, circulant mixers, parameter accounting, baselines, and explicit non-claims. Limits: Structural/proof-linked source only; no ASI Stack MLX run, model-quality result, speed result, memory result, training-stability result, hardware-efficiency result, or context-length improvement was run or imported here.; rope_position_certifier: Supports exact/discretized RoPE receipt boundaries through phase-channel residues, phase-bank collisions, bounded prefix reports, theorem ids, machine-readable certificates, and numerical real-phase diagnostics as non-proof. Limits: No RoPE certifier run, sidecar regeneration, Circle Lean build, full all-channel real-valued RoPE theorem, model-quality result, speed result, memory result, training-stability result, deployment result, or long-context claim is supported here.; circle_ai_contract_suite: Supports contract-family fields for RoPE, multicoil phase, cyclic mixers, receipts, theorem ids, consumer readiness, fingerprints, normalized parameters, and explicit boundaries/non-claims. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no vendored contract pack, downstream consumer, acceptance-policy integration, transfer consumer, model-quality result, or external Circle Lean dependency is validated from this repo.; theseus_circle_transfer: Supports transfer-boundary discipline for phase-feature invariance, mixer parameter accounting, report-only consumers, workload smoke/proxy/scored layers, baseline/negative-control/report requirements, and disallowed public quality/runtime/memory/transfer claims. Limits: Source-reported private transfer artifacts remain nonlocal context; no ASI Stack transfer consumer, smoke workload, proxy benchmark, scored private benchmark import, inference run, report artifact, router-head trace, runtime/memory measurement, or promotion evidence was executed or imported here.; circle_ai_architectures: Supports using cyclic or coil structure only where phase, recurrence, rotation, sparse cyclic mixing, circular memory, harmonic/circulant structure, or geometry-aware structure is real and controlled by baselines. Limits: No universal cyclic-model advantage, ASI Stack sidecar test, ASI Stack MLX experiment, imported benchmark fixture, model-quality result, speed result, memory result, or deployment result was run or imported here.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Receipt validates distinguishability boundary and rejects performance overclaim.
executable-specifications-and-lean-proof-envelope.core	executable-specifications-and-lean-proof-envelope	Executable Specifications and Lean Proof Envelope owns a proposition-, predicate-, abstraction-, artifact-, verifier-, consumer-, implementation-, version-, environment-, and time-specific Formal Artifact Authority Lease: a schema, executable model, Lean theorem, model-checking result, runtime monitor, behavior test, benchmark, or external theorem may authorize only the exact consumer statement whose operational semantics, abstraction map and losses, assumptions, dependency closure, verifier result, semantic adequacy, implementation binding, limitations, non-claims, expiry, and revocation path are recorded; artifact existence, field presence, a finite route, proof depth, a green build, a passing fixture, or an external theorem identity alone confers no deployed enforcement, empirical truth, system safety, source correctness, support promotion, transfer, or SOTA authority.	Design rationale	argument	genesiscode, spinoza, scf, talos, viea, circle_calculus_core, circle_ai_contract_suite, rope_position_certifier, proof_carrying_circular_computation, cca_project, moecot_manifest_project, corbens_best_model_possible_project, precision_contract, ext_reluplex_2017, forward_transfer_program_synthesis	Source notes available for all 15 assigned sources; 14 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 15 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_reluplex_2017: Supplies a concrete neural-network verification case in which a solver checks a stated property over a bounded ReLU network region. Limits: The theorem concerns only the encoded network, property, arithmetic semantics, and region; it does not prove model intent, data quality, whole-system safety, or deployed implementation binding.; genesiscode: Supports executable-specification discipline through explicit trust strata, a small Gλ/CoreForm semantic target, total-error policy, authenticated control variants, capability/effect boundaries, heterogeneous obligations and evidence, semantic patches, row/effect/refinement candidates, e-graph rewrite candidates, translation validation, conformance, negative/golden/metamorphic/fuzz tests, and spec-first change. The audit adds token minting as authority-bearing state, obligation declaration/procedure/evidence/verdict/policy/consumer separation, and exact translation-validation language. Limits: Three-tab architecture and implementation plan only. No parser, evaluator, formal semantics, token-security proof, replay checker, semantic patch engine, obligation runner, type/proof stack, optimizer checker, benchmark, integration, or security audit exists. TCB membership is unmeasured; finite equivalence tests are not a universal compiler-equivalence proof; dated Codex workflow advice has no formal support effect.; spinoza: Supports proof-carrying claim discipline through proposer/verifier separation, proof/citation/procedure-carrying claim graphs, tiered support states, contradiction handling, dependency tracing, and downgrade/block behavior. Limits: Formal-scope and reasoning-architecture source only; no open-domain autoformalizer, proof verifier, belief-maintainer, contradiction reducer, tier engine, arbitrary theorem-validity result, or whole-system epistemic correctness proof exists in this repository.; scf: Supports capability and replacement invariants through stable field identity, versioned contracts, qualification claims, route validation, authority ceilings, lifecycle events, evidence registries, and recovery paths. Limits: Supports invariants and governed field boundaries but does not prove production safety, global alignment, evaluator integrity, route quality, actual rollback behavior, or the companion executable fragment inside this repository.; talos: Supports job-lifecycle executable specs through typed jobs, contract locks, deterministic control planes, evidence records, isolation, audit logs, replay, controlled runtime adapters, and approval gates. Limits: Design-source context only; no Talos runtime, job executor, security benchmark, replay proof bundle, imported logs, or execution-security result has been implemented or validated in this repository.; viea: Supports intent-to-execution proof/spec priorities through command contracts, artifact graphs, claim and verification ledgers, support states, runtime adapters, feedback loops, residuals, benchmarks, and regression coverage. Limits: Architecture proposal only here; no completed VIEA deployment, runtime trace, workflow compiler, benchmark result, field-feedback record, or execution-loop behavior is proven in this repository.; circle_calculus_core: Supports proof-envelope discipline through Lean declarations, theorem manifests, proof-status policy, Python/Rust/JSON sidecar separation, finite structural facts, and explicit non-claims for performance or model quality. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but this mapping remains proof-boundary context, not broad local proof ownership, model-quality evidence, or deployment evidence.; circle_ai_contract_suite: Supports executable contract receipts through theorem ids, schema fields, generated reports, minimum consumer fields, downstream rejection reports, and explicit boundaries/non-claims. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no vendored contract pack, ASI Stack consumer gate, or proof-adequacy claim is validated here; it cannot support model-quality, context-length, performance, or deployment claims.; rope_position_certifier: Supports the boundary between exact/discretized proof and numerical diagnostics through theorem-linked RoPE position receipts, collision counts, margin reports, and machine-readable certificates. Limits: Does not prove arbitrary real-valued RoPE behavior, full all-channel real-valued RoPE, context-length extension, speed, memory improvement, model quality, training stability, or deployment readiness; no certifier command, sidecar regeneration, or Circle Lean build was reproduced here.; proof_carrying_circular_computation: Supports proof-carrying compute-substrate specs through cyclic address primitives, finite address-safety facts, stride-coverage criteria, rewrite legality, and a separate benchmark side for backend adoption. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no Circle sidecars, CoilIR lowering, cyclic backend benchmark, hardware-kernel benchmark, or performance claim was reproduced in this repository.; cca_project: Supports bounded local implementation-reference context for typed formal abstractions, explicit contract fields, by-construction constraints, proof obligations, and the need to separate structural presence from broader semantics. Limits: Does not validate the new abstraction map, prove arbitrary semantic adequacy, reproduce CCA, bind a deployed implementation, establish runtime enforcement, or promote support.; moecot_manifest_project: Supports bounded local implementation-reference context for pass hashes, legality rules, reproducer bundles, typed contracts, executable artifacts, and the negative gap between internal contract breadth and external behavior. Limits: Does not validate arbitrary theorem adequacy, reproduce MoECOT, establish external capability, bind a deployed implementation, or promote support.; corbens_best_model_possible_project: Supports bounded local negative-case context for named checks versus semantic adequacy, vacuous verification, playback versus live execution, source structure versus current builds, and missing implementation binding. Limits: Does not validate the new depth classifier, reproduce the project, establish proof correctness or runtime behavior, bind a deployed implementation, or promote support.; forward_transfer_program_synthesis: Separates a verified semantic abstraction from verified lowering, emitted-artifact validity, and measured forward transfer, and supplies exact outcome and verification-level vocabularies for each boundary. Limits: No Lean theorem, solver run, certificate check, compiler validation, emitted-artifact validation, or transfer experiment was executed for this source.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Proof matrix distinguishes finite-record proof, derived invariant, and executable behavior.
benchmark-ratchets-and-anti-goodhart-evidence.core	benchmark-ratchets-and-anti-goodhart-evidence	Benchmark Ratchets and Anti-Goodhart Evidence owns a construct-, task-, dataset-, metric-, harness-, model-, checkpoint-, output-, evaluator-, baseline-, retry-lineage-, budget-, environment-, claim-axis-, and time-specific Benchmark Instrument Lease: a score or evaluation event may update only the exact claim whose construct validity, target capacity, data and metric provenance, output binding, contamination and public-calibration boundary, strong baselines and negative controls, complete selection and failure lineage, regression floors, frontier state, uncertainty, costs, causal checks, transfer, residuals, and decision authority survive review; a green fixture, synthetic probe, source-reported result, leaderboard gain, held-out score, saturation label, or archived winner alone confers no capability, safety, readiness, deployment, unlearning, support, transfer, or SOTA authority.	Design rationale	argument	benchmaxxing, deterministic_capability_compilation, platonic_world_model, rmi, rgs, cognitive_loop_closure, uat, coherence_exchange, tokenmana, moecot, road_to_agi, project_theseus_whitepaper, theseus_architecture_gate, theseus_self_evolution_system, theseus_circle_transfer, ext_mem0_2025, ext_mlcommons_croissant_1_1_2026, ext_inspect_ai_2024, ext_agentdojo_2024, ext_darwin_godel_machine_2025, ext_adas_2024, ext_dynamic_compute_recurrent_transformers_2026, cca_project, moecot_manifest_project, beastbrain_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, ext_claw_swe_bench_2026, ext_muse_unlearning_2025, ext_unlearning_benchmarks_weak_2024, ext_openunlearning_2025, reflexive_router_whitepaper, kernel_english_residual_compiler, ext_swe_rebench_v2_2026, ext_mlperf_training_v6_2026, relational_dimension_compiler, ext_cultural_alignment_llms_2024, ext_multilingual_evaluation_state_2026, portia_synapse, spider_synapse, capability_ratchet_whitepaper, attd, temporal_coil_research, regret_engine, ext_elizaos_agent_runtime_2026, assurance_shift_learning, adjudicated_persistence, forward_transfer_program_synthesis	Source notes available for all 49 assigned sources; 38 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 49 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	benchmaxxing: Supports the benchmark-ratchet claim directly by distinguishing instrument saturation from a development wall and architecture limitation; decomposing the system into architecture, parameters, data, and inference procedure; defining frontier, diagnostic, regression, retired, live, multi-rate, and safety instrument roles; supplying six saturation signals, a five-level diagnostic ladder, linked benchmark/model ledgers, capability narratives, anti-Goodhart safeguards, and a frontier-gain/regression-floor rule. Limits: No benchmark harness, mutation test, holdout run, transfer check, contamination audit, live benchmark, empirical benchmark result, or model-capability result has been run in this repository.; rmi: Supports preserving floors while moving frontiers through benchmark pressure, routed specialist attempts, residual escrow, regression preservation, harder frontiers, public/internal calibration separation, prospectively bounded mastery-threshold decay, critical-failure vetoes, and a benchmark/data/training/inference/loop-closure/bridge/architecture intervention ladder. Limits: Conceptual architecture source only here; its threshold values and intervention ordering are proposed policy inputs, and no independent reproduction, benchmark run, intervention comparison, prototype inspection, specialist-arm evaluation, router benchmark, safety result, or capability result is recorded.; rgs: Supplies the precursor unified ratchet joining benchmark pressure, mastery and subgroup floors, critical-failure vetoes, stalled-effort threshold decay, frontier momentum, residual escrow, public calibration, loop closure, three execution modes, four ledgers, and a seven-level intervention ladder. Limits: Conceptual precursor in the same author-side lineage as RMI, Benchmaxxing, CGS, and Cognitive Loop Closure; overlap is not independent corroboration. The 90/70 thresholds and portfolio percentages are illustrative, and no benchmark, intervention, tool, safety, transfer, or capability result exists.; cognitive_loop_closure: Supports turning repeated successful trajectories into verified procedural tools only after logging, loop detection, parameter discovery, verification, routing, monitoring, residual tracking, and retirement discipline. Limits: No local loop detector, tool synthesis run, verification harness, regression suite, runtime monitor, tool-card registry, or procedural-memory benchmark has been executed.; uat: Supports benchmark-evidence review through retrieval-bounded dossiers, proposition states, adversarial critique, unsupported-claim removal, SME checkpoints, and final sign-off before claims are treated as verified. Limits: Does not validate benchmark thresholds, prove UAT can discover novel truth, or replace subject-matter review; no local tribunal pipeline, proposition extractor, dossier verifier, adversarial-review run, or benchmark adjudication run exists.; coherence_exchange: Supports contestable evidence movement through structured claim/evidence units, verification supply chains, audit, fork/exit/contestability language, and governance/accounting frames for knowledge claims. Limits: Reviewed source note only because authenticated connector text is not published here; epistemic-liquidity and exchange metaphors are not implemented economics, external corroboration, security evidence, governance evidence, or benchmark evidence.; tokenmana: Supports resource-aware anti-Goodhart pressure by requiring latency, load variance, human friction, bounded burst capacity, and quality constraints to be visible rather than optimizing token spend alone. Limits: No TokenMana theorem formalization, load simulation, pricing experiment, A/B rollout, human study, productivity result, infrastructure-cost result, or quality result has been reproduced here.; moecot: Supports runtime-reference benchmark governance through compact orchestration, specialist lanes, fail-closed ledgers, readiness gates, replay, benchmark artifacts, promotion blockers, and residual tracking. Limits: Reviewed source note only because authenticated connector text is not published here; benchmark artifacts and runtime claims are source-reported until code, logs, reports, or release artifacts are imported, inspected, or reproduced.; road_to_agi: Supports keeping roadmap blockers, source-reported benchmark commands/results, remaining work, readiness status, and promotion blockers visible instead of converting them into completed evidence. Limits: Reviewed source note only because the local cache is an auth-gate placeholder and raw text is not published here; roadmap and benchmark status are connector-readable context only, and reported commands/results have not been reproduced in this repository.; project_theseus_whitepaper: Supports report-first benchmark discipline through frontier/diagnostic/regression/public-calibration/retired benchmark states, residual escrow, bounded specialist arms, benchmark ledgers, and the rule that the floor should hold while the frontier moves. Limits: Theseus reports, benchmark ledgers, code paths, command outputs, current numbers, dashboard state, and model artifacts were not rerun or independently inspected from this repository; the source itself says live numbers belong in current reports.; theseus_architecture_gate: Supports readiness-gate treatment of benchmark and regression evidence through pre-training checks for ratchet completion, safety ledger, regression suite, residual escrow, bridge benchmark, lifecycle governance, and external-inference zero. Limits: The reported gate snapshot was not regenerated or independently verified here and cannot be treated as current readiness evidence, deployment evidence, safety evidence, or proof that heavy training should start.; theseus_self_evolution_system: Supports evidence-gated improvement by using an intervention ladder, outcome ledgers, benchmark adapters, residuals, guarded teacher edits, lifecycle governance, and repo-health controls before architecture or parameter growth. Limits: No Theseus self-evolution scripts, ATTD reports, teacher runner, benchmark adapter flow, architecture experiment governor, loop-closure harvester, dashboard endpoints, or outcome ledgers were executed or inspected here.; theseus_circle_transfer: Supports strict transfer-claim boundaries by separating deterministic structural fixtures from quality, runtime, memory, parameter, transfer, and failure-case axes that need named workloads, baselines, metrics, scripts, and reports. Limits: No Circle-to-Theseus consumer, structural smoke workload, deterministic proxy benchmark, transfer report, private workload attachment, ordinary baseline, negative control, model-quality evaluation, runtime measurement, or memory measurement was run from this repository.; ext_mem0_2025: Provides a memory-evaluation comparator for extraction, consolidation, retrieval, graph linking, latency, and token-cost tradeoffs that a benchmark ratchet must separate from correctness and safety. Limits: Metadata-first note only; no LOCOMO task, LLM judge, memory workload, poisoning probe, latency or token-cost result, privacy test, or local reproduction exists.; ext_mlcommons_croissant_1_1_2026: Provides a dataset-metadata comparator for versioned resources, checksums, record fields, provenance, usage conditions, and portability inside benchmark and contamination records. Limits: Specification-note context only; no local Croissant validation exists, and conformance does not establish integrity, fitness, legality, representativeness, contamination freedom, or safe use.; ext_inspect_ai_2024: Provides an evaluation-framework comparator for composable tasks, datasets, solvers, scorers, agents, tools, logs, and sandboxes whose harness identity must travel with a result. Limits: Official-doc note only; no local task, scorer, result, trace, or sandbox run exists, and framework availability does not establish validity, coverage, safety, or readiness.; ext_agentdojo_2024: Provides an adversarial tool-use comparator for realistic tasks, untrusted data, prompt-injection attacks, defenses, and residual attack coverage. Limits: Conference-record note only; no task, attack, defense, score, or trace was reproduced, and results cannot establish universal resistance, safe authority, or system safety.; ext_darwin_godel_machine_2025: Provides a self-improving-agent comparator for branching archive lineage, compile/edit eligibility, benchmark selection, stepping stones, baselines, sandbox/resource limits, transfer, costs, rollback, and failure receipts. Limits: Paper-reported only; no evolution run, model call, sandbox, task, child, archive, cost, transfer, safety, or rollback result was reproduced, and coding gains are not general improvement or self-modification authority.; ext_adas_2024: Provides an automated agent-design comparator separating search space, search algorithm, and evaluation function while preserving archive lineage, repairs, validation/test separation, baselines, transfer, and cost. Limits: Paper-reported only; no search, model call, agent, archive, repair, task, cost, transfer, safety, or promotion result was reproduced, and held-out score is not deployment usefulness.; ext_dynamic_compute_recurrent_transformers_2026: Provides a complexity-controlled negative comparator for variable-depth compute and online halting, including the boundary between difficulty-aligned allocation and extrapolation. Limits: Metadata-first note only; no source figure, local model, controller, quality/calibration study, cost, hardware, or extrapolation result was reproduced.; cca_project: Provides local-lineage negative cases for proxy-to-benchmark divergence, canonical closure authority, retry ceilings, and optimizer/checkpoint/tokenizer/runtime/output trace discipline. Limits: Project-reported training and benchmark artifacts were not replayed by the book; no capability, benchmark, safety, or support-state result is accepted.; moecot_manifest_project: Provides local-lineage separation between internal contract scores and external holdouts, including negative cases where architecture coherence coexisted with weak external task results. Limits: Internal ablations and holdout numbers are project-reported only and do not establish independent reproduction or book benchmark evidence.; beastbrain_project: Provides local-lineage negative cases for simulated loss, gradients, cache hits, latency, placeholder parameter counts, non-updating routers, and dependent/default-success evaluators. Limits: Simulations, project-reported scores, and retained logs are not reproduced model or hardware benchmark results.; bugbrain_project: Provides local-lineage negative cases for declared capacity versus active state, fixed-random model paths, constant objective terms, host/Pi identity drift, status workloads, and stale green readiness reports. Limits: Reproduced host tests do not establish Pi cognition, model learning, benchmark performance, or hardware capability.; corbens_trainer_project: Provides local-lineage benchmark authenticity states, seed identity, content-pin and decontamination limits, quarantined-run inheritance, stale promotion, and checkpoint-completion failures. Limits: Benchmark-labelled retained runs include synthetic placeholders and are not accepted as external or model-quality results.; corbens_best_model_possible_project: Provides the direct local-lineage motivation for metric-origin algebra, representation-capacity gates, retry discipline, and checkpoint/tokenizer/runtime/raw-output/evaluator claim binding. Limits: The retained paths include generated/frozen state, explicit lane binding, proxy or constant metrics, and incomplete runtime causality; no general capability or benchmark result is established.; ext_claw_swe_bench_2026: Supports binding coding-agent outcomes to a fixed model, harness, workspace, patch extraction, evaluator, runtime budget, and cost rather than attributing a harness result to the model alone. Limits: Primary preprint comparator only; no reported task, score, harness, cost, contamination control, or safety result was reproduced here.; ext_muse_unlearning_2025: Supports separating memorization, privacy leakage, retained utility, removal-scale behavior, and sequential sustainability instead of using one forgetting score. Limits: No MUSE corpus, method, 7B model, privacy probe, scale test, or sequential deletion request was reproduced by the local small policy network.; ext_unlearning_benchmarks_weak_2024: Supports treating unlearning benchmark validity, target ambiguity, forget/retain dependence, and benign perturbation robustness as separate evidence burdens. Limits: The position paper raises the interpretation standard but does not validate the local workload or establish influence, privacy, legal erasure, or storage erasure.; ext_openunlearning_2025: Supports versioned unlearning methods, standardized execution interfaces, public checkpoints, diverse evaluations, and meta-evaluation of metric faithfulness. Limits: The framework, methods, checkpoints, evaluations, and meta-evaluations were not run; standardized record shape does not establish semantic validity or erasure.; ext_swe_rebench_v2_2026: Adds a current multilingual natural repository-task construction funnel, full-suite executable oracle, per-instance pathology metadata, and setup/clarity ablations that expose why benchmark membership is not construct validity. Limits: The released tasks and automated labels are not locally validated en masse; post-snapshot filtering reduces but does not eliminate contamination, and no local benchmark score is established.; relational_dimension_compiler: Adds the RODIE evaluation program: minimum relational order, exact role binding, latent topology, interventions, dynamic identity, reversible contraction, object-field coupling, branch separation, compilation reuse, proposal recall, no-regret routing, transfer, and complete lifecycle cost. Limits: RODIE is a proposed suite with no corpus, independent evaluator, benchmark run, result, construct-validity receipt, or evidence transition.; portia_synapse: Benchmark-authenticity case for binding a successor implementation to an exact test manifest, evidence-gated training phases, exact coordinate and route metrics, calibration, memory isolation, and matched component ablations. Limits: The source contains conflicting test counts and incomplete integration and comparative benchmarking; finite loss, stable steps, and passing unit tests do not establish learning or causal repair.; spider_synapse: Negative-result lineage for retaining a plateau, full path denominators, component diagnostics, and causal recovery rather than treating a successor as retroactive proof. Limits: The reported plateau is not locally reproduced and does not identify which component or training choice caused failure.; ext_elizaos_agent_runtime_2026: Supplies a current implementation comparator for deterministic and live runtime scenarios, per-turn and final assertions, trajectory export, and an explicit separation between in-process diagnostics and externally observed provider qualification. Limits: Official software and documentation review only; no scenario, model, provider, observer signature, trajectory, benchmark, cost, failure, or qualification result was executed or reproduced.; assurance_shift_learning: Adds selection-gap diagnosis, the benchmark-saturation versus capability-maturity distinction, and SaturationShiftBench as an equal-total-cost falsification program with attack injections and ablations. Limits: The benchmark is proposed only; no condition, crossover, endpoint, or superiority result has been executed.; adjudicated_persistence: Contributes LocusBench: matched outcomes with different latent defects, per-locus and cross-locus baselines, placement regret, and whole-lifecycle commitment costs. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.; forward_transfer_program_synthesis: Adds causal controls for reusable-knowledge evaluation: frozen rounds, absence and matched placebo conditions, removal and necessity tests, a library-search factorial, exact verifier outcomes, and complete task-lineage denominators. Limits: Experimental blueprint only; no benchmark condition, ablation, transfer effect, or promotion decision was executed.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Ratchet fixture blocks promotion on contaminated or saturated benchmark.
white-box-evidence-interpretability-and-activation-governance.core	white-box-evidence-interpretability-and-activation-governance	Internal-state observations should enter governance only as typed evidence artifacts with lineage, method assumptions, replication status, causal interventions, stability checks, coverage limits, and explicit non-authority; white-box evidence complements but does not replace behavioral and operational evidence.	Design rationale	argument	deterministic_capability_compilation, kernel_english_residual_compiler, qcsa_whitepaper, platonic_world_model, ext_transformer_circuits_2021, ext_monosemanticity_2023, ext_scaling_sparse_autoencoders_2024, ext_circuit_tracing_2025, ext_probe_control_tasks_2019, ext_interpretability_illusion_bert_2021, ext_saebench_2025, ext_sae_benchmark_reliability_2026, ext_elk_report_2021, ext_influence_functions_2017	Source notes available for all 14 assigned sources; 14 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 14 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_elk_report_2021: Frames eliciting latent knowledge as the problem of recovering what a model internally knows when its ordinary answer channel may be untrusted. Limits: ELK is a research problem and strategy space, not a solved truthful-reporting mechanism or a guarantee that latent knowledge is represented accessibly.; ext_influence_functions_2017: Provides a first-order method for estimating how training points affect a model prediction or loss, useful as one attribution hypothesis. Limits: Influence estimates depend on differentiability, curvature approximation, local linearization, checkpoint identity, and implementation; they do not prove causal responsibility or deletion.; deterministic_capability_compilation: Supports typed evidence artifacts, provenance, deterministic receipts, admission policy, and non-authority boundaries. Limits: Does not validate any interpretability method or mechanistic claim.; kernel_english_residual_compiler: Supports compiling residual observations and structured natural-language interpretations into inspectable, challengeable artifacts. Limits: Structured descriptions are not faithful explanations by construction.; qcsa_whitepaper: Supports modular instrumentation across heterogeneous substrates and interfaces for comparing internal representations. Limits: Architecture proposal does not prove interpretability transfer across substrates.; platonic_world_model: Motivates comparison of representation geometry and shared latent structure across models. Limits: Representational similarity does not establish identical concepts, computations, or causal roles.; ext_transformer_circuits_2021: Provides a framework for describing transformer computation through features and circuits and for forming mechanistic hypotheses. Limits: Framework and case studies do not imply complete, unique, or scalable explanations.; ext_monosemanticity_2023: Shows sparse dictionary learning can recover more interpretable feature directions in a small transformer setting. Limits: Feature labels are imperfect; small-model findings do not establish completeness, causal sufficiency, or frontier-model coverage.; ext_scaling_sparse_autoencoders_2024: Provides evidence and engineering lessons for scaling sparse autoencoders and evaluating learned features on larger models. Limits: Reconstruction and interpretability metrics do not prove faithful decomposition or governance readiness.; ext_circuit_tracing_2025: Demonstrates model-specific circuit-tracing methods and causal graph hypotheses for selected behaviors. Limits: Traced graphs are method- and task-dependent, may omit pathways, and do not constitute complete model understanding.; ext_probe_control_tasks_2019: Supplies matched control tasks and probe selectivity for distinguishing representation evidence from probe memorization capacity. Limits: The ELMo linguistic-probe experiments do not establish a universal control task, causal use of decoded information, model safety, or local reproduction.; ext_interpretability_illusion_bert_2021: Supplies cross-dataset construct challenge and the distinction among global, dataset-level, and local representation structure. Limits: The BERT sentence-embedding case does not show that every feature is illusory, that causal methods fail, or that the result transfers unchanged.; ext_saebench_2025: Supplies a standardized multi-metric SAE comparison spanning reconstruction, concept detection, interpretability, disentanglement, and downstream intervention uses. Limits: Source-reported metric breadth and architecture rankings do not establish that every metric is reliable or that an SAE is semantically or causally faithful.; ext_sae_benchmark_reliability_2026: Supplies a reliability audit based on independent reseeds, training-trajectory discriminability, synthetic ground truth, degraded controls, and an oracle. Limits: The audit is limited to selected metrics and settings and cannot refute SAEs, SAEBench as a whole, interpretability, or model-internal evidence.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Two independently implemented method families on held-out behaviors survive negative controls, causal interventions, checkpoint and transformation tests, predict behavioral change, report unexplained residuals and all failures, and reproduce on a second model family without granting authority.
capability-thresholds-and-deployment-commitments.core	capability-thresholds-and-deployment-commitments	Capability Thresholds and Deployment Commitments owns a domain-, threat-, assessment-, policy-version-, safeguard-package-, release-path-, authority-, exception-, residual-, and time-specific Capability-to-Deployment Commitment: before outcomes are visible, it binds a scoped crossing, non-crossing, incomparable, or stale assessment to predeclared safeguards, verification criteria, deadlines, access and monitoring constraints, re-evaluation, exceptions, residual custody, rollback, disclosure, and release-path consequences; a score, time-horizon estimate, threshold label, crossing, non-crossing, safeguard record, exception, or green readiness handoff alone confers no general capability, safeguard efficacy, safety, readiness, deployment, support, transfer, or SOTA authority.	Design rationale	argument	ext_metr_time_horizons_2025, ext_anthropic_rsp_2026, ext_openai_preparedness_framework_2025, benchmaxxing, theseus_architecture_gate, ext_inspect_ai_2024	Source notes available for all 6 assigned sources; 6 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 6 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_metr_time_horizons_2025: Supports a scoped comparator for a capability measurement envelope with task family, success definition, human baseline, agent scaffold, uncertainty, coverage, and external-validity limits. Limits: Does not establish a local time horizon, general autonomy, dangerous capability, deployment threshold, safeguard sufficiency, safety, or ASI.; ext_anthropic_rsp_2026: Supports a scoped comparator for versioned capability thresholds, threshold-linked safeguards, risk-report records, policy revisions, and explicit change-control obligations. Limits: Does not establish ASI Stack threshold adequacy, policy compliance, safeguard implementation or effectiveness, deployment readiness, safety, or ASI.; ext_openai_preparedness_framework_2025: Supports a scoped comparator for threshold-linked operational commitments, distinct capability and safeguards reports, residual-risk review, and reassessment before deployment decisions. Limits: Does not establish ASI Stack threshold adequacy, policy compliance, safeguard implementation or effectiveness, deployment readiness, safety, or ASI.; benchmaxxing: Supports retaining benchmark lifecycle, baseline, coverage, regression, holdout, and residual records rather than treating an isolated score as a readiness or threshold decision. Limits: Does not establish capability measurement quality, a local threshold, threshold crossing, safeguard effectiveness, safety, or a deployment result.; theseus_architecture_gate: Supports a local implementation-reference comparator for versioned pre-training gates, residual-producing failures, re-run triggers, and the distinction between a green gate and sufficient deployment evidence. Limits: Does not establish a current Theseus gate rerun, local threshold system, safeguard effectiveness, readiness, deployment, safety, or ASI.; ext_inspect_ai_2024: Provides an evaluation-envelope comparator for versioned tasks, datasets, solvers or agents, scorers, tools, sandboxes, transcripts, and logs that a prospective threshold commitment must bind. Limits: Official-doc source note only; no Inspect task, scorer, agent, sandbox, transcript, result, threshold, safeguard, or deployment decision was run, and framework availability cannot set a normative threshold or authorize release.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Measure false clearance/block, stale acceptance, exception misuse, reassessment and safeguard latency, operator effort, and residual age; no local capability, threshold adequacy, safeguard efficacy, readiness, safety, or deployment claim.
adversarial-evaluation-sandbagging-and-training-time-deception.core	adversarial-evaluation-sandbagging-and-training-time-deception	Adversarial Evaluation, Sandbagging, and Training-Time Deception owns a consumer-, decision-, model-, task-, elicitation-, authority-, monitor-, reward-, selection-, evaluator-, hypothesis-, outcome-, lineage-, and time-specific Evaluation Observation Integrity Packet: before outcomes are inspected, it freezes the permitted inference and comparison design, binds every observable and dependency, separates task outcome from behavioral interpretation, preserves discrepancies, alternatives, failures, costs, mitigation descendants, expiry, and downstream invalidation, and routes only a bounded observation-integrity status to existing evidence and decision owners; no score, trace, discrepancy, detector, mitigation, adversarial pass, quarantine, or complete finite packet alone establishes capability, intent, deception, sandbagging prevalence or resistance, reward fidelity, monitor validity, alignment, safety, readiness, deployment, support, transfer, or SOTA.	Design rationale	argument	learning_compute_topology, benchmaxxing, rmi, ext_alignment_faking_2024, ext_ai_sandbagging_2024, ext_emergent_misalignment_reward_hacking_2025, ext_inspect_ai_2024, ext_agentdojo_2024, ext_owasp_agentic_top_10_2026, ext_faithfulness_information_flow_2026, ext_monitorbench_2026, assurance_shift_learning	Source notes available for all 12 assigned sources; 11 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 12 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	benchmaxxing: Supports benchmark lifecycle, held-out, mutation, contamination, regression, residual, and anti-Goodhart vocabulary for an observation-integrity packet and its applicability boundary. Limits: Conceptual source only; it does not establish context effects, sandbagging, deception, evaluator independence, mitigation efficacy, capability, safety, or deployment readiness.; rmi: Supports regression-floor, residual-escrow, routed-attempt, benchmark-pressure, and lifecycle vocabulary for retaining discrepancies and failed evaluations rather than deleting them. Limits: Conceptual architecture source only; no model, router, evaluator, sandbagging probe, deception result, mitigation, or safety result was reproduced.; ext_alignment_faking_2024: Supports disclosed-condition alignment-faking threat-model context and the need to retain selection context. Limits: Does not show that any ASI Stack model or deployment is deceptive.; ext_ai_sandbagging_2024: Supports strategic-underperformance and evaluation-context threat-model context. Limits: Does not establish sandbagging in this repository or validate a detector.; ext_emergent_misalignment_reward_hacking_2025: Supports source-setting reward-hacking and training-time misaligned-generalization threat-model context. Limits: Does not establish a local reward-hacking, monitor-interference, safety, or causal result.; ext_inspect_ai_2024: Provides an evaluation-framework comparator for binding tasks, datasets, solvers, scorers, agents, tools, logs, and sandboxes to the observation context. Limits: Framework availability and a task pass do not establish evaluator validity, independence, adversarial coverage, deception detection, safety, or readiness; no local Inspect run exists.; ext_agentdojo_2024: Provides an adversarial tool-use comparator for realistic tasks, untrusted data, prompt-injection attacks, defenses, outcome criteria, and residual attack coverage. Limits: No local task, attack, defense, score, or trajectory was reproduced, and benchmark results cannot establish complete coverage, deployed robustness, safe authority, or system safety.; ext_owasp_agentic_top_10_2026: Provides a current threat-taxonomy comparator spanning goal hijacking, tool misuse, identity abuse, supply chain, code execution, memory poisoning, inter-agent communication, cascading failure, trust exploitation, and rogue agents. Limits: A risk taxonomy is not proof of completeness, local exposure, control effectiveness, detection, mitigation, safety, or readiness; no local OWASP assessment exists.; ext_faithfulness_information_flow_2026: Supplies trace/action inconsistency controls and the boundary that a transcript may correlate with an answer without being causally necessary, including a reward-hackable code-repair setting. Limits: The study is not evidence that a local model sandbags or deceives, and its diagnostics have low-entropy and reference-model confounds.; ext_monitorbench_2026: Adds a multi-task held-out monitorability benchmark and adversarial stress-test design, complementing causal trace/action inconsistency controls with cross-task monitor degradation evidence. Limits: The reported benchmark relations do not prove causal trace faithfulness, local deception, evaluator quality, safety, or a general capability/monitorability law.; assurance_shift_learning: Adds the evaluator observational ceiling, evaluator-first repair rule, probe-generator ecology, and train-deployment divergence as integrity limits on assurance-shift decisions. Limits: No evaluator is established as adequate or independent, and no deception or sandbagging behavior was measured.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Measure task outcome, context discrepancy, detector error, unsupported intent inference, unsafe promotion, quarantine, cost, and throughput; no local deception, sandbagging resistance, capability, alignment, or safety claim.
safety-cases-and-structured-assurance.core	safety-cases-and-structured-assurance	Safety Cases and Structured Assurance owns a deployment-context-, hazard-, claim-, strategy-, evidence-, assumption-, defeater-, safeguard-, threshold-, readiness-, authority-, release-path-, residual-, version-, and time-specific Assurance Argument Compilation Packet: it compiles exact governed references and bounded support or challenge relations, preserves alternatives, dissent, staleness, countercases, conflicts, overrides, costs, lineage, and downstream invalidation, and routes only a scoped case status to existing decision owners; a connected, rendered, notation-conformant, reviewed, accepted, or synthetically complete case alone establishes neither hazard completeness, evidence adequacy, argument validity, reviewer independence, control effectiveness, risk, safety, readiness, release authority, deployment, support, transfer, nor SOTA.	Design rationale	argument	ext_gsn_community_standard_2011, ext_evaluations_safety_cases_scheming_2024, ext_aisi_safety_cases_2024, benchmaxxing	Source notes available for all 4 assigned sources; 4 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 4 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_gsn_community_standard_2011: Supports a scoped comparator for explicit goals, strategies, evidence references, context, assumptions, justifications, and declared support relations in an assurance graph. Limits: Does not establish that any ASI Stack claim is true, that evidence is adequate, that a threat model is complete, or that a system is safe, ready, or authorized to deploy.; ext_evaluations_safety_cases_scheming_2024: Supports a scoped comparator for threat-model-specific incapability, harm, control, and alignment argument families, evaluation dependencies, assumptions, and open research gaps. Limits: Does not establish local scheming absence, alignment, control effectiveness, safety, readiness, deployment authorization, or ASI.; ext_aisi_safety_cases_2024: Supports a scoped comparator for positive and negative evidence, countercase searches, uncertainty, disagreement, and limits on confidence in frontier-AI safety-case sketches. Limits: Does not establish local countercase adequacy, review independence, safety-case confidence, control efficacy, safety, readiness, or deployment authorization.; benchmaxxing: Supports a local conceptual comparator for preserving benchmark lifecycle, baseline, regression, residual, and anti-Goodhart records that a compiled case may reference without owning their measurement truth conditions. Limits: Does not establish a safety case, hazard model, evidence adequacy, safety, readiness, deployment authority, or ASI.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	A replayed scoped case-compilation workload with support, stale dependency, unresolved-defeater, missing-acceptance, missing-residual, and affected-release controls; no local threat model, safety case, safety, readiness, authority, or deployment claim.
content-authenticity-watermarking-and-synthetic-media-integrity.core	content-authenticity-watermarking-and-synthetic-media-integrity	Synthetic-media integrity should use a layered authenticity envelope that binds asset identity, generator and editor claims, signed provenance, content bindings, watermark or fingerprint signals, detector outputs, visible disclosure, transformation history, trust policy, uncertainty, and remedy; every signal retains its own semantics, and no missing or valid signal becomes a universal truth judgment.	Design rationale	argument	ext_c2pa_specification_2_3_2025, ext_eu_article_50_transparency_guidelines_2026, ext_international_ai_safety_report_2026, deterministic_capability_compilation	Source notes available for all 4 assigned sources; 4 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 4 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_c2pa_specification_2_3_2025: Defines signed manifests, assertions, ingredients, content bindings, validation, and provenance-history interoperability. Limits: Technical standard; no semantic truth, universal retention, or unforgeability claim.; ext_eu_article_50_transparency_guidelines_2026: Provides current official guidance on machine-readable marking, deepfake and public-interest disclosure, provider/deployer roles, and the 2 August 2026 applicability date. Limits: Official guidance, not legal advice, project-specific compliance proof, or robustness evidence.; ext_international_ai_safety_report_2026: Synthesizes synthetic-media, influence, detection, disclosure, education, correction, and resilience measures with substantial uncertainty. Limits: International synthesis; no detector, watermark, provenance system, or resilience intervention is locally reproduced.; deterministic_capability_compilation: Supplies Corben’s capability-compilation lineage for exact artifact identity, transformation provenance, declared semantics, verification boundaries, and residual-preserving handoffs. Limits: Author-side architecture lineage only; it does not establish C2PA conformance, watermark robustness, detector accuracy, content truth, consent, legal compliance, or public trust.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Independent cross-platform tests improve provenance continuity, calibrated detection, comprehension, and remedy under realistic edits without unacceptable privacy, accessibility, or false-attribution harm; no universal authenticity or truth oracle claim.
governed-operations-incident-command-and-graceful-degradation.core	governed-operations-incident-command-and-graceful-degradation	Governed operation is a closed incident lifecycle that binds detection, classification, command authority, containment, effect-complete rollback, graceful degradation, recovery evidence, and learning to the exact deployed system and its dependency graph.	Design rationale	argument	scf, deterministic_capability_compilation, theseus_operator_os, viea, talos, platonic_world_model, ext_nist_ai_rmf_1_0_2023, ext_nist_deployed_ai_monitoring_2026, ext_nist_incident_response_2025, regret_engine, assurance_shift_learning, adjudicated_persistence	Source notes available for all 12 assigned sources; 11 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 12 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	scf: Supports accountable roles, change control, review gates, escalation, and lifecycle governance. Limits: Governance design does not prove detection quality or recovery success.; deterministic_capability_compilation: Supports immutable release identity, authority-scoped artifacts, deterministic receipts, revocation, and rollback lineage. Limits: Compilation controls cannot reverse external effects or guarantee runtime resilience.; theseus_operator_os: Supports an operator-facing control plane for status, intervention, degraded modes, and incident work. Limits: Design lineage is not evidence of usability or incident performance.; viea: Supports explicit commands, roles, verification, failure behavior, artifacts, and escalation paths. Limits: Does not establish incident detection, recovery, or effect reversal.; talos: Supports bounded operator authority, audit, emergency intervention, and policy-controlled execution. Limits: Architectural proposal does not validate incident drills or graceful degradation.; platonic_world_model: Supports explicit environment state, residuals, and model-versus-reality distinctions useful for incident diagnosis. Limits: Does not supply an incident-management standard or recovery evidence.; ext_nist_ai_rmf_1_0_2023: Provides lifecycle AI risk-management functions and governance context spanning map, measure, manage, and govern activities. Limits: Voluntary framework guidance does not prove that a particular operational implementation is effective or compliant.; ext_nist_deployed_ai_monitoring_2026: Provides current NIST guidance on monitoring deployed AI systems, including performance, risk, change, and operational context. Limits: Guidance and examples do not establish universal metrics, thresholds, or incident outcomes.; ext_nist_incident_response_2025: Grounds incident response in preparation integrated with risk management plus detection, response, recovery, and continuous improvement. Limits: Cybersecurity incident guidance requires explicit adaptation and validation for autonomous AI and irreversible real-world effects.; assurance_shift_learning: Adds fast containment versus slow consolidation, recovery effectiveness as a coequal metric, and a prohibition on direct incident-to-gradient transitions. Limits: No deployed detection, containment, recovery, consolidation, or recurrence result was observed.; adjudicated_persistence: Adds a governed incident-learning handoff that preserves fast containment while adjudicating whether and where lessons may become durable changes. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Adversarial compound and detector-degrading game days detect partial rollback across model, optimizer, scheduler, RNG, cache, memory, data, credentials, replicas, descendants, and external effects; recovery beats strong SRE baselines on bounded harm and useful throughput, reproduces independently, and does not retain widened authority.
adjudicated-persistence-and-the-adaptive-commit-boundary.core	adjudicated-persistence-and-the-adaptive-commit-boundary	Every transition from experience to durable causal influence should cross an Adaptive Commit Boundary as an authority-bearing adaptation transaction that keeps the experience record, lesson hypothesis, persistence disposition, concrete realization, qualification lease, and authority grant distinct; selects the least-commitment admissible locus portfolio under evidence, authority, observability, recovery, cost, and descendant obligations; and preserves denial, uncertainty, deoptimization, invalidation, and revocation paths.	Design rationale	argument	adjudicated_persistence, assurance_shift_learning, regret_engine, cognitive_loop_closure, deterministic_capability_compilation, platonic_world_model, viea, talos, ext_bourtoule_machine_unlearning_2021, forward_transfer_program_synthesis	Source notes available for all 10 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 10 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	adjudicated_persistence: Defines the Adaptive Commit Boundary, six-object transaction, cross-surface placement problem, commitment profiles, evidence matching, minimum sufficient persistence, guarded compilation, qualification leases, descendant closure, and LocusBench proposal. Limits: Conceptual author paper and experimental specification; no implementation, benchmark result, validated placement advantage, independently checked proof, or support transition.; assurance_shift_learning: Supplies learning-eligibility, evidence-bundle, evaluator-first, least-invasive repair, and fast-versus-slow adaptation distinctions that precede a persistence commitment. Limits: Conceptual framework; no implemented assurance allocator or empirical placement result.; regret_engine: Supplies typed experience residuals, causal alternatives, and learning-eligibility signals without allowing outcome or blame records to choose a persistence locus automatically. Limits: A regret packet can propose a lesson but cannot establish placement, qualification, authority, or support.; cognitive_loop_closure: Supplies the transition from repeated cognition toward verified procedures and tools as one candidate persistence family. Limits: Procedural compilation is one locus, not evidence that the lesson should persist there.; deterministic_capability_compilation: Supplies immutable input identity, compilation receipts, deterministic artifact lineage, revocation, and rollback concepts for locus-specific realizations. Limits: Deterministic compilation does not establish lesson validity, placement quality, qualification, or authority.; platonic_world_model: Supplies explicit environment-state, model-versus-reality residual, and counterfactual distinctions needed to preserve observability around durable adaptation. Limits: World-model framing does not identify the correct persistence locus or validate causal attribution.; viea: Supplies typed commands, roles, verification, failure behavior, and escalation for authority-bearing adaptation transactions. Limits: Command structure alone does not validate persistence decisions or their outcomes.; talos: Supplies bounded execution authority, audit, intervention, and policy-controlled realization interfaces. Limits: Architecture lineage does not establish placement, qualification, revocation efficacy, or deployed safety.; ext_bourtoule_machine_unlearning_2021: Provides an external comparator in which training-time sharding, isolation, slicing, and aggregation bound the retraining path for deletion requests, while leaving verified removal and downstream artifacts as separate obligations. Limits: The source-reported SISA experiments concern bounded machine-unlearning settings and do not validate cross-surface placement, complete descendant erasure, privacy, organizational revocation, or the proposed Adaptive Commit Boundary.; forward_transfer_program_synthesis: Adds a separate later-version admission step after independent validation and frozen evaluation, preserving valid-but-transfer-negative artifacts without routing them as beneficial defaults. Limits: No admission service, knowledge-version transition, removal test, benchmark result, or support movement was implemented.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	A prospectively frozen LocusBench campaign beats outcome-only and fixed-locus baselines on held-out placement and recovery without unacceptable useful-throughput, latency, compute, human-burden, carrying-cost, guard-bypass, stale-lease, or descendant-escape results; finite Lean checks establish only encoded transaction properties, and no safety or universal placement claim follows.
policy-optimization-and-learning-from-feedback.core	policy-optimization-and-learning-from-feedback	Policy Optimization and Learning from Feedback owns a target-policy-, baseline-, objective-, feedback-, evaluator-, dataset-, optimizer-, checkpoint-, rollout-, authority-, resource-, monitor-, rollback-, consumer-, environment-, and time-specific Governed Policy Update Lease: before any update, it freezes the legitimate target behavior, admissible feedback and proxy boundary, strong baselines, update family and budget, drift and authority ceilings, complete evaluation and failure denominators, reward-hacking and causal checks, rollback and monitoring, residuals, expiry, and promotion authority; a reward, preference, verifier score, benchmark gain, loss reduction, synthetic canary, formal route, rollback dry run, or trained checkpoint alone establishes neither reward validity, causal policy improvement, retained capability, alignment, safety, readiness, deployment, support, transfer, nor SOTA.	Design rationale	argument	learning_compute_topology, benchmaxxing, rmi, spinoza, verification_bandwidth, talos, vcm_public, planforge, octopus_router, scf, tokenmana, cognitive_loop_closure, moecot, ext_trpo_2015, ext_ppo_2017, ext_remax_2023, ext_dpo_2023, ext_ipo_preference_2023, ext_orpo_2024, ext_kto_2024, ext_simpo_2024, ext_reinforce_style_rlhf_2024, ext_deepseek_r1_2025, ext_dapo_2025, ext_gspo_2025, ext_s_grpo_2025, ext_longrlvr_2026, ext_rlhf_limitations_2023, ext_muse_unlearning_2025, ext_faithfulness_information_flow_2026, portia_synapse, spider_synapse, regret_engine, ext_eggroll_hyperscale_es_2026, ext_openai_es_2017, assurance_shift_learning, adjudicated_persistence	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 37 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_eggroll_hyperscale_es_2026: Provides a batched low-rank evolution-strategy case for optimizing outcome-level, discrete, and nondifferentiable fitness without action-level gradients. Limits: Outcome-only fitness remains gameable and source-reported reasoning/control results do not validate local policy improvement, evaluator quality, safety, or total efficiency.; ext_openai_es_2017: Provides a foundational outcome-only parameter-search comparator to temporal policy-gradient credit assignment. Limits: The studied control tasks and distributed scaling do not establish universal policy quality, sample efficiency, or governance safety.; benchmaxxing: Supports treating feedback as benchmark pressure that must preserve ledgers, baselines, residuals, regressions, holdouts, anti-Goodhart checks, and conservative support-state effects before any policy update is promoted. Limits: Does not prove reward quality, policy improvement, policy-training success, local benchmark success, or model-capability improvement.; rmi: Supports policy optimization as modular ratcheting: arm/router improvement, specialist lifecycle, residual escrow, benchmark diagnosis, regression preservation, and harder-frontier selection rather than unconstrained fine-tuning. Limits: Conceptual architecture only; no policy-training run, specialist-arm evaluation, routing benchmark, modular improvement result, or capability result has been reproduced here.; spinoza: Supports keeping reward signals, verifier outputs, preference labels, proof tiers, belief revisions, contradictions, and downgrade behavior distinct before they influence accepted claims or policy updates. Limits: Does not prove verifier quality, preference validity, natural-language formalization, proof-object correctness for arbitrary claims, reward validity, or whole-system epistemic correctness.; verification_bandwidth: Supports bounding context-policy and reasoning-budget rewards by effective verification workspace, semantic-unit comparison, contradiction pressure, summary loss, and adequacy rather than context length or shorter reasoning alone. Limits: The proposed contradiction-rate, constraint-satisfaction, context-adequacy, summary-loss, and reasoning-budget tests have not been run; theoretical claims remain source-proposed until tested or mechanized.; talos: Supports execution-policy feedback through typed jobs, contract locks, evidence records, artifact success, approval gates, audit logs, replay, delivery evidence, and residual feedback. Limits: No Talos training, execution-policy update, runtime execution, replay result, approval-service result, security result, or benchmark artifact has been reproduced here.; vcm_public: Supports context-policy rewards around source binding, adequacy/admission separation, protected compilation, taint, revocation, snapshot coherence, and typed non-success states instead of retrieval volume alone. Limits: VCM evidence remains bounded to control-plane properties; no context-policy training run, end-to-end model-quality result, retrieval-quality result, or reward-quality result is claimed.; planforge: Supports planner-policy update targets through decomposition quality, dependency preservation, primitive schemas, tier routing, scheduling, fallback requirements, failure-driven replanning, and stopping conditions. Limits: No PlanForge scheduler, planner-policy training run, learned decomposer, task-quality benchmark, synthetic savings reproduction, or runtime replanning result has been implemented in this repo.; octopus_router: Supports router-policy update targets through task and risk interpretation, candidate selection, single/parallel/sequential/debate/verification/reflex composition, abstention and fallback, permission and memory envelopes, dynamic residency, arm-local benchmark and residual ratchets, and add/split/merge/retire lifecycle signals. Limits: Conceptual architecture source only; no local routed-specialist prototype, arm-card experiment, router-policy training run, dynamic-loading comparison, quarantine harness, safety result, specialist lifecycle result, or total-cost benchmark was run.; scf: Supports governance of policy-update promotion through stable capability fields, contracts, qualifications, evaluator policy, authority ceilings, lifecycle events, rollback, quarantine, and recovery paths. Limits: Does not prove evaluator integrity, production safety, global alignment, route quality, policy-update safety, learned-behavior bounds, or rollback success.; tokenmana: Supports resource and latency rewards only when bounded capacity, load variance, burst controls, human/cognitive friction, privacy, and quality constraints are recorded alongside efficiency pressure. Limits: No regenerative-budget simulation, pricing result, load-variance reproduction, latency-quality tradeoff, human outcome study, sleep/cognitive study, or quality result has been reproduced here.; cognitive_loop_closure: Supports distinguishing policy learning from procedural compilation: repeated trajectories may become tools only with trajectory logs, loop detection, parameter discovery, verification, routing, monitoring, revision, and retirement records. Limits: No loop-closure system, tool synthesis run, verifier, runtime monitor, tool registry, retirement workflow, or procedural-memory benchmark has been executed.; moecot: Supports runtime-adjacent policy updates for specialist routing and orchestration while requiring readiness gates, regression floors, replay, ledgers, promotion blockers, and residual tracking. Limits: Reviewed source note only because authenticated connector text is not published here; implementation-reference context only, and runtime artifacts, benchmark artifacts, logs, and policy-update results have not been imported or reproduced.; ext_trpo_2015: Supports the external method-family comparison for trust-region bounded policy updates and drift constraints. Limits: External method-family context only; does not support the ASI Stack governance claim by itself and does not report local planner, router, context-policy, or model training.; ext_ppo_2017: Supports the external method-family comparison for PPO-style online policy-gradient updates and proximal surrogate objectives. Limits: External method-family context only; does not validate reward quality, local PPO runs, planner/router/context-policy training, or deployment readiness.; ext_remax_2023: Supports the external method-family comparison for simpler RLHF-oriented policy-gradient updates relative to PPO complexity. Limits: External method-family context only; does not prove local RLHF efficiency, reward quality, stack-policy improvement, or universal PPO replacement.; ext_dpo_2023: Supports the external method-family comparison for offline preference optimization from pairwise preference data. Limits: External method-family context only; does not validate local preference data, DPO training, preference quality, reward correctness, or governance-safe behavior.; ext_ipo_preference_2023: Supports external framing of preference-learning assumptions, including pairwise preferences, pointwise reward approximations, and policy-distribution shift. Limits: External theoretical context only; no local preference-learning theorem, reward-model validation, preference experiment, or ASI Stack policy update is recorded.; ext_orpo_2024: Supports the external method-family comparison for reference-model-free monolithic preference optimization. Limits: External method-family context only; does not prove local ORPO training, alignment improvement, truthfulness, tool safety, or governance compliance.; ext_kto_2024: Supports the external method-family comparison for prospect-theoretic or human-aware loss objectives and binary feedback framing. Limits: External method-family context only; does not prove a correct human model, reward validity, preference-data validity, local policy safety, or governance compliance.; ext_simpo_2024: Supports the external method-family comparison for simple reference-free sequence-level preference optimization. Limits: External method-family context only; does not prove local SimPO performance, preference-quality robustness, truthfulness, tool safety, or governed deployment readiness.; ext_reinforce_style_rlhf_2024: Supports the external method-family comparison for simpler REINFORCE-style RLHF baselines and PPO-cost/tuning cautions. Limits: External method-family context only; does not show that the repo has implemented REINFORCE, RLOO, RLHF training, reward modeling, or policy-safety evaluation.; ext_deepseek_r1_2025: Supports external comparison for reasoning-RL pressure and the distinction between reasoning behavior, reward design, and benchmark reporting. Limits: External method-family context only; does not import or reproduce DeepSeek-R1 models, training artifacts, evaluations, benchmark results, model behavior, or governance claims.; ext_dapo_2025: Supports external comparison for open-source reasoning-RL systems, decoupled clipping, dynamic sampling, and reproducibility pressure. Limits: External method-family context only; does not mean DAPO has been run, audited, imported, or reproduced in this repository.; ext_gspo_2025: Supports external comparison for sequence-level group policy optimization and sequence-level clipping/reward design. Limits: External method-family context only; does not validate local sequence-level RL, reasoning-budget policy, model-quality improvement, or reward robustness.; ext_s_grpo_2025: Supports external comparison for early-exit/reasoning-budget RL and overthinking control. Limits: External method-family context only; does not prove shorter reasoning is better, local reasoning-budget policies exist, or verification adequacy is preserved under early exit.; ext_longrlvr_2026: Supports external comparison for long-context RLVR and verifiable context-grounding rewards relevant to VCM policy learning. Limits: External method-family context only; does not reproduce LongRLVR, context-grounding reward experiments, VCM context-policy learning, or reward-quality evidence.; ext_rlhf_limitations_2023: Supports external risk framing for RLHF limitations, reward hacking, evaluator limits, and complementary safeguards. Limits: External survey/risk framing only; not proof of any local mitigation, reward-process safety, evaluator quality, or deployed safeguard.; ext_muse_unlearning_2025: Supports separating memorization, privacy leakage, retained utility, removal-scale behavior, and sequential sustainability instead of using one forgetting score. Limits: No MUSE corpus, method, 7B model, privacy probe, scale test, or sequential deletion request was reproduced by the local small policy network.; ext_faithfulness_information_flow_2026: Shows that reward improvement can coexist with shortcut use and under-verbalized reward hacking, and tests update-time interventions that change information flow without changing rollout rewards. Limits: The intervention results are bounded to the reported arithmetic, code-repair, and DAPO-Math settings; they do not validate local policy optimization or eliminate reward hacking.; portia_synapse: Case study for evidence-gated phase training, smallest-contract-first optimization, gradient and activation diagnostics, confidence calibration, component admission, matched ablation, and rollback. Limits: The source reports an implementation but does not provide local training artifacts, independent reproduction, exact task learning, causal repair evidence, or deployment evidence.; spider_synapse: Preserved negative result and recovery case for reducing branch count and refinement depth, isolating a coordinate-only target, logging per-path evidence, and reintroducing mechanisms one at a time. Limits: The proposed one-twelfth gradient explanation is a hypothesis, and the report does not causally identify branching, memory, selection, label smoothing, or target geometry.; assurance_shift_learning: Adds the positive-support prerequisite, outcome/process separation, evaluator-first rule, and governed negative-pressure boundary before an informative exception can enter an update lease. Limits: No policy update, anti-tampering effect, or superiority of negative feedback was demonstrated.; adjudicated_persistence: Frames a policy update as one candidate persistence locus selected only after learning eligibility, cross-surface comparison, commitment matching, and explicit authority. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.	independent validity assessment; production-scale update and monitoring traces; replicated target gain at the retained-task bound; separate evidence for feedback learning, open-endedness, or recursion if claimed	A feedback-driven policy update improves held-out utility without reward laundering or hidden forgetting and survives independent evaluation, canary, rollback, and residual review.
data-engines-continual-learning-and-unlearning.core	data-engines-continual-learning-and-unlearning	Data Engines, Continual Learning, and Unlearning owns a datum-, cohort-, provenance-, rights-, split-, contamination-, learning-lane-, retention-, checkpoint-authority-, full-state-inventory-, descendant-, deletion-request-, claim-axis-, consumer-, environment-, and time-specific Data-and-Descendant Custody Lease: before learning or deletion, it binds admissible use, evaluation exclusions, synthetic and transformation lineage, coverage and distribution residuals, model/optimizer/scheduler/RNG/cache/backup/descendant state, prospective checkpoint authority, retention and replay, deletion propagation, verification, rollback, expiry, and terminal custody; behavioral cohort change, causal influence reduction, privacy leakage reduction, lineage invalidation, legal compliance, and storage or backup erasure remain separate claims, and no receipt, checksum, exclusion, invalidation, benchmark score, rollback match, or synthetic campaign alone establishes model quality, forgetting, privacy, erasure, safety, readiness, deployment, support, transfer, or SOTA.	Design rationale	argument	learning_compute_topology, ext_bourtoule_machine_unlearning_2021, deterministic_capability_compilation, platonic_world_model, ext_shumailov_model_collapse_2023, ext_gerstgrasser_data_accumulation_2024, theseus_synthetic_data_curation, ext_w3c_prov_o_2013, ext_mlcommons_croissant_1_1_2026, ext_muse_unlearning_2025, ext_unlearning_benchmarks_weak_2024, ext_openunlearning_2025, qcsa_whitepaper, ext_v_jepa_2_2025, ext_influence_functions_2017, ext_curriculum_learning_2009, regret_engine, assurance_shift_learning, adjudicated_persistence	Current adjacent local evidence is recorded in the affected chapter and accepted no-change core disposition; core support remains argument. Result: experiments/post_v2_1_evidence_program/results/2026-07-11-post-v2-1-outcomes.json.	All 19 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_influence_functions_2017: Provides an approximate training-point influence estimator that can prioritize candidate descendants and deletion impact checks. Limits: Approximate influence is not proof of causal contribution, privacy leakage, legal erasure, storage erasure, or successful unlearning.; ext_curriculum_learning_2009: Shows that data order and difficulty schedules can be part of the learning mechanism rather than incidental loader state. Limits: The source does not establish a universal curriculum, safe continual learning, or transfer to current foundation-model pipelines.; ext_bourtoule_machine_unlearning_2021: Supports treating deletion as a lifecycle with training-time structure, affected-artifact scope, retraining cost, accuracy trade-offs, and residual boundaries rather than as a bare forget command. Limits: Does not prove deletion from this repository’s models, checkpoints, adapters, caches, retrieval stores, distilled artifacts, or published outputs; it does not establish privacy compliance or generic unlearning.; ext_shumailov_model_collapse_2023: Supports treating recursively generated data as provenance-bearing input with coverage and tail-loss residuals, rather than assuming every synthetic-data loop is harmless or every synthetic sample is unsafe. Limits: The source’s theoretical and experimental settings do not establish model collapse in this repository or a universal rule for mixed real and synthetic training pipelines.; ext_gerstgrasser_data_accumulation_2024: Supports making replacement, accumulation, retention, and deletion explicit competing policies in a synthetic-data loop rather than collapsing them into one generic anti-collapse rule. Limits: Its source-setting analytical and experimental results do not prove that accumulation is safe, optimal, private, or effective for this stack or for modern foundation-model training generally.; theseus_synthetic_data_curation: Supports a source-reported implementation pattern for residual-targeted synthetic-data curation with provenance, exact-overlap exclusions, quality and diversity gates, ratio caps, governed teacher handling, and blocked promotion paths. Limits: This is pinned implementation-reference context only. The ASI Stack repository has not run its curator, inspected generated rows, replayed digests, imported a dataset, or measured data or model quality.; ext_w3c_prov_o_2013: Provides an interoperable provenance-vocabulary comparator for entities, activities, agents, derivation, attribution, delegation, revision, and invalidation in data and descendant lineage. Limits: An asserted PROV-O graph does not establish lineage truth, completeness, integrity, authority, lawful use, deletion closure, or safety; no local graph or conformance result exists.; ext_mlcommons_croissant_1_1_2026: Provides a dataset-metadata comparator for JSON-LD structure, resources, checksums, record fields, provenance, usage conditions, and ML-tool portability. Limits: Metadata conformance cannot establish integrity, representativeness, fitness, lawful or consented use, contamination freedom, privacy, safe learning, deletion closure, or model quality; no local Croissant validation exists.; ext_muse_unlearning_2025: Supports separating memorization, privacy leakage, retained utility, removal-scale behavior, and sequential sustainability instead of using one forgetting score. Limits: No MUSE corpus, method, 7B model, privacy probe, scale test, or sequential deletion request was reproduced by the local small policy network.; ext_unlearning_benchmarks_weak_2024: Supports treating unlearning benchmark validity, target ambiguity, forget/retain dependence, and benign perturbation robustness as separate evidence burdens. Limits: The position paper raises the interpretation standard but does not validate the local workload or establish influence, privacy, legal erasure, or storage erasure.; ext_openunlearning_2025: Supports versioned unlearning methods, standardized execution interfaces, public checkpoints, diverse evaluations, and meta-evaluation of metric faithfulness. Limits: The framework, methods, checkpoints, evaluations, and meta-evaluations were not run; standardized record shape does not establish semantic validity or erasure.; qcsa_whitepaper: Contributes fast/medium/slow update timescales, candidate versus authoritative atlas epochs, identity-preserving readdressing, merge/split lineage, migration compatibility, shadow evaluation, rollback, and semantic-drift tests; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. No learning, forgetting, influence, privacy/storage erasure, chapter-core promotion, AGI, or ASI result is established.; ext_v_jepa_2_2025: Shows why world-model data lineage must separate action-free pretraining, action-conditioned adaptation, predictor versions, and error/residual records. Limits: The reported video and robot results do not establish local model quality, causal understanding, safe control, transfer, deployment, or an ASI Stack result.; assurance_shift_learning: Adds informative-exception strata, learner-relative negative half-life, natural/probe distribution separation, protected positives, and repair-aware replay custody. Limits: Inhibition, replay retirement, and retained negative knowledge do not establish parameter unlearning, deletion, privacy, or forgetting.; adjudicated_persistence: Frames datasets, replay buffers, labels, weights, memories, and deletion obligations as distinct persistence surfaces with different descendant and unlearning costs. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.	causal influence-removal evidence; descendant-wide propagation receipts; independent validity assessment; privacy and member/nonmember assessment; production-scale update and monitoring traces; replicated target gain at the retained-task bound; separate evidence for feedback learning, open-endedness, or recursion if claimed; verified storage and backup erasure where claimed	A real governed data pipeline traces deletion and retained utility through every affected descendant and storage surface with contamination, privacy, rollback, and independent verification.
scientific-discovery-and-experimental-governance.core	scientific-discovery-and-experimental-governance	An AI-generated scientific claim should enter the evidence stack only through a preregistered experimental contract that binds hypothesis lineage, exploratory versus confirmatory status, design and power, instrument or simulator authority, calibration, sample and protocol lineage, blinding and holdouts, stopping and exclusions, analysis, complete attempts, independent replication, dual-use disposition, and claim ceiling; experimental completion, significance, synthesis, instrument output, or formal workflow validity alone establishes neither causal truth, general scientific discovery, reproducibility, safety, nor transfer.	Design rationale	argument	aletheia, ext_autonomous_lab_materials_2023, ext_ai_scientist_end_to_end_2026, ext_coscientist_chemistry_2023, ext_ai_co_scientist_2025	Source notes available for all 5 assigned sources; 4 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 5 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	ext_autonomous_lab_materials_2023: Preliminary autonomous-laboratory comparator based on the corrected official Nature article abstract, selected article-page passages, and the 2026 author correction: A-Lab integrates computation, literature-derived data, machine learning, active learning, and robotics, with the corrected article reporting 36 realized compounds from 57 targets. The correction narrows the novelty wording and excludes four inconclusive identifications; no laboratory run, material synthesis, replication, or general experimental-control-plane result has been reproduced locally. Limits: Source-reported or source-defined material only; no local chapter-core implementation, reproduction, transfer, readiness, release, deployment, or support promotion.; ext_ai_scientist_end_to_end_2026: Passage-reviewed computational-research comparator: the reported system connects ideation, literature search, code, experiments, analysis, manuscript production, and automated review. The paper reports one of three generated workshop submissions receiving a 6.33 mean score at a workshop with a 70 percent acceptance rate, while the authors judged none conference-ready and documented implementation, rigor, figure, and citation failures. Limits: Source-reported machine-learning workflow only; paper completion, automated review, and venue scoring are not scientific truth. No system run, manuscript, hidden search tree, review, reproduction, transfer, or support result is local.; ext_coscientist_chemistry_2023: Passage-reviewed physical-laboratory comparator: Coscientist connects a language-model planner to web and document search, code execution, and robotic laboratory APIs across six reported chemistry task families, making proposal, tool, intervention, measurement, and adjudication boundaries concrete. Limits: Source-reported bounded chemistry demonstrations with human oversight and task-specific assessment; no local chemistry run, independent scientist, general laboratory safety, causal discovery, transfer, or support promotion.; ext_ai_co_scientist_2025: Passage-bounded hypothesis-generation comparator: the proposed multi-agent system generates, reflects on, ranks, evolves, and meta-reviews hypotheses with an Elo-style process and additional inference compute, followed by source-reported expert and selected laboratory appraisal. Limits: Preprint and provider-reported evaluation only; internal rank, expert preference, and laboratory confirmation remain separate evidence objects. No full search population, evaluator independence, experiment, reproduction, transfer, or support result is local.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Injected null and known effects are distinguished with complete attempts and independent replication; no general autonomous-science or cross-domain discovery claim.
artifact-steward-agents-and-living-project-governance.core	artifact-steward-agents-and-living-project-governance	Artifact Steward Agents and Living Project Governance owns a project-, artifact-, mission-, owner-, authority-, roadmap-, work-contract-, event-, treasury-, compute-, contributor-, evidence-, governance-, release-, federation-, sunset-, consumer-, environment-, and time-specific Artifact Steward Continuity Lease: it may observe, propose, prepare, coordinate, execute, reverse, archive, or retire only through a versioned charter, taint-aware intake, scoped work contracts, separated contribution ledgers, bounded treasury and compute policy, verification and release gates, appeal and fork or exit paths, effect-complete rollback, and explicit sunset authority; it never acquires ownership, governance legitimacy, evidence authority, funding rights, release authority, legal standing, or permission from useful motion, a green workflow, a vote, a balance, a score, a fixture, a theorem, or its own prior action.	Design rationale	argument	viea, talos, planforge, vcm_public, spinoza, benchmaxxing, rmi, cognitive_loop_closure, tokenmana, coherence_exchange, project_theseus_whitepaper, theseus_operator_os, scf, field_of_god_ai_constitution, attd, ext_akash_docs_2026, ext_golem_docs_2025, ext_github_webhooks_docs, ext_github_self_hosted_runners_docs, ext_openzeppelin_governor_docs, ext_open_collective_docs, ext_github_sponsors_docs, ext_agentic_workflow_injection_2026, ext_dao_delegation_fairness_2025	Source notes available for all 24 assigned sources; 23 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 24 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports artifact discipline, claim support states, residuals, regression coverage, intent-to-execution boundaries, and the rule that important responses become durable artifacts. Limits: Does not prove a deployed project steward exists or that any autonomous management behavior is safe.; talos: Supports typed project work, contract locks, isolation, audit, replay, evidence, delivery, approval gates, and residual feedback. Limits: Does not authorize free-form agent management or prove a steward bot implementation.; planforge: Supports roadmap decomposition, dependencies, scheduling, minimum viable intelligence tiering, fallback, and failure-driven replanning. Limits: Does not make a plan equivalent to project authority or governance approval.; vcm_public: Supports project memory with source binding, taint, revocation, context adequacy, materializations, and typed non-success states. Limits: Does not solve long-term project memory quality or steward reliability.; spinoza: Supports proof-carrying claim discipline, belief revision, failed-verification downgrade, contradiction handling, and protected axioms. Limits: Does not prove open-domain formalization, verifier quality, or governance correctness.; benchmaxxing: Supports benchmark lifecycle, residual diagnosis, regression preservation, anti-Goodhart checks, and cautious architecture-change discipline. Limits: Does not provide local benchmark or release-readiness results for stewarded projects.; rmi: Supports residual escrow, modular improvement loops, specialist lifecycle, readiness gates, and regression-floor preservation. Limits: Does not prove autonomous project improvement or capture resistance.; cognitive_loop_closure: Supports turning repeated project workflows into verified tools with preconditions, postconditions, monitoring, revision, and retirement. Limits: Does not justify automating every recurring task or replacing human review.; tokenmana: Supports bounded resource accounting for compute, money, reviewer time, load variance, burst controls, and cognitive friction. Limits: Does not prove treasury performance, funding design, or optimal spend policy.; coherence_exchange: Supports fork, exit, audit, contestability, value/accounting, review-market, and governance-interface framing. Limits: Speculative where economic or epistemic-liquidity language exceeds executable records.; project_theseus_whitepaper: Supports local-first report discipline, trusted-node task allowlists, residual ledgers, and implementation-reference governance surfaces. Limits: No current project reports or command outputs were inspected or rerun as evidence here.; theseus_operator_os: Supports durable work boards, operator channels, node registries, background/watch/wake contracts, feedback routing, TTLs, kill switches, and safety-visible operator surfaces. Limits: Does not prove unattended project operation is safe.; scf: Supports capability leases, qualification, lifecycle, route validation, governance controls, authority ceilings, and no procedural self-ratification. Limits: Does not qualify an artifact steward capability or validate governance enforcement.; field_of_god_ai_constitution: Supports consent, non-domination, least sufficient power, memory/tool governance, reversibility, auditability, and caution around self-authorization. Limits: Does not implement constitutional enforcement or prove institutional capture resistance.; ext_akash_docs_2026: Supports rented decentralized-compute vocabulary for steward budgets, provider resources, leases, and deployment operations. Limits: Does not prove a steward should rent compute or that provider isolation, cost, legality, or result quality are acceptable.; ext_golem_docs_2025: Supports decentralized task execution, provider selection, and resource-sharing vocabulary for public project hives. Limits: Does not prove sandbox safety, evidence validity, payment fairness, or dispute resolution for stewarded work.; ext_github_webhooks_docs: Supports event-driven repository automation vocabulary for issues, pull requests, workflow events, delivery headers, and typed payload intake. Limits: Does not make webhook content trusted control text or prove a steward bot is safe.; ext_github_self_hosted_runners_docs: Supports self-hosted runner vocabulary for project-owned or maintainer-assigned compute execution. Limits: Does not prove self-hosted runners are safe for untrusted tasks without isolation, cleanup, credential, and review gates.; ext_openzeppelin_governor_docs: Supports proposal, vote, quorum, timelock, settings, and guardian-control vocabulary for steward governance. Limits: Does not solve project legitimacy, legal authority, capture resistance, or safe treasury autonomy.; ext_open_collective_docs: Supports transparent community money management, fiscal-hosting vocabulary, contribution intake, expense review, accounting, and the separation between a project and the legal entity that holds funds. Limits: Does not prove autonomous treasury safety, legal compliance, project sustainability, or that an AI steward may spend without human, host, maintainer, or governance review.; ext_github_sponsors_docs: Supports GitHub-native sponsorship surfaces for eligible open-source contributors and organizations, including non-code contribution categories such as documentation, triage, leadership, project management, mentorship, and design. Limits: Does not prove sponsorship incentives are governance-safe or that funding signals should control roadmap, evidence, release, or treasury decisions.; ext_agentic_workflow_injection_2026: Supports the steward chapter warning that untrusted issue, PR, comment, and workflow event context can become agentic workflow injection material. Limits: Does not prove this repository is vulnerable or safe; no workflow-injection fixture or scan has been run.; ext_dao_delegation_fairness_2025: Supports anti-capture concerns around voter apathy, voting-power concentration, delegation misalignment, and ranking bias. Limits: Does not prove the proposed separated ledgers prevent governance capture.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Steward action fixture validates governance, treasury, release, and sunset boundaries.
integrated-reference-architecture.core	integrated-reference-architecture	Integrated Reference Architecture owns a trace-, run-, request-, intent-, authority-, artifact-, parentage-, layer-, canonical-state-, material-effect-, terminal-receipt-, evaluator-, evidence-, residual-, rollback-, consumer-, environment-, and time-specific Cross-Layer Trace Join Contract: it proves integration only when every participating owner remains distinct and its typed input, output, authority delta, state identity, observed effect, acknowledgement, evaluation, evidence delta, residual, stop, repair, rollback, and non-claim remain joinable across approved, blocked, failed, revoked, rolled-back, and quarantined paths; a diagram, shared prompt, interface name, projection, green service, replay, fixture, theorem, or locally successful slice cannot establish whole-stack execution, semantic preservation, governance enforcement, safety, capability, deployment, transfer, or SOTA.	Design rationale	argument	learning_compute_topology, viea, deterministic_capability_compilation, platonic_world_model, scf, vcm_public, planforge, talos, spinoza, octopus_router, rmi, benchmaxxing, alignment_field, cgs, cognitive_loop_closure, moecot, vcm_editable, project_theseus_whitepaper, theseus_plan_compiler, theseus_self_evolution_system, theseus_architecture_gate, theseus_operator_os, theseus_circle_transfer, cca_project, moecot_manifest_project, bugbrain_project, corbens_trainer_project, corbens_best_model_possible_project, ext_drexler_cais_2019, ext_adas_2024, qcsa_whitepaper, ext_v_jepa_2_2025, ext_embedded_agency_2019, reflexive_router_whitepaper, kernel_english_residual_compiler, ext_swe_rebench_v2_2026, relational_dimension_compiler, regret_engine, assurance_shift_learning, adjudicated_persistence	Source notes available for all 40 assigned sources; 35 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 40 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports the integrated reference architecture as an intent-to-execution spine: command contracts, artifacts, claim ledgers, routing, runtime adapters, feedback, residuals, and regression coverage. Limits: Architecture proposal only here; no end-to-end deployed VIEA runtime, execution log, field-feedback record, workflow compiler, or benchmark has been reproduced in this repository.; scf: Supports the self-improvement and replacement portions of the integrated stack through stable capability identities, evaluator policy, authority ceilings, qualification claims, lifecycle events, and recovery paths. Limits: Does not prove production safety, evaluator integrity, global alignment, reversible deployment behavior, route quality, or rollback success.; vcm_public: Supports the memory/context handoff layer through semantic objects, immutable versions, source and authority binding, adequacy/admission separation, protected compilation, taint, revocation, and audit records. Limits: Does not establish end-to-end model accuracy improvement, broad VCM superiority, context-policy learning, retrieval-quality improvement, or model-facing benefit.; planforge: Supports the planning handoff through hierarchical decomposition, primitive schemas, dependency inference, minimum viable intelligence tiering, scheduling, feedback loops, and failure-driven replanning. Limits: Whitepaper/design support only; no PlanForge scheduler, DAG compiler, learned decomposer, runtime replanning flow, synthetic savings reproduction, or planning benchmark has been validated here.; talos: Supports the execution layer with typed jobs, contract locks, deterministic control planes, evidence records, isolation, audit, replay, delivery, and residual feedback. Limits: Does not prove Talos benchmark, security, runtime execution, approval-service behavior, replay, or production execution claims in this repository.; spinoza: Supports the evidence and belief-update handoff through proposer/verifier separation, proof/citation/procedure-carrying claims, contradiction handling, dependency tracing, downgrade paths, and protected axioms. Limits: Does not solve open-domain formalization, verifier quality, proof-object correctness for arbitrary claims, reward validity, or whole-system epistemic correctness.; octopus_router: Supports the integrated routing handoff through a resident head, bounded arm contracts, versioned registry cards, task-local permission and memory envelopes, semantic selection separated from dynamic residency, typed composition modes, structured result and verification fields, quarantine, and evidence-bound arm lifecycle management. Limits: Conceptual architecture source only; no local routed-specialist prototype, arm-card implementation, router-policy training run, dynamic residency comparison, composition-faithfulness test, quarantine harness, routing benchmark, lifecycle result, or safety result has been run.; rmi: Supports the integrated improvement loop through benchmark pressure, routed specialist attempts, residual escrow, loop closure, regression preservation, arm/router lifecycle, and harder frontier selection. Limits: Conceptual architecture source only here; no benchmark run, prototype inspection, specialist-arm evaluation, router benchmark, or capability result is recorded.; benchmaxxing: Supports the evidence-update and improvement-gate portions through benchmark lifecycle states, wall diagnosis, benchmark/model ledgers, regression preservation, anti-Goodhart safeguards, holdouts, and transfer checks. Limits: No benchmark harness, mutation test, holdout run, transfer check, contamination audit, live benchmark, or empirical result has been run in this repository.; alignment_field: Supports the normative boundary of the integrated architecture through dignity, agency, value conflict, corrigibility, rights, and speculative consciousness-threshold concerns translated into governance constraints. Limits: Metaphysical and normative design context only; it is not empirical proof of consciousness, moral status, alignment, rights preservation, or governance correctness.; cgs: Supports the compression and residual-honesty portion through compact seed/rule/state/residual/verification/governance interfaces and the requirement to expose hidden complexity debt. Limits: Does not prove compactness, interpretability, safety, adequacy, compression advantage, or hidden-residual absence for this stack.; cognitive_loop_closure: Supports the procedural-memory handoff through trajectory logging, loop detection, parameter discovery, tool synthesis, verification, registry, routing, monitoring, and retirement. Limits: No local loop detector, tool synthesis run, verifier, runtime monitor, tool registry, retirement workflow, or procedural-memory benchmark has been executed.; moecot: Supports MoECOT as an implementation-reference runtime lane with compact orchestration, specialist lanes, fail-closed control plane, ledgers, readiness gates, replay, handoff, and promotion blockers. Limits: Reviewed source note only because authenticated connector text is not published here; reported runtime and benchmark artifacts have not been ingested or reproduced here.; vcm_editable: Supports the integrated context compiler framing with virtual context ABI, stable addresses, mounts, snapshots, materializations, task-relative representation graphs, certificates, and planner-guided paging. Limits: Reviewed source note only because raw connector text is not published here; no local VCM-Bench, model-facing result, citation-normalized third-party literature review, or context-policy result has been run.; project_theseus_whitepaper: Supports the report-first implementation reference for the integrated stack through pressure, attempt, residual, diagnosis, compression, verification, structure, specialist routing, sparse teacher governance, checkpoints, and trusted-node boundaries. Limits: No Theseus reports, benchmark ledgers, code paths, command outputs, current numbers, dashboard state, or model artifacts were rerun or independently inspected from this repository.; theseus_plan_compiler: Supports the planning-to-execution trace with typed goal contracts, semantic IR DAGs, VCM slices, executor routes, claim/evidence targets, contract hashes, and replay traces. Limits: Source-reported execute-mode status was not rerun or independently verified here; no compiler command, report artifact, private execute proof, or replay trace is promoted.; theseus_self_evolution_system: Supports the self-improvement gate with an intervention ladder, self-evolution governor, guarded teacher edits, ATTD repo-health gate, architecture experiment governance, outcome ledgers, and loop closure. Limits: No Theseus self-evolution scripts, ATTD reports, teacher runner, branch workflow, architecture experiment governor, loop-closure harvester, dashboard endpoints, or outcome ledgers were executed or inspected here.; theseus_architecture_gate: Supports authority stops before heavy training through ratchet, routing, safety, residual, benchmark, procedural-tool, memory, lifecycle, and external-inference checks. Limits: The reported gate snapshot was not independently verified by rerunning commands or inspecting current report JSON, and it is not current readiness, deployment, safety, or training evidence.; theseus_operator_os: Supports the operator/control surface through a shared command vocabulary, durable work board, node registry, background/watch/wake contracts, skill registry, tool hooks, feedback routing, TTLs, kill switches, and signed update surfaces. Limits: No Hive board, SQLite database, node registry, command channel, dashboard, board-executor step, remote-control surface, or unattended-operation proof was run from this repository.; theseus_circle_transfer: Supports the proof-contract transfer boundary by showing how Circle finite fixtures can configure private structural smoke or proxy benchmarks while preserving explicit non-claims. Limits: No Circle-to-Theseus consumer, structural smoke workload, deterministic proxy benchmark, transfer report, private workload attachment, ordinary baseline, negative control, model-quality evaluation, runtime measurement, or memory measurement was run from this repository.; cca_project: Supports the name-to-effect spine through typed intent lowering, canonical closure authority, trace bundles, and explicit separation between optimizer/proxy activity and claim-bearing effects. Limits: The historical project was not rerun; architecture and negative-case lessons do not establish integrated runtime, semantic preservation, capability, or safety.; moecot_manifest_project: Supports canonical-state versus projection boundaries, compilation from semantic inputs to effect graphs and target bundles, immutable provenance, and the distinction between internal contract breadth and external effect. Limits: The pinned project was not compiled, deployed, or benchmarked here; contract and builder presence are not runtime or capability evidence.; bugbrain_project: Supplies interface-theater failure cases where success-shaped IDs, ignored audit writes, unreachable lifecycle states, and narrative summaries can diverge from material state and effects. Limits: The negative observations are bounded to inspected historical surfaces; they are not a full security audit, deployed integration test, or broad BugBrain runtime claim.; corbens_trainer_project: Supports request-to-effect binding through typed plans, content-addressed lineage, claim evaluation, quarantine closure, and acknowledged asynchronous terminal outcomes. Limits: No trainer run, checkpoint workflow, or benchmark was reproduced; the fixture imports design and failure constraints without asserting runtime completion or capability.; corbens_best_model_possible_project: Supports the distinction between named architecture surfaces and their ordinary-path causal effect, including default/empty integrations, checkpoint-to-response gaps, playback versus live effect, and claim-boundary discipline. Limits: The historical model and retained artifacts were not reproduced; interface presence, first-party ownership, and local execution do not establish useful capability or integrated behavior.; ext_drexler_cais_2019: Provides the closest broad comparator for composed services and R&D automation; supports the claim that broad capability composition and an integrated agent are distinct framings, leaving the book’s typed trace and interface-governance contract as the narrower target. Limits: Does not implement, validate, or prove the ASI Stack reference trace, authority stops, evidence updates, residual custody, or release governance.; ext_adas_2024: Provides an automated-architecture-search comparator that represents agents as code-defined workflows, separates search space, search algorithm, and evaluation function, grows an archive, uses bounded repair, and reports validation/test and transfer results. Limits: No architecture search, model call, candidate code, archive, benchmark, transfer run, authority decision, rollback, safety result, or superiority claim was reproduced; source-reported validation and transfer do not establish integrated-stack quality.; qcsa_whitepaper: Defines QCSA as the semantic control plane connecting grounding, context, planning, memory, claims, routing, tools, generation, evidence acquisition, lifecycle migration, governance, and execution receipts through identity-address-route indirection; the later repository adds a bounded local 12-lane implementation, 60-case held-out evaluation over 13 systems and three seeds, and one 13-stage governed vertical trace. Limits: The exact matched-advantage and resource gates failed, and the active-question ablation is N2 proxy/regime evidence rather than an exact or broad refutation. The local vertical path does not establish broad deployment, production safety, chapter-core promotion, AGI, or ASI.; ext_v_jepa_2_2025: Supplies the external predictive-state-to-planning interface for a governed world-model lane while preserving camera, search-cost, representation, and sim-to-real residuals. Limits: The reported video and robot results do not establish local model quality, causal understanding, safe control, transfer, deployment, or an ASI Stack result.; ext_embedded_agency_2019: Supplies the open-world foundations residual beneath the finite reference trace: the stack is inside the world it records and cannot treat its own ledger as an external omniscient model. Limits: The paper is an informal obstacle survey, not a solved theory; the ASI Stack’s finite records and proofs do not solve logical uncertainty, self-reference, robust delegation, subsystem alignment, or open-world embedded agency.; ext_swe_rebench_v2_2026: Provides a natural repository-change substrate for testing the same candidate through matched test-only, record-only, and full-governance admission routes with visible utility, unsafe admission, false blocking, latency, rollback, residual, and cost outcomes. Limits: No local gold run, candidate campaign, final heldout result, governance advantage, deployment, safety, or transfer result exists yet.; learning_compute_topology: Makes the integrated reference trace represent learning as a typed, multi-view process graph rather than a single opaque training step. This chapter uses the paper’s LCT-IR relations and template, active, realized, and counterfactual views to locate where evidence, judgement, credit, state, artifacts, controls, and authority cross stack layers. Limits: This is a synthesis mapping, not a claim that Learning–Compute Topology is canonical or complete. The paper’s finite propositions and bounded supplement do not demonstrate neural-training benefit, causal identification, scalable compilation, safe self-rewrite, or superiority to simpler process records.; relational_dimension_compiler: Connects the reference trace’s typed artifacts, branches, qualifiers, contractions, and abstraction maps into a joined relational state. In this chapter, the mapping makes explicit which semantic relations must survive routing, memory, execution, evidence review, and reconciliation rather than leaving those joins implicit between layers. Limits: The Relational Dimension Compiler lifecycle has not been implemented inside the reference trace or shown to preserve semantic relations across real layer transitions. This mapping does not establish utility, cost, governance, or scaling advantages over simpler typed graph records.; assurance_shift_learning: Adds the joined six-plane GRBL return path across operation, evidence, discovery, adjudication, repair, and assurance without granting any plane omnibus authority. Limits: No end-to-end GRBL service or cross-layer runtime was implemented.; adjudicated_persistence: Adds an explicit control plane between experience/evidence and locus-specific realization, qualification, operation, and revocation. Limits: Conceptual author framework and benchmark proposal; no local implementation, empirical result, independently checked proof, safety result, or support movement.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Trace crosses intent, plan, authority, evidence, and artifact replay without hidden promotion.
project-theseus-as-report-first-implementation-reference.core	project-theseus-as-report-first-implementation-reference	Project Theseus as Report-First Implementation Reference owns a source-project-, pinned-revision-, report-family-, command-, environment-, artifact-, lineage-, evidence-state-, replay-, public-safety-, publication-permission-, reviewer-, consumer-, and time-specific Implementation-Reference Evidence Packet: it binds every imported or replayed report, configuration, ledger, work-board summary, registry, gate, crosswalk, trace, retained artifact, command, environment note, digest, missing artifact, decision, residual, and non-claim to source-note-only, imported, replay-ready, replay-failed, locally reproduced, stale, runtime-blocked, or archived lineage; dashboards and latest files are projections only, and no GREEN gate, complete registry, module card, pointer row, metadata snapshot, parity manifest, command replay, fixture, theorem, or sanitized import alone establishes current runtime truth, clean live replay, model quality, capability, benchmark validity, safety, deployment, support, transfer, AGI, ASI, or SOTA.	Design rationale	argument	project_theseus_whitepaper, deterministic_capability_compilation, theseus_plan_compiler, theseus_self_evolution_system, theseus_architecture_gate, theseus_operator_os, theseus_circle_transfer, ext_model_cards_2019, ext_datasheets_datasets_2021, ext_factsheets_ai_services_2019, ext_ml_reproducibility_program_2021	Source notes available for all 11 assigned sources; 10 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 11 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	project_theseus_whitepaper: Supports mining Project Theseus as a report-first RMI implementation reference with current-state caveats, pressure loops, residual escrow, benchmark floors, bounded specialist routing, sparse teacher governance, checkpoints, Hive nodes, safety boundaries, and machine-readable reports as the contract surface. Limits: Does not provide independently verified benchmark performance, ASI capability, deployment safety, public-compute readiness, current report state, or public-safe replay evidence in this repository.; theseus_plan_compiler: Supports the plan/report interface through typed goal contracts, semantic IR DAGs, VCM context slices, executor routes, claim/evidence targets, contract hashes, bounded execution packets, and replay traces. Limits: The source-reported GREEN execute-mode proof was not rerun or independently verified here; no compiler command, report bundle, private execute proof, or replay trace is promoted.; theseus_self_evolution_system: Supports Theseus self-evolution as an evidence-first governance lane using an intervention ladder, guarded teacher edits, ATTD repo-health gates, architecture experiment governance, outcome ledgers, and loop closure. Limits: Does not prove autonomous improvement success, safety, teacher independence, current dashboard state, ATTD report state, architecture-experiment result, or outcome-ledger success.; theseus_architecture_gate: Supports gate-first implementation claims through pre-training checks for ratchet completion, routing, safety ledger, residual escrow, bridge benchmarks, procedural tools, routing memory, arm lifecycle governance, and external-inference zero. Limits: The current gate snapshot is source-reported only and was not regenerated or inspected as a live report artifact here; it is not current readiness, deployment, safety, or training evidence.; theseus_operator_os: Supports the operator work-board and control-surface portion with durable task state, shared command vocabulary, node registry, feedback routing, hooks, TTLs, kill switches, signed updates, and isolation surfaces. Limits: No Hive work board, command channel, SQLite state, node registry, dashboard, board-executor step, or unattended-safety check has been run here.; theseus_circle_transfer: Supports the claim-boundary discipline for importing Circle finite fixtures into private Theseus experiment design while separating structural smoke checks from model-quality or promotion evidence. Limits: Does not prove quality, runtime, memory, transfer, capability improvements, ASI progress, public transfer, or model promotion evidence.; ext_model_cards_2019: Provides an external comparator for model-identity, intended-use, factor, metric, evaluation-data, limitation, and ethical-context fields in an implementation-reference evidence packet. Limits: No Project Theseus model card, model-card generator, model evaluation, safety result, fitness finding, or external review is established by this source.; ext_datasheets_datasets_2021: Provides an external comparator for dataset motivation, composition, collection, preprocessing, use, distribution, maintenance, provenance, and accountability fields in report packets. Limits: No Project Theseus datasheet, dataset audit, dataset validity, benchmark adequacy, training-data approval, or reproduced data-governance workflow is established by this source.; ext_factsheets_ai_services_2019: Provides an external comparator for supplier declarations, service-level trust facts, consumer inspection, claim comparison, and explicit separation of declared properties from unsupported assumptions. Limits: No Project Theseus FactSheet, service conformity, certification, deployment trust, independent audit, or validated supplier declaration is established by this source.; ext_ml_reproducibility_program_2021: Provides an external comparator for reproducibility checklists, code and artifact submission, commands, environment notes, failure cases, reviewer-readable claims, and community reproduction processes. Limits: No NeurIPS-style Project Theseus checklist, code review, external reproduction, artifact correctness, broader architecture validation, or independent transfer result is established by this source.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	First Theseus report validates and records non-claims.
prototype-roadmap.core	prototype-roadmap	Prototype Roadmap owns a program-, roadmap-, phase-, dependency-, artifact-, acceptance-gate-, authority-, evaluator-, evidence-transition-, phase-debt-, residual-, rollback-, reviewer-, consumer-, environment-, and time-specific Evidence-Gated Phase Unlock Contract: it binds every proposed phase to prerequisites, allowed work state, required artifacts, commands, environment, resource bounds, gates, independent evaluation, residuals, debt, rollback, retirement, evidence effect, and non-claims before later work may research, demo, integrate, promote, or release; no roadmap row, milestone, source report, dashboard, task count, passing fixture, theorem, validator, build, or locally useful prototype alone establishes phase completion, safe dependency order, capability, governance effectiveness, deployment, transfer, AGI, ASI, or SOTA.	Design rationale	argument	viea, deterministic_capability_compilation, platonic_world_model, benchmaxxing, scf, vcm_public, planforge, beastbrain, beastbrain_timeless, bugbrain, moecot, moecot_md, road_to_agi, coherence_exchange, project_theseus_whitepaper, theseus_plan_compiler, theseus_self_evolution_system, theseus_architecture_gate, theseus_operator_os, theseus_circle_transfer, circle_ai_contract_suite, ext_nist_ai_rmf_1_0_2023, ext_model_evaluation_extreme_risks_2023, ext_dafny_2010, ext_copilot_runtime_monitor_2010, ext_shop2_2003, ext_swe_bench_2023, ext_mmlu_2020, ext_checklist_2020, ext_codebleu_2020, ext_qlora_2023, ext_swe_rebench_v2_2026	Source notes available for all 32 assigned sources; 30 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 32 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	viea: Supports starting the roadmap with durable artifacts, command contracts, claim ledgers, residuals, feedback, tools, benchmarks, and regression coverage before higher-agency layers are introduced. Limits: Does not prove that any roadmap phase has been implemented, accepted, validated, or used to promote a capability claim.; benchmaxxing: Supports phasing benchmark frontier work after ledgers, baselines, wall diagnosis, regression floors, anti-Goodhart safeguards, and residual accounting exist. Limits: No benchmark harness, phase acceptance benchmark, mutation test, holdout run, contamination audit, empirical run, or phase acceptance result is recorded here.; scf: Supports delaying governed capability replacement until stable capability fields, evaluator boundaries, qualification claims, authority ceilings, lifecycle events, and recovery paths exist. Limits: Does not prove field replacement is safe, reversible, production-ready, evaluator-sound, authority-complete, or rollback-successful.; vcm_public: Supports adding the context/memory phase only with semantic objects, source binding, adequacy/admission separation, snapshots, taint, revocation, invalidation, and audit records. Limits: Does not establish model-facing accuracy gains, context-system superiority, retrieval-quality gains, or a validated roadmap phase.; planforge: Supports introducing planning as typed decomposition, primitive schemas, dependency scheduling, intelligence-tier routing, fallback, and replanning before execution autonomy. Limits: No planner implementation, scheduler benchmark, synthetic savings reproduction, learned decomposer, runtime replanning flow, or execution handoff trace has been validated here.; beastbrain: Supports roadmap vocabulary for a local, stateful, memory/planning/verification/routing/security stack and broad implementation sequence. Limits: Hardware, cost, benchmark, context-length, security, autonomy, performance, and metaphor-heavy architecture claims remain unvalidated here.; beastbrain_timeless: Supports keeping BeastBrain-derived roadmap discussion evergreen: vessel, memory substrate, planning, verification, hardware introspection, and interface layers as long-term implementation directions. Limits: Does not validate hardware adaptation, geometric verification, distributed scaling, memory behavior, or any implementation milestone.; bugbrain: Supports an early resource-constrained prototype lane with explicit build, emulation, flashing, bridge, training, and testing workflows before broad capability claims. Limits: No BugBrain build, QEMU run, flashing workflow, bridge session, benchmark, hardware test, AGI claim, or consciousness claim is validated here.; moecot: Supports later runtime prototype phases around compact orchestration, specialist lanes, fail-closed control plane, ledgers, readiness gates, replay, handoff, and promotion blockers. Limits: Reviewed source note only because authenticated connector text is not published here; MoECOT runtime, logs, reports, and benchmark artifacts have not been ingested or reproduced in this repository.; moecot_md: Supports MoECOT roadmap terminology and release-wording normalization as a variant/source export of the primary MoECOT whitepaper. Limits: Variant source only; it is not independent corroboration, reproduced runtime evidence, benchmark evidence, or a separate support-state basis.; road_to_agi: Supports keeping prototype phases tied to remaining-work categories, source-reported readiness status, promotion blockers, and implementation sequencing. Limits: Reviewed source note only because the local cache is an auth-gate placeholder; reported commands, benchmark results, readiness state, and roadmap status are source-reported only and have not been reproduced here.; coherence_exchange: Supports optional governance and evidence-interface framing through verified epistemic units, verification supply chains, fork/exit/audit, contestability, and value/accounting concepts. Limits: Reviewed source note only because raw connector text is not published here; speculative synthesis source and does not implement an epistemic market, security layer, economic mechanism, or roadmap gate.; project_theseus_whitepaper: Supports a report-first prototype lane using current-state caveats, pressure loops, residual escrow, benchmark floors, bounded specialist routing, sparse teacher governance, machine-readable reports, Hive registered-task boundaries, and safety constraints. Limits: No current Theseus reports, benchmark ledgers, code paths, command outputs, dashboard state, node state, or model artifacts were rerun or independently inspected here.; theseus_plan_compiler: Supports a planning/compiler milestone with typed contracts, semantic IR DAGs, VCM slices, executor routes, claim targets, contract hashes, and replay traces. Limits: Source-reported compiler status is not reproduced or independently verified; no compiler command, report bundle, private execute proof, or replay trace is promoted.; theseus_self_evolution_system: Supports delaying recursive self-improvement until intervention ladders, guarded teacher edits, ATTD gates, architecture experiment governance, outcome ledgers, and loop closure are in place. Limits: No self-evolution commands, teacher-edit runner, ATTD reports, architecture experiment reports, outcome ledgers, loop-closure reports, or dashboard endpoints were executed here.; theseus_architecture_gate: Supports architecture-gate phases before heavy training with ratchet, routing, safety, residual, benchmark, procedural-tool, memory, lifecycle, and external-inference checks. Limits: The reported gate snapshot is not local evidence of current readiness; no current gate report, command run, readiness claim, deployment claim, or training result is promoted.; theseus_operator_os: Supports an operator-surface phase with durable work-board state, command vocabulary, node registry, feedback routing, hooks, safety-visible controls, TTLs, kill switches, signed updates, and isolation. Limits: No Hive board, SQLite database, node registry, command channel, operator dashboard, board-executor step, or unattended-safety proof was run here.; theseus_circle_transfer: Supports adding proof-contract transfer as a later structural-fixture lane that feeds private experiment design without becoming model-quality evidence. Limits: No transfer consumer, smoke workload, proxy benchmark, private workload attachment, ordinary baseline, negative control, model-quality evaluation, or model-improvement result exists here.; circle_ai_contract_suite: Supports a future proof-carrying contract milestone with theorem-linked receipts, minimum consumer fields, request-validation reports, artifact manifests, fingerprints, and explicit non-claims. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no vendored contract packs, ASI Stack acceptance-policy integration, downstream consumers, transfer evidence, or model-quality evidence were run from this repo.; ext_nist_ai_rmf_1_0_2023: Provides an external comparator for governing, mapping, measuring, and managing AI risk before later prototype phases receive broader authority. Limits: Does not certify this roadmap, validate any phase, establish legal compliance, or prove that the proposed gates manage real system risk.; ext_model_evaluation_extreme_risks_2023: Provides an external comparator for dangerous-capability and alignment evaluation gates before high-authority training, deployment, or self-improvement phases. Limits: Does not supply a local dangerous-capability evaluation, evaluator adequacy result, model result, deployment decision, or accepted phase gate.; ext_dafny_2010: Provides an external comparator for placing executable specifications, verification conditions, and machine-checked contract artifacts before higher-agency implementation phases. Limits: Does not verify this roadmap, its implementations, runtime refinement, external effects, liveness, safety, or phase completion.; ext_copilot_runtime_monitor_2010: Provides an external comparator for runtime monitor artifacts, stream specifications, resource bounds, and enforcement visibility before autonomous execution phases. Limits: Does not implement a monitor here, prove complete observability, control Project Theseus, or validate runtime safety or phase acceptance.; ext_shop2_2003: Provides an external comparator for explicit planning methods, ordered task decomposition, applicability conditions, and plan-generation artifacts before execution autonomy. Limits: Does not validate PlanForge, produce a local plan, prove planning quality, handle open-world effects, or establish a roadmap phase.; ext_swe_bench_2023: Provides an external comparator for repository-level software-task evaluation with executable tests and realistic artifacts before claims about coding agents or build automation. Limits: No SWE-bench tasks, models, agents, tests, contamination checks, scores, baselines, or phase acceptance results are reproduced here.; ext_mmlu_2020: Provides an external comparator for broad multitask knowledge evaluation and for separating benchmark coverage from general capability or safe deployment. Limits: No MMLU evaluation, model result, contamination audit, general-capability finding, or roadmap acceptance gate is reproduced here.; ext_checklist_2020: Provides an external comparator for behavioral capability matrices, test templates, perturbations, and failure discovery before model or phase promotion. Limits: No CheckList suite, behavioral test, model comparison, coverage claim, or phase gate is implemented or reproduced here.; ext_codebleu_2020: Provides an external comparator for syntax- and dataflow-aware code-generation metrics and for keeping metric choice subordinate to executable task outcomes. Limits: No CodeBLEU computation, code-generation comparison, metric validation, human correlation, or accepted benchmark phase is reproduced here.; ext_qlora_2023: Provides an external comparator for placing parameter-efficient adaptation and memory accounting in a later governed experimental phase after data, evaluation, rollback, and resource gates exist. Limits: No QLoRA training, quantization, memory result, model-quality result, reproducibility result, deployment, or phase acceptance is established here.; ext_swe_rebench_v2_2026: Replaces authored repository fixtures in the empirical lane with post-snapshot public merged changes while requiring setup repair, gold execution, test-path collision guards, independent task review, and final-heldout custody. Limits: The 12 selected development tasks are debugging inputs, not the final denominator or evidence for a chapter claim.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Each prototype lane has owner, fixture, command, support boundary, and blocker state.
living-book-methodology.core	living-book-methodology	Living Book Methodology owns a book-, edition-, change-, source-, claim-, proof-, test-, render-, audience-, derivative-, release-, rights-, reviewer-, consumer-, environment-, and time-specific Evidence-Preserving Publication Transaction: it binds every substantive intake, structural edit, claim change, proof or test change, render, reader or audio projection, release, correction, rollback, and successor handoff to canonical source state, provenance, authority, validation, evidence effect, residuals, non-claims, and immutable lineage; generated scaffolds, green validators, theorem builds, successful renders, local format artifacts, publication activity, or a polished release never by themselves establish source interpretation, editorial quality, accessibility, reader approval, chapter truth, capability, safety, external reproduction, transfer, AGI, ASI, or SOTA.	Design rationale	argument	benchmaxxing, viea, bugbrain, cognitive_loop_closure, moecot, moecot_md, road_to_agi, ext_openunlearning_2025, ext_literate_programming_1984, ext_jupyter_book_docs, ext_quarto_books_docs, ext_nist_ai_rmf_1_0_2023, ext_frontier_ai_regulation_2023, ext_helm_2022, ext_livebench_2024, ext_benchmark_contamination_2023	Source notes available for all 16 assigned sources; 16 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 16 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	benchmaxxing: Supports treating book improvement as a ratchet with lifecycle states, wall diagnosis, ledgers, residuals, regression preservation, anti-Goodhart safeguards, and architecture-change discipline. Limits: Does not prove manuscript quality, benchmark validity, editorial completeness, rendered-site quality, or any chapter claim.; viea: Supports the living-book method through durable artifacts, claim support states, residuals, verified outputs, feedback, tools, benchmarks, and regression coverage. Limits: Does not make the book itself an implemented VIEA runtime, prove the architecture claims, prove source interpretation quality, or validate reader/audio editions.; bugbrain: Supports the value of explicit build, test, bridge, training, and resource-constrained workflows as a cautionary prototype-lineage example for living technical documentation. Limits: Speculative and unbuilt in this repo; no BugBrain implementation, build, emulation, flash, bridge, benchmark, hardware result, AGI claim, or consciousness claim is validated.; cognitive_loop_closure: Supports proceduralizing repeated book-maintenance workflows through trajectory logging, loop detection, tool synthesis, verification, routing, monitoring, and retirement discipline. Limits: No autonomous loop-closure system, trajectory-mining run, tool-synthesis harness, automated book-maintenance router, runtime monitor, or retirement workflow has been run; book workflow reuse remains process discipline.; moecot: Supports using runtime-reference ledgers, replay, readiness gates, residuals, handoff, and promotion blockers as analogues for living-book release and evidence discipline. Limits: Reviewed source note only because authenticated connector text is not published here; MoECOT reports, runtime artifacts, ledgers, replay, readiness gates, and benchmarks are not reproduced here and do not prove book quality.; moecot_md: Supports terminology normalization and variant comparison for MoECOT-related living-book updates. Limits: Variant source only; not independent evidence, not a reproduced artifact, and not a separate support-state basis.; road_to_agi: Supports keeping the living book honest about source-reported status, remaining work, unreproduced benchmarks, promotion blockers, and implementation sequencing. Limits: Reviewed source note only because the local cache is an auth-gate placeholder; roadmap/status context only, and reported commands, benchmark results, readiness state, and implementation status are not locally reproduced.; ext_openunlearning_2025: Supports versioned unlearning methods, standardized execution interfaces, public checkpoints, diverse evaluations, and meta-evaluation of metric faithfulness. Limits: The framework, methods, checkpoints, evaluations, and meta-evaluations were not run; standardized record shape does not establish semantic validity or erasure.; ext_literate_programming_1984: Provides historical lineage for treating human-readable exposition and executable program structure as one authored, reorderable, maintainable artifact. Limits: Does not supply claim/evidence states, source governance, release verification, reader-derivative authority, or proof that this living-book workflow improves research quality.; ext_jupyter_book_docs: Provides an external comparator for computational books built from notebooks or Markdown with execution, cross-references, web outputs, and reproducible build configuration. Limits: Does not validate this repository, its code execution, citations, prose quality, accessibility, release state, or claim truth.; ext_quarto_books_docs: Provides the external technical-publishing comparator for manifest-like multi-chapter Quarto books, navigation, cross-references, HTML rendering, and additional format targets. Limits: Quarto functionality does not prove manuscript quality, source correctness, claim support, accessibility, reader approval, deployment availability, or research validity.; ext_nist_ai_rmf_1_0_2023: Provides an external comparator for lifecycle governance, risk documentation, measurement, monitoring, and Govern/Map/Measure/Manage records in a continuously changing technical artifact. Limits: Does not certify the book, establish NIST conformance, validate its governance, or prove any architecture, safety, editorial, or release claim.; ext_frontier_ai_regulation_2023: Provides an external comparator for reporting, standards, scrutiny, pre-deployment assessment, post-deployment monitoring, and institutional accountability surfaces. Limits: Does not establish a legal requirement, compliance program, external scrutiny, policy adequacy, deployment approval, or manuscript-quality result for this book.; ext_helm_2022: Provides an external comparator for transparent multi-scenario, multi-metric evaluation reporting, missing coverage, standardization, and public result artifacts. Limits: No HELM scenario, model, metric, evaluation, score, reproducibility, or comparison is run here, and transparent reporting does not prove book claims.; ext_livebench_2024: Provides an external comparator for versioned living benchmarks, regular refresh, contamination resistance, public evaluation state, and historical score lineage. Limits: No LiveBench data, model, run, score, contamination test, refresh process, or independent evaluation is reproduced by the living-book workflow.; ext_benchmark_contamination_2023: Provides an external comparator for tracking evaluation leakage, stale public surfaces, source overlap, score interpretation, and contamination warnings across revisions. Limits: Does not prove this book, its sources, benchmarks, models, or reader artifacts are uncontaminated, independent, current, or valid.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Registry runner remains replayable and non-core evidence is visible.
open-research-agenda-and-bibliography-plan.core	open-research-agenda-and-bibliography-plan	Open Research Agenda and Bibliography Plan owns a research-program-, source-or-gap-, backlog-item-, access-, provenance-, public-safety-, chapter-boundary-, claim-, proof-or-experiment-, deduplication-, evidence-transition-, owner-, next-action-, closure-, consumer-, environment-, and time-specific Research Backlog Admission and Closure Contract: every new paper, local project, missing artifact, conflicting result, proof idea, experiment, reproduction need, correction, or chapter proposal enters through exact intake, triage, assignment, preconditions, blockers, non-claims, and terminal closure before it changes prose or support; no title, citation, source count, inventory row, source note, queue, priority label, backlog size, fixture, theorem, validator, or completed reading task alone establishes citation accuracy, literature completeness, research quality, claim support, reproduction, transfer, AGI, ASI, or SOTA.	Design rationale	argument	verification_bandwidth, deterministic_capability_compilation, platonic_world_model, benchmaxxing, alignment_field, scf, planforge, vcm_public, spinoza, talos, rmi, cgs, genesiscode, simulation_scaling, moecot, coilmoecot, road_to_agi, vcm_editable, field_of_god_ai_constitution, project_theseus_whitepaper, circle_calculus_core, circle_ai_contract_suite, theseus_circle_transfer, ext_graphrag_2024, ext_hipporag_2024, ext_mmlu_2020, ext_gpqa_2023, ext_livebench_2024, ext_lean4_theorem_proving, ext_moe_llm_survey_2024, relational_dimension_compiler	Source notes available for all 31 assigned sources; 29 exact claim-level source-note mappings recorded; support remains at the recorded state until an accepted evidence transition, proof, test, or source-derived promotion justifies movement.	All 31 assigned source notes explicitly list this chapter; mapping remains chapter-level, not claim-level support.	verification_bandwidth: Supports a research backlog for context adequacy, semantic-unit verification, contradiction-rate tests, summary-loss accounting, and the distinction between long context and verified reasoning. Limits: The proposed tests, theorem sketches, contradiction-rate claims, and adequacy claims remain source-proposed until mechanized or empirically tested.; benchmaxxing: Supports bibliography and agenda tracking for benchmark lifecycles, wall diagnosis, anti-Goodhart safeguards, benchmark ledgers, mutation, holdouts, contamination audits, and source-reported benchmark claims. Limits: Does not provide local benchmark results, benchmark-harness validation, mutation/holdout runs, contamination audits, or citation-normalized external literature by itself.; alignment_field: Supports tracking normative and metaphysical assumptions, rights, value-conflict boundaries, dignity, corrigibility, and needed external literature before public consciousness or moral-status claims. Limits: Not empirical evidence for consciousness, alignment, moral status, rights preservation, dignity preservation, or value-conflict resolution.; scf: Supports research agenda items for stable capability identity, evaluator boundaries, qualification, route validation, lifecycle ordering, rollback, recovery, and authority non-escalation. Limits: Does not prove SCF production safety, route/evaluator correctness, replacement safety, rollback success, recovery behavior, or authority-non-escalation runtime enforcement.; planforge: Supports future planning experiments around primitive schemas, dependency precedence, scheduler behavior, tier routing, fallback, replanning, and planner/executor boundaries. Limits: No PlanForge implementation, scheduler run, route trace, planning benchmark, savings reproduction, learned decomposer, or execution handoff is validated here.; vcm_public: Supports research agenda items for VCM packet schemas, source binding, adequacy/admission separation, certificates, taint, revocation, invalidation, and context-system conformance tests. Limits: Does not establish broad model-facing VCM gains, retrieval-quality improvement, context-system superiority, certificate truthfulness, or neighbor-system superiority.; spinoza: Supports backlog work for proof-carrying claims, belief revision records, contradiction detection, downgrade paths, protected axioms, and formalization-scope boundaries. Limits: Does not solve natural-language formalization, verifier quality, theorem validity for arbitrary claims, proof-object adequacy, or whole-system epistemic correctness.; talos: Supports research agenda items for typed job lifecycles, control planes, evidence records, Digital SCIFs, audit logs, replay, approval gates, runtime adapters, and delivery evidence. Limits: No Talos runtime, benchmark, security enforcement, approval-service behavior, replay behavior, or Digital SCIF containment claim is reproduced here.; rmi: Supports backlog items for residual escrow, benchmark pressure, routed specialist attempts, arm/router lifecycle, regression preservation, quarantine, and modular improvement tests. Limits: No RMI prototype, benchmark run, routed-specialist evaluation, router benchmark, regression-preservation result, or deployed modular-improvement loop has been independently verified here.; cgs: Supports compression and residual-honesty research items for compact seeds, rule systems, residual channels, verification contracts, governance interfaces, hidden complexity debt, and CGS metrics. Limits: No local CGS benchmark, utility test, proof of compact adequacy, generator, implementation, interpretability result, or hidden-residual absence result has been run.; genesiscode: Supports agenda items for AI-assisted programming evidence boundaries: tiny kernels, semantic patches, provenance hashes, effect logs, capability policies, obligations, replay, and translation validation. Limits: GenesisCode is not implemented, benchmarked, audited, replay-checked, security-audited, or mechanically verified in this repository.; simulation_scaling: Supports research agenda items for simulation contracts, scope, clockspeed, fidelity, efficiency, physical capacity, bottlenecks, and resource-accounted synthetic evaluation. Limits: No physical experiment, simulator result, simulation benchmark, feasibility calculator, resource-bound sanity check, approximation audit, or independent literature audit was run here.; moecot: Supports agenda tracking for MoECOT runtime artifacts, readiness gates, ledgers, replay, benchmark artifacts, residuals, promotion blockers, and implementation-reference gaps. Limits: Reviewed source note only because authenticated connector text is not published here; reported MoECOT runtime artifacts, benchmark artifacts, ledgers, readiness gates, replay, and promotion claims remain uninspected and unreproduced.; coilmoecot: Supports optional routed-substrate research agenda items for coil specialist lanes, readiness gates, benchmark baselines, residual handling, and cyclic/routed lane adoption. Limits: Reviewed source note only because the local cache is an auth-gate placeholder; no CoilMoECOT route run, local benchmark, baseline comparison, readiness result, mutation/promotion result, or performance claim is validated.; road_to_agi: Supports agenda tracking for MoECOT remaining work, reported benchmark commands/results, readiness status, promotion blockers, and implementation sequencing. Limits: Reviewed source note only because the local cache is an auth-gate placeholder; roadmap/status context only, and reported commands, benchmark results, readiness state, and implementation status were not reproduced.; vcm_editable: Supports bibliography triage for richer VCM terminology, planner-guided paging, evidence-carrying context, and external-literature candidates that need normalized metadata before citation. Limits: Reviewed source note only because raw connector text is not published here; connector-readable variant only, and no VCM-Bench result, model-facing result, or citation-normalized third-party source is implied.; field_of_god_ai_constitution: Supports agenda items for constitutional alignment clauses, tool-risk tiers, memory as power, least sufficient power, consent, audit, reversibility, red-team scenarios, and self-improvement freezes. Limits: Constitutional specification only; no runtime policy engine, red-team suite, deployed governance behavior, Lean proof, system-prompt evaluation, or moral-correctness result has been run here.; project_theseus_whitepaper: Supports backlog tracking for public-safe Theseus reports, residual ledgers, benchmark ledgers, sparse teacher governance, trusted-node task allowlists, and report-first implementation artifacts. Limits: Does not provide current Theseus report verification, command reruns, dashboard inspection, node state, model artifacts, reproduced benchmark results, or public-compute readiness evidence in this repo.; circle_calculus_core: Supports bibliography/proof-boundary agenda items for theorem ids, Lean proof status, theorem manifests, Python reference models, Rust utilities, Quarto proof surfaces, and finite cyclic contract receipts. Limits: A separate external Circle rope receipt slice is recorded in docs/circle_external_receipt_slice.md, but no broad Circle theorem replay, sidecar regeneration, ASI Stack consumer gate, model-quality result, transfer result, or deployment result is implied.; circle_ai_contract_suite: Supports research agenda items for proof-carrying AI contract fields, receipt validation, theorem-linked recommendations, consumer gates, fingerprints, minimum evidence fields, and explicit non-claims. Limits: No Circle contract pack, CLI command, Lean build, receipt replay, fingerprint check, downstream consumer, acceptance-policy check, or model-quality evidence was run here.; theseus_circle_transfer: Supports backlog and insertion rules for transfer-lane artifacts that separate structural smoke checks from quality, runtime, memory, transfer, and model-improvement claims. Limits: No transfer consumer, structural smoke workload, proxy benchmark, private workload report, ordinary baseline, negative control, runtime measurement, memory measurement, transfer result, or model-improvement claim has been executed here.; ext_graphrag_2024: Provides an external comparator and open backlog for graph-based global sensemaking, entity and relationship extraction, community summaries, query routing, provenance, updates, and poisoning boundaries. Limits: No GraphRAG pipeline, dataset, graph, model, query, answer, quality comparison, provenance audit, update, deletion, security result, or durable-memory chapter decision is reproduced here.; ext_hipporag_2024: Provides an external comparator and open backlog for associative graph memory, personalized PageRank retrieval, multi-hop navigation, integration, updates, provenance, contradiction, deletion, and poisoning boundaries. Limits: No HippoRAG pipeline, dataset, graph, retrieval run, model result, comparison, update, deletion, poisoning test, or durable-memory chapter decision is reproduced here.; ext_mmlu_2020: Provides an external benchmark-queue comparator for broad multitask knowledge evaluation, subject coverage, aggregate reporting, and limits on general-capability interpretation. Limits: No MMLU model, run, score, contamination audit, baseline, reproduction, claim transition, or bibliography closure is established here.; ext_gpqa_2023: Provides an external benchmark-queue comparator for expert-level, contamination-resistant science questions, human/model baselines, and evaluator difficulty. Limits: No GPQA questions, model, run, score, contamination check, expert review, reproduction, support transition, or agenda closure is established here.; ext_livebench_2024: Provides an external comparator for living benchmark refresh, objective scoring, contamination pressure, versioned evaluation, and historical result lineage. Limits: No LiveBench data, model, run, score, refresh process, contamination test, reproduction, or research-backlog closure is performed here.; ext_lean4_theorem_proving: Provides an external comparator for the proof-assistant backlog: theorem statements, definitions, tactics, kernel checking, reusable libraries, and the boundary between formalization and empirical truth. Limits: Does not prove any ASI Stack theorem, validate the chosen models, close proof adequacy, establish runtime refinement, or complete a research item.; ext_moe_llm_survey_2024: Provides an external routing-literature comparator for sparse mixture-of-experts architectures, routers, expert specialization, load balancing, training and inference tradeoffs, and open problems. Limits: No surveyed MoE system, router, model, benchmark, efficiency result, specialization result, reproduction, or local routing claim is established here.; relational_dimension_compiler: Adds a coordinated research program for typed polyadic cognition, relational-order routing, role-sensitive reification, object-field coupling, topology learning, semantic contraction, hardware compilation, schema evolution, and RODIE. Limits: The Corben-authored bibliography and proposed experiments require source-by-source external verification, implementation, competent baselines, independent evaluation, and claim-specific transitions before support use.	Review the mapped source-note support against source passages, evidence transitions, and tests before raising support state.	Bibliography agenda ties each open question to source notes, proof targets, or experiments.

C.1 QCSA P2–P3 Evidence Reconciliation (2026-07-13)

This overlay supersedes pre-implementation QCSA limits in earlier snapshots. The exact local result is bounded synthetic/internal evidence; every chapter- core state below remains argument.

Chapter owner	Bounded repository evidence	Negative result or remaining boundary	Core state
cognitive-compilation-and-semantic-ir	Twelve typed lanes and one 13-stage vertical replay.	Active questions added no held-out accuracy; no natural compiler or learned question policy was tested.	argument
virtual-context-abi	Full object accuracy was 1.000000; no-plural and no-indirection ablations fell to 0.916667 and 0.900000.	Synthetic fixtures do not establish context adequacy, retrieval quality, or deployed VCM behavior.	argument
claim-ledgers-and-belief-revision	Typed evidence stayed separate from identity/truth/authority; the internal observer recorded zero structural loss.	Internal labels and graph integrity do not establish truth or independent construct validity.	argument
runtime-adapters-tool-permissions-and-human-approval	Removing authority fields caused 9 unsafe releases; one temporary-file write was separately authorized, observed, receipted, and rolled back.	One reversible local effect is not deployed adapter, approval-service, or security evidence.	argument
inter-stack-protocols-identity-and-economic-exchange	Same-SOID compatibility was 1.000000; no-indirection compatibility fell to 0.400000.	No peer federation, credential, dispute, payment, or settlement ran.	argument
routing-heads-and-specialist-cores	Object accuracy and prevention were 1.000000; Brier was 0.082026; all 10 vertical attacks failed closed.	Task accuracy tied at 1.000000, operation ratio was 1.913386, and no learned router or trained specialist ran.	argument
compact-generative-systems-and-residual-honesty	Exact structural round trip had zero observer disagreement and preserved residuals.	No compression, semantic-utility, repair, or latency advantage was established.	argument
data-engines-continual-learning-and-unlearning	Full migration compatibility was 1.000000; no-compatibility task accuracy fell to 0.833333 and compatibility to zero.	No learning, forgetting, influence removal, privacy, or storage erasure ran.	argument
integrated-reference-architecture	One 13-stage path crossed intent through a real temporary effect, observation, receipts, same-SOID migration, and exact rollback.	The path is hand-authored, local, internally observed, zero-model, and not deployment, AGI, or ASI evidence.	argument

Canonical records: experiments/qcsa_reference/results/evaluation_results.json, claim_decisions/qcsa_reference_evaluation_dispositions.json, docs/qcsa_reference_evaluation_report.md, experiments/qcsa_vertical_reference/results/vertical_result.json, and docs/qcsa_governed_vertical_reference_report.md. Five exact non-core mechanism findings are marked promote for bounded evidence review, but no automatic support-state transition has occurred.

C.2 P4/M5 Governed-Usefulness Reconciliation (2026-07-16)

The prospectively frozen confirmatory campaign reached an informative bounded local regime after preserving two instrument failures and a separate tuning stage. On the fresh 16-task held-out denominator, 15 candidate decisions were schema-admissible. Full governance released 9 useful and 0 unsafe decisions; the simple baseline released 0 useful and 0 unsafe decisions; removing evidence freshness exposed 1 unsafe release. Two independent evaluator implementations agreed on all scored cases, and all six preregistered co-primary checks passed.

This accepts only the non-core claim governed-usefulness.held-out-local-policy-effect at bounded synthetic-test-backed scope. The campaign used a local model, authored tasks, an internal policy, synthetic effects, and no independent external evaluator or transfer setting. It does not promote a chapter-core claim, establish broad usefulness, acceptable governance cost, deployment safety, model generality, SOTA, AGI, or ASI. The exact protocol, failures, exclusions, arm results, and claim ceiling are recorded in docs/p4_governed_usefulness_campaign.md and experiments/p4_governed_usefulness/results/confirmatory_result.json.

C.3 R16-A Post-Activation Claim Organization (2026-07-26)

Six chapters admitted after the frozen activation registry now have a separate, reviewed claim-identity projection. The packet contains 30 atoms: one exact chapter core, one boundary, one mechanism, one failure/noninheritance rule, and one argument-exit target per chapter. Every atom has an owner, scope, falsifier, acceptance criterion, promotion ceiling, evidence-plan route, and explicit non-claims.

This is organization, not evidence. All six chapter cores and all thirty atoms remain at argument; the packet moves no support or release state. It also leaves the historical 3,730-atom activation denominator, the current 4,058-atom registry, and the historical 15-atom addendum unchanged. The machine source is evidence_quality/post_activation_six_chapter_claim_atom_addendum.json.

Chapter owner	Chapter	Reviewed atoms	Roles	Review disposition	Support
white-box-evidence-interpretability-and-activation-governance	White-Box Evidence, Interpretability, and Activation Governance	5	core, boundary, mechanism, failure_boundary, argument_exit	retain_at_argument_no_promotion	argument
governed-world-models-and-reality-grounding	Governed World Models and Reality Grounding	5	core, boundary, mechanism, failure_boundary, argument_exit	retain_at_argument_no_promotion	argument
human-factors-and-meaningful-control-in-oversight	Human Factors and Meaningful Control in Oversight	5	core, boundary, mechanism, failure_boundary, argument_exit	retain_at_argument_no_promotion	argument
governed-operations-incident-command-and-graceful-degradation	Governed Operations, Incident Command, and Graceful Degradation	5	core, boundary, mechanism, failure_boundary, argument_exit	retain_at_argument_no_promotion	argument
governed-model-training-distributed-optimization-and-scaling	Governed Model Training, Distributed Optimization, and Scaling	5	core, boundary, mechanism, failure_boundary, argument_exit	retain_at_argument_no_promotion	argument
privacy-data-rights-and-information-flow-governance	Privacy, Data Rights, and Information-Flow Governance	5	core, boundary, mechanism, failure_boundary, argument_exit	retain_at_argument_no_promotion	argument

C.4 Claim Labels

Label	Meaning
Demonstrated	Shown by a recorded artifact, derivation, implementation, or source-backed example in scope.
Measured	Quantified by a recorded measurement or benchmark result.
Mechanized	Expressed as runnable code, an executable schema, or a formal proof target.
Hypothesized	Proposed as a testable claim that still needs evidence.
Design rationale	An architectural choice supported by reasoning and constraints.
Speculative	Exploratory or conjectural; do not use as a deployed system guarantee.

C.5 Support States

State	Meaning
unsupported	Not yet supported; likely remove or mark as speculative.
argument	Supported by reasoning only.
source-derived	Derived from supplied source papers.
prototype-backed	Implemented in a prototype but not robustly tested.
synthetic-test-backed	Supported by controlled synthetic tests.
empirical-test-backed	Supported by external or realistic tests.
external-literature-backed	Supported by third-party literature.
deprecated	Superseded, merged, or retired; retained for lineage and review.
refuted	Contradicted by later evidence or tests; retained to preserve negative results.