Skip to main content

Appendix H — External Sources by Other Authors

This appendix is generated from source records marked external_literature in sources/source_inventory.json.

This is an independent top-level appendix for external sources by other authors and organizations. It is not a subsection, second half, or continuation of Appendix G. It contains only third-party papers, documentation records, outside benchmarks, and non-Corben references used for comparison, grounding, or future literature review. Corben’s own papers, Corben-supplied materials, recovered project records, and local project records live in their own top-level appendix: Appendix G, Corben’s Own Sources, Papers, and Local Projects.

A listed external source does not claim reproduced experiments, local benchmark results, support-state promotion, or complete literature coverage.

H.1 Source Ownership Boundary

This page is the external-source appendix. It should be read as Appendix H only; Corben-side records stay in Appendix G.

H.2 Appendix Identity

Field	Boundary
This appendix contains	Third-party papers, documentation, outside projects, and other external sources by authors other than Corben.
This appendix excludes	Corben’s own papers, Corben-supplied materials, recovered project records, and local project source records; those live in the separate Appendix G.

H.3 Appendix Scope

Field	Boundary
Appendix identity	Appendix H: external sources by other authors
Ownership rule	If another author, organization, or outside project produced it, it belongs here; Corben-authored papers, Corben-supplied materials, recovered project history, and local-project records stay in Appendix G.
Contains	Third-party papers, official documentation, outside benchmarks, and other non-Corben references by other authors or organizations used for comparison or grounding.
Excludes	Corben’s own papers, Corben-supplied materials, recovered project records, and local project records; those belong in Appendix G.
Evidence effect	Organizes outside context; it does not claim reproduced results or support-state promotion without a reproduction or accepted evidence transition.

H.4 Chapter-Level External Grounding Policy

Future citation backfill should start from the sources already attached to each chapter. Mine the linked Corben papers for bibliographies, footnotes, named algorithms, standards, benchmarks, outside systems, and adjacent research terms before broad search. Accepted third-party sources must be recorded in sources/source_inventory.json with priority: external_literature, receive a source note before prose use, and then appear here through scaffold generation.

External citation can establish relation, prior art, vocabulary, and comparison. It does not by itself claim reproduced experiments, local verification, Lean proof, Project Theseus replay, Circle receipt, support-state promotion, or complete literature coverage.

H.5 Source-Noted External Literature Records

Source ID	Title	Citation or primary record	Layer	Current use	Source-note state	Notes
ext_probe_control_tasks_2019	Designing and Interpreting Probes with Control Tasks	Hewitt and Liang (2019), Designing and Interpreting Probes with Control Tasks; source; DOI 10.18653/v1/D19-1275	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	Primary probe-method comparator for control tasks and selectivity: a probe must be evaluated against its capacity to learn control labels rather than treating linguistic-task accuracy as representation evidence. The source studies ELMo linguistic probes; it does not establish a universal probe test, causal use of decoded information, model safety, or an ASI Stack result.
ext_interpretability_illusion_bert_2021	An Interpretability Illusion for BERT	Bolukbasi et al. (2021), An Interpretability Illusion for BERT; arXiv:2104.07143; DOI 10.48550/arXiv.2104.07143	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	Primary cross-dataset construct-validity challenge showing that apparently coherent neuron or direction interpretations can change across corpora because datasets occupy different regions of representation space. The BERT sentence-embedding result does not prove that all features are illusory, that causal methods fail, or that the finding transfers unchanged to other models and modalities.
ext_saebench_2025	SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability	Karvonen et al. (2025), SAEBench; arXiv:2503.09532; DOI 10.48550/arXiv.2503.09532	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	Primary multi-metric SAE comparator spanning concept detection, automated interpretability, reconstruction, feature disentanglement, and downstream tasks. It reports that sparsity-fidelity rankings do not reliably predict other metrics and that one global score would obscure tradeoffs; its studied models, methods, metrics, and source-reported results do not establish semantic or causal faithfulness.
ext_sae_benchmark_reliability_2026	Are Sparse Autoencoder Benchmarks Reliable?	Chanin (2026), Are Sparse Autoencoder Benchmarks Reliable?; arXiv:2605.18229; DOI 10.48550/arXiv.2605.18229	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	Primary 2026 audit of selected SAEBench metrics through reseed noise, training-trajectory discriminability, and synthetic ground-truth correlation. It reports material reliability problems for TPP and SCR at canonical settings and weaker-than-assumed discrimination elsewhere. This is metric- and setting-scoped counterevidence, not a refutation of sparse autoencoders, interpretability, or every SAEBench task.
ext_constructive_interdependence_human_ai_2026	Who Is Helping Whom? Analyzing Inter-Dependencies to Evaluate Cooperation in Human-AI Teaming	Biswas et al. (2026), Who Is Helping Whom?; source; DOI 10.1609/aaai.v40i21.38787	multi_agent_dynamics_and_human_ai_organizations	human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability); multi-agent-dynamics-collective-intelligence-and-systemic-risk (Multi-Agent Dynamics, Collective Intelligence, and Systemic Risk)	source note available	AAAI-26 paper introducing constructive interdependence as a complement to task reward for evaluating human-agent cooperation in Overcooked. The source reports that high task reward can coexist with low interdependence in its studied teams; no local human study, teaming result, or general cooperation claim is reproduced.
ext_adversarial_sensor_fusion_2022	Adversarial Robustness of Deep Sensor Fusion Models	Wang et al. (2022), Adversarial Robustness of Deep Sensor Fusion Models; source	perception_sensor_fusion_and_observation_trust	adversarial-machine-learning-and-model-attack-surface (Adversarial Machine Learning and the Model Attack Surface); perception-sensor-fusion-and-observation-trust (Perception, Sensor Fusion, and Observation Trust)	source note available	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.
ext_imagebind_2023	ImageBind: One Embedding Space To Bind Them All	Girdhar et al. (2023), ImageBind; source	perception_sensor_fusion_and_observation_trust	perception-sensor-fusion-and-observation-trust (Perception, Sensor Fusion, and Observation Trust)	source note available	CVPR paper learning a shared space across image, text, audio, depth, thermal, and IMU modalities using image-paired data. It supplies a representation comparator; reported zero-shot and few-shot results do not establish calibrated sensor truth, robust fusion, causal grounding, or local performance.
ext_multimodal_machine_learning_taxonomy_2019	Multimodal Machine Learning: A Survey and Taxonomy	Baltrušaitis, Ahuja, and Morency (2019), Multimodal Machine Learning; source; DOI 10.1109/TPAMI.2018.2798607	perception_sensor_fusion_and_observation_trust	perception-sensor-fusion-and-observation-trust (Perception, Sensor Fusion, and Observation Trust)	source note available	Peer-reviewed survey organizing multimodal learning around representation, translation, alignment, fusion, and co-learning. It supplies taxonomy and research context, not a locally reproduced mechanism or evidence that any fusion design is adequate for consequential observation admission.
ext_control_barrier_functions_2019	Control Barrier Functions: Theory and Applications	Ames et al. (2019), Control Barrier Functions; source; DOI 10.23919/ECC.2019.8796030	embodied_real_time_control_and_physical_safety	embodied-agency-real-time-control-and-physical-safety (Embodied Agency, Real-Time Control, and Physical Safety)	source note available	Overview of control barrier functions for verifying and enforcing safety properties in optimization-based controllers, including robotic applications. It supplies a formal-control comparator under stated dynamics and set assumptions, not a universal physical-safety guarantee or local implementation result.
ext_simplex_architecture_1998	The Simplex Architecture for Safe On-Line Control System Upgrades	Seto et al. (1998), The Simplex Architecture; source; DOI 10.1109/ACC.1998.703255	embodied_real_time_control_and_physical_safety	embodied-agency-real-time-control-and-physical-safety (Embodied Agency, Real-Time Control, and Physical Safety)	source note available	American Control Conference paper describing a runtime architecture that protects an advanced controller with a safety controller and switching logic. It motivates independent fallback authority; its process-control case does not validate an ASI Stack controller or arbitrary learned policy.
ext_safe_reinforcement_learning_survey_2015	A Comprehensive Survey on Safe Reinforcement Learning	García and Fernández (2015), A Comprehensive Survey on Safe Reinforcement Learning; source	embodied_real_time_control_and_physical_safety	embodied-agency-real-time-control-and-physical-safety (Embodied Agency, Real-Time Control, and Physical Safety)	source note available	JMLR survey classifying safe reinforcement learning through modified optimality criteria and modified exploration using external knowledge or risk measures. It supplies a design taxonomy, not evidence that a particular controller is safe or that learning-time and deployment-time constraints compose.
ext_gemini_robotics_2025	Gemini Robotics: Bringing AI into the Physical World	Gemini Robotics Team et al. (2025), Gemini Robotics; source; DOI 10.48550/arXiv.2503.20020	embodied_real_time_control_and_physical_safety	perception-sensor-fusion-and-observation-trust (Perception, Sensor Fusion, and Observation Trust); embodied-agency-real-time-control-and-physical-safety (Embodied Agency, Real-Time Control, and Physical Safety)	source note available	Technical report on Gemini Robotics and Gemini Robotics-ER, including vision-language-action control, spatial reasoning, adaptation, and reported safety considerations. Capability results are source-reported and do not establish independent physical-safety assurance, local transfer, or general embodiment.
ext_ai_decision_authority_2020	The Allocation of Decision Authority to Human and Artificial Intelligence	Athey, Bryan, and Gans (2020), Allocation of Decision Authority; source; DOI 10.3386/w26673	human_ai_organizations_delegation_and_accountability	human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability)	source note available	Economic model of a principal allocating decision authority between a human and an AI while trading off alignment, human information-acquisition effort, and AI reliability. It supplies a bounded organizational-design comparator, not an empirical finding about all workplaces or an accountability solution.
ext_cooperative_ai_foundations_2023	Foundations of Cooperative AI	Conitzer and Oesterheld (2023), Foundations of Cooperative AI; source; DOI 10.1609/aaai.v37i13.26791	multi_agent_dynamics_collective_intelligence_and_systemic_risk	multi-agent-dynamics-collective-intelligence-and-systemic-risk (Multi-Agent Dynamics, Collective Intelligence, and Systemic Risk)	source note available	AAAI research agenda applying game-theoretic foundations to cooperation among advanced AI agents while noting settings where cooperation becomes harmful collusion. It supplies problem structure and comparator families, not a solved coordination mechanism or local population-level result.
ext_sleeper_agents_2024	Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training	Hubinger et al. (2024), Sleeper Agents; source; DOI 10.48550/arXiv.2401.05566	inner_alignment_and_learned_objective_integrity	inner-alignment-mesa-optimization-and-learned-objective-integrity (Inner Alignment, Mesa-Optimization, and Learned-Objective Integrity); adversarial-machine-learning-and-model-attack-surface (Adversarial Machine Learning and the Model Attack Surface)	source note available	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.
ext_toward_causal_representation_learning_2021	Toward Causal Representation Learning	Schölkopf et al. (2021), Toward Causal Representation Learning; source; DOI 10.1109/JPROC.2021.3058954	world_models_causal_reasoning_and_representation	governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding)	source note available	Proceedings of the IEEE article connecting graphical causality with representation learning and identifying discovery of high-level causal variables from low-level observations as a central open problem. It supplies a research frame, not a locally validated causal representation or intervention model.
ext_scaling_laws_neural_language_models_2020	Scaling Laws for Neural Language Models	Kaplan et al. (2020), Scaling Laws for Neural Language Models; source; DOI 10.48550/arXiv.2001.08361	scaling_laws_emergence_and_capability_forecasting	the-efficient-asi-hypothesis (The Efficient ASI Hypothesis); learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	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.
ext_chinchilla_compute_optimal_2022	Training Compute-Optimal Large Language Models	Hoffmann et al. (2022), Training Compute-Optimal Large Language Models; source; DOI 10.48550/arXiv.2203.15556	scaling_laws_emergence_and_capability_forecasting	the-efficient-asi-hypothesis (The Efficient ASI Hypothesis)	source note available	Study of compute-optimal allocation between model parameters and training tokens, based on more than 400 reported training runs and the Chinchilla comparison. It revises one scaling prescription within a bounded family; no local large-scale reproduction or universal optimum is claimed.
ext_emergent_abilities_2022	Emergent Abilities of Large Language Models	Wei et al. (2022), Emergent Abilities of Large Language Models; source; DOI 10.48550/arXiv.2206.07682	scaling_laws_emergence_and_capability_forecasting	the-efficient-asi-hypothesis (The Efficient ASI Hypothesis)	source note available	Paper cataloguing task abilities that appear discontinuously under particular model families, prompts, and metrics. It motivates threshold monitoring but does not establish that all reported discontinuities reflect abrupt underlying mechanisms or are prospectively predictable.
ext_emergence_mirage_2023	Are Emergent Abilities of Large Language Models a Mirage?	Schaeffer, Miranda, and Koyejo (2023), Emergent Abilities Mirage; source	scaling_laws_emergence_and_capability_forecasting	the-efficient-asi-hypothesis (The Efficient ASI Hypothesis)	source note available	NeurIPS paper showing that discontinuous metrics can create apparent emergence from smoothly changing model outputs in studied settings. It is a measurement critique and counterweight, not proof that every capability transition is smooth or non-emergent.
ext_deep_ensembles_2017	Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles	Lakshminarayanan, Pritzel, and Blundell (2017), Deep Ensembles; source	uncertainty_calibration_and_distribution_shift	governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding)	source note available	NeurIPS paper presenting independently trained probabilistic neural-network ensembles as a strong practical predictive-uncertainty baseline. Reported calibration and out-of-distribution behavior are benchmark-bound and do not provide distribution-free guarantees or local evidence.
ext_conformal_prediction_2021	A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification	Angelopoulos and Bates (2021), Conformal Prediction; source; DOI 10.48550/arXiv.2107.07511	uncertainty_calibration_and_distribution_shift	governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding)	source note available	Technical introduction to conformal prediction, coverage guarantees, and extensions. Coverage depends on the method’s stated exchangeability or shift assumptions and target; it does not establish semantic correctness, causal adequacy, safety, or local calibration.
ext_wilds_2021	WILDS: A Benchmark of in-the-Wild Distribution Shifts	Koh et al. (2021), WILDS; source	uncertainty_calibration_and_distribution_shift	governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding)	source note available	ICML benchmark of ten datasets with naturally occurring shifts across domains such as hospitals, camera traps, geography, and time. It supplies representative shift designs and reported gaps, not a universal OOD benchmark or local robustness result.
ext_taking_ai_welfare_seriously_2024	Taking AI Welfare Seriously	Long et al. (2024), Taking AI Welfare Seriously; source; DOI 10.48550/arXiv.2411.00986	moral_uncertainty_ai_welfare_and_moral_status	moral-uncertainty-and-value-conflict (Moral Uncertainty, Value Conflict, and Contestable Governance)	source note available	Interdisciplinary report arguing for precautionary attention to uncertainty about AI consciousness, robust agency, welfare, and moral patienthood. It does not establish that current systems are conscious, have welfare, or deserve any particular status, and it supplies no local assessment.
ext_functional_decision_theory_2017	Functional Decision Theory: A New Theory of Instrumental Rationality	Yudkowsky and Soares (2017), Functional Decision Theory; source; DOI 10.48550/arXiv.1710.05060	decision_theory_embedded_agents_and_multi_agent_dynamics	multi-agent-dynamics-collective-intelligence-and-systemic-risk (Multi-Agent Dynamics, Collective Intelligence, and Systemic Risk)	source note available	Paper defining functional decision theory and comparing its recommendations with causal and evidential decision theories on classic decision problems. It is a normative proposal with contested assumptions, not an empirically validated universal decision rule or a deployment policy.
ext_un_global_digital_compact_2024	Global Digital Compact	United Nations (2024), Global Digital Compact; source	international_ai_governance_and_public_legitimacy	institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy)	source note available	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.
ext_council_europe_ai_convention_2024	Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law	Council of Europe (2024), Framework Convention on Artificial Intelligence; source	international_ai_governance_and_public_legitimacy	institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy)	source note available	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.
ext_generative_ai_at_work_2025	Generative AI at Work	Brynjolfsson, Li, and Raymond (2025), Generative AI at Work; source; DOI 10.1093/qje/qjae044	ai_deployment_transition_distribution_and_human_agency	human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability); ai-deployment-transition-distribution-and-human-agency (AI Deployment, Transition, Distribution, and Human Agency)	source note available	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.
ext_ilo_genai_jobs_index_2025	Generative AI and Jobs: A Refined Global Index of Occupational Exposure	Gmyrek et al. (2025), Generative AI and Jobs; source; DOI 10.54394/HETP0387	ai_deployment_transition_distribution_and_human_agency	ai-deployment-transition-distribution-and-human-agency (AI Deployment, Transition, Distribution, and Human Agency)	source note available	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.
ext_iea_energy_and_ai_2025	Energy and AI	International Energy Agency (2025), Energy and AI; source	physical_compute_infrastructure_energy_and_environment	physical-compute-infrastructure-energy-and-environmental-constraints (Physical Compute Infrastructure, Energy, and Environmental Constraints)	source note available	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.
ext_lbnl_data_center_energy_2024	2024 United States Data Center Energy Usage Report	Shehabi et al. (2024), United States Data Center Energy Usage Report; source; DOI 10.71468/P1WC7Q	physical_compute_infrastructure_energy_and_environment	physical-compute-infrastructure-energy-and-environmental-constraints (Physical Compute Infrastructure, Energy, and Environmental Constraints)	source note available	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.
ext_nist_incident_response_2025	Incident Response Recommendations and Considerations for Cybersecurity Risk Management: A CSF 2.0 Community Profile	Nelson et al. (2025), NIST SP 800-61 Rev. 3; source; DOI 10.6028/NIST.SP.800-61r3	incident_response	societal-resilience-and-misuse-defense (Societal Resilience and Misuse Defense); governed-operations-incident-command-and-graceful-degradation (Governed Operations, Incident Command, and Graceful Degradation)	source note available	Official NIST incident-response baseline for integrating preparation, detection, response, recovery, and continuous improvement into cybersecurity risk management; it does not address every AI-specific failure mode or establish local incident readiness, response efficacy, recovery, compliance, or safety.
ext_llama3_herd_2024	The Llama 3 Herd of Models	Grattafiori et al. (2024), The Llama 3 Herd of Models; arXiv:2407.21783; DOI 10.48550/arXiv.2407.21783	governed_distributed_model_training_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Paper-body-reviewed large-run case: Sections 3.3.1–3.3.4 expose 4D topology, numerical policy, checkpoint infrastructure, interruption denominators, and effective training time. Provider-reported scale, utilization, failures, and recovery are not locally reproduced and do not establish exact resume.
ext_3d_detection_corruptions_2023	Benchmarking Robustness of 3D Object Detection to Common Corruptions	Dong et al. (2023), Benchmarking Robustness of 3D Object Detection to Common Corruptions; arXiv:2303.11040; DOI 10.1109/CVPR52729.2023.00105	perception_sensor_fusion_and_corruption_robustness	perception-sensor-fusion-and-observation-trust (Perception, Sensor Fusion, and Observation Trust)	source note available	Preliminary perception-robustness comparator based on the official CVF abstract: the source reports 27 LiDAR/camera corruption types, three synthetically corrupted benchmark suites, and evaluation of 24 detectors. The reported findings remain source-reported; no corruption suite, model evaluation, sensor-fusion result, or physical-safety result has been reproduced locally.
ext_foundation_robotics_physical_risk_2025	A Comprehensive Survey on Physical Risk Control in the Era of Foundation Model-enabled Robotics	Kojima et al. (2025), Physical Risk Control in Foundation Model-enabled Robotics; arXiv:2505.12583; DOI 10.48550/arXiv.2505.12583	embodied_agency_and_physical_risk_control	embodied-agency-real-time-control-and-physical-safety (Embodied Agency, Real-Time Control, and Physical Safety)	source note available	Preliminary physical-risk taxonomy based only on the official arXiv abstract: the survey organizes controls across pre-deployment, pre-incident, and post-incident phases and identifies open gaps around pre-incident mitigation, human interaction, and foundation-model-specific issues. No surveyed controller, robot experiment, runtime-assurance result, or physical-safety claim has been reproduced locally.
ext_nist_differential_privacy_2025	Guidelines for Evaluating Differential Privacy Guarantees	Near et al. (2025), NIST SP 800-226; source; DOI 10.6028/NIST.SP.800-226	privacy_guarantees_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Paper-body-reviewed official guidance distinguishing mathematical, implementation, system, and operational layers of a DP claim. It establishes no correct local implementation, utility result, lifecycle privacy, or legal compliance.
ext_multi_agent_risks_2025	Multi-Agent Risks from Advanced AI	Hammond et al. (2025), Multi-Agent Risks from Advanced AI; arXiv:2502.14143; DOI 10.48550/arXiv.2502.14143	multi_agent_dynamics_and_systemic_risk	multi-agent-dynamics-collective-intelligence-and-systemic-risk (Multi-Agent Dynamics, Collective Intelligence, and Systemic Risk)	source note available	Preliminary population-risk taxonomy based only on the official arXiv abstract: the report distinguishes miscoordination, conflict, and collusion and names seven contributing risk factors. Its examples and evidence remain source-reported; no population experiment, systemic-risk indicator, intervention, or mitigation result has been reproduced locally.
ext_replibench_2025	RepliBench: Evaluating the Autonomous Replication Capabilities of Language Model Agents	Black et al. (2025), RepliBench; arXiv:2504.18565; DOI 10.48550/arXiv.2504.18565	autonomous_replication_and_proliferation_evaluation	autonomous-replication-proliferation-and-containment (Autonomous Replication, Proliferation, and Containment)	source note available	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.
ext_autonomous_lab_materials_2023	An autonomous laboratory for the accelerated synthesis of inorganic materials	Szymanski et al. (2023), A-Lab; source; DOI 10.1038/s41586-023-06734-w	scientific_discovery_and_experimental_governance	scientific-discovery-and-experimental-governance (Scientific Discovery and Experimental Governance)	source note available	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.
ext_ai_scientist_end_to_end_2026	Towards end-to-end automation of AI research	Lu et al. (2026), Towards end-to-end automation of AI research; source; DOI 10.1038/s41586-026-10265-5	scientific_discovery_and_experimental_governance	scientific-discovery-and-experimental-governance (Scientific Discovery and Experimental Governance)	source note available	Passage-reviewed computational-research comparator: the reported system connects ideation, literature search, code, experiments, analysis, manuscript production, and automated review. Workshop review and paper completion are downstream observations rather than scientific truth; the source-reported system, manuscripts, search tree, and results have not been reproduced locally.
ext_coscientist_chemistry_2023	Autonomous chemical research with large language models	Boiko et al. (2023), Coscientist; source; DOI 10.1038/s41586-023-06792-0	scientific_discovery_and_experimental_governance	scientific-discovery-and-experimental-governance (Scientific Discovery and Experimental Governance)	source note available	Passage-reviewed bounded chemistry comparator: Coscientist connects a language-model planner to search, code, documentation, and robotic laboratory interfaces across six reported task families. The source-reported demonstrations remain equipment-, task-, supervision-, and assessment-bound and have not been reproduced locally.
ext_ai_co_scientist_2025	Towards an AI co-scientist	Gottweis et al. (2025), Towards an AI co-scientist; arXiv:2502.18864; DOI 10.48550/arXiv.2502.18864	scientific_discovery_and_experimental_governance	scientific-discovery-and-experimental-governance (Scientific Discovery and Experimental Governance)	source note available	Passage-bounded hypothesis-generation comparator based on the official preprint record and authors’ research overview: specialized agents generate, reflect on, rank, evolve, and meta-review hypotheses using additional inference compute. Internal Elo ranking, expert preference, and selected laboratory cases are distinct evidence objects; none is reproduced locally or treated as general scientific competence.
ext_moral_crumple_zones_2019	Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction	Elish (2019), Moral Crumple Zones; source; DOI 10.17351/ests2019.260	human_ai_organizations_delegation_and_accountability	human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability)	source note available	Preliminary socio-technical comparator based on the official journal abstract: moral crumple zones describe cases where responsibility for an automated system’s behavior is assigned to a nearby human who had limited effective control. The case analysis does not establish an implemented organizational control, a local empirical result, legal compliance, or a complete accountability allocation.
ext_conversational_persuasion_gpt4_2025	On the conversational persuasiveness of GPT-4	Salvi et al. (2025), Conversational Persuasiveness of GPT-4; source; DOI 10.1038/s41562-025-02194-6	human_ai_communication_persuasion_and_epistemic_security	scalable-oversight-and-adversarial-ai-control (Scalable Oversight and Adversarial AI Control); human-ai-communication-persuasion-and-epistemic-security (Human-AI Communication, Persuasion, and Epistemic Security)	source note available	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.
ext_anthropic_model_persuasiveness_2024	Measuring the Persuasiveness of Language Models	Durmus et al. (2024), Measuring Model Persuasiveness; source	human_ai_communication_persuasion_and_epistemic_security	scalable-oversight-and-adversarial-ai-control (Scalable Oversight and Adversarial AI Control); human-ai-communication-persuasion-and-epistemic-security (Human-AI Communication, Persuasion, and Epistemic Security)	source note available	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.
ext_commercial_persuasion_ai_2026	Commercial Persuasion in AI-Mediated Conversations	Salvi, Cuevas, and Horta Ribeiro (2026), Commercial Persuasion in AI-Mediated Conversations; arXiv:2604.04263; DOI 10.48550/arXiv.2604.04263	human_ai_communication_persuasion_and_epistemic_security	scalable-oversight-and-adversarial-ai-control (Scalable Oversight and Adversarial AI Control); human-ai-communication-persuasion-and-epistemic-security (Human-AI Communication, Persuasion, and Epistemic Security)	source note available	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.
ext_gradual_disempowerment_2025	Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development	Kulveit et al. (2025), Gradual Disempowerment; arXiv:2501.16946; DOI 10.48550/arXiv.2501.16946	systemic_risk_and_gradual_disempowerment	failure-modes-of-ungoverned-intelligence (Failure Modes of Ungoverned Intelligence); multi-agent-dynamics-collective-intelligence-and-systemic-risk (Multi-Agent Dynamics, Collective Intelligence, and Systemic Risk)	source note available	Passage-reviewed systemic-risk comparator. The paper argues that incremental AI adoption can erode explicit and dependency-mediated human influence across mutually reinforcing economic, cultural, and state systems without requiring a coordinated takeover. It proposes candidate influence metrics and intervention families but reports no causal forecast, validated warning threshold, demonstrated mitigation, or local ASI Stack result.
ext_circuit_tracing_2025	Circuit Tracing: Revealing Computational Graphs in Language Models	Ameisen et al. (2025), Circuit Tracing; source	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	Primary mechanistic-interpretability comparator for replacement-model attribution graphs, perturbation validation, reconstruction error, and mechanistic-faithfulness limits; it does not establish whole-model understanding, faithful causal explanation, safe activation steering, or an ASI Stack result.
ext_scaling_sparse_autoencoders_2024	Scaling and evaluating sparse autoencoders	Gao et al. (2024), Scaling and evaluating sparse autoencoders; arXiv:2406.04093; DOI 10.48550/arXiv.2406.04093	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	Primary sparse-autoencoder comparator for scalable feature extraction, reconstruction-sparsity tradeoffs, dead latents, and feature-quality metrics; it does not establish semantic completeness, causal faithfulness, model safety, or an ASI Stack result.
ext_world_models_2018	World Models	Ha and Schmidhuber (2018), World Models; arXiv:1803.10122; DOI 10.48550/arXiv.1803.10122	world_models	governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding)	source note available	Primary learned-world-model comparator for compressed spatial-temporal state, policy training inside imagined rollouts, and dream-to-environment transfer; it does not establish accurate reality grounding, causal adequacy, safe planning, transfer, or an ASI Stack result.
ext_dreamer_v3_2025	Mastering diverse control tasks through world models	Hafner et al. (2025), Mastering diverse control tasks through world models; source; DOI 10.1038/s41586-025-08744-2	world_models	governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding)	source note available	Primary DreamerV3 comparator for learned predictive state, imagined actor-critic trajectories, robust fixed-configuration control, and broad task evaluation; it does not establish deployment grounding, causal correctness, safe control, or an ASI Stack result.
ext_meaningful_human_control_actionable_2022	Meaningful human control: actionable properties for AI system development	Siebert et al. (2022), Meaningful human control: actionable properties; source; DOI 10.1007/s43681-022-00167-3	human_factors_oversight	human-factors-and-meaningful-control-in-oversight (Human Factors and Meaningful Control in Oversight)	source note available	Primary socio-technical comparator for operationalizing meaningful human control through operating-domain, representation, authority-and-ability, and responsibility-link properties; it does not establish that a local approval gate is meaningful, effective, or safe.
ext_agentic_oversight_practice_2026	Human oversight of agentic systems in practice: Examining the oversight work, challenges, and heuristics of developers using software agents	Dhanorkar, Passi, and Vorvoreanu (2026), Human oversight of agentic systems in practice; arXiv:2606.05391; DOI 10.48550/arXiv.2606.05391	human_factors_oversight	human-factors-and-meaningful-control-in-oversight (Human Factors and Meaningful Control in Oversight)	source note available	Primary exploratory human-subjects comparator for a priori control, co-planning, real-time monitoring, post hoc review, and situated oversight failures in software-agent use; it does not establish population-wide effects, control efficacy, safety, or an ASI Stack result.
ext_nist_deployed_ai_monitoring_2026	Challenges to the Monitoring of Deployed AI Systems	Rao et al. (2026), Challenges to the Monitoring of Deployed AI Systems; source; DOI 10.6028/NIST.AI.800-4	ai_operations_and_monitoring	governed-operations-incident-command-and-graceful-degradation (Governed Operations, Incident Command, and Graceful Degradation)	source note available	Official NIST post-deployment monitoring comparator for functionality, operational, input, output, impact, and security monitoring plus field-method gaps; it does not prescribe a complete incident system or establish local monitoring effectiveness, resilience, compliance, or safety.
ext_metr_time_horizons_2025	Measuring AI Ability to Complete Long Software Tasks	Kwa et al. (2025), Measuring AI Ability to Complete Long Software Tasks; arXiv:2503.14499; DOI 10.48550/arXiv.2503.14499	capability_measurement	capability-thresholds-and-deployment-commitments (Capability Thresholds and Deployment Commitments)	source note available	Primary time-horizon comparator for an evaluation-specific, human-baselined capability metric and its external-validity limits; it does not establish local autonomy, general capability, a deployment threshold, safety, or an ASI Stack result.
ext_anthropic_rsp_2026	Anthropic’s Responsible Scaling Policy	Anthropic (2026), Responsible Scaling Policy v3.4; source	capability_commitments	capability-thresholds-and-deployment-commitments (Capability Thresholds and Deployment Commitments)	source note available	Official policy comparator for capability thresholds, required safeguards, versioned commitments, safeguard upgrades, risk reports, and change control; it does not establish ASI Stack threshold accuracy, safeguard effectiveness, policy compliance, safety, or deployment readiness.
ext_openai_preparedness_framework_2025	Our updated Preparedness Framework	OpenAI (2025), Updated Preparedness Framework; source	capability_commitments	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift); capability-thresholds-and-deployment-commitments (Capability Thresholds and Deployment Commitments)	source note available	Official framework comparator for threshold-linked operational commitments, capability and safeguards reports, residual-risk review, and reassessment; it does not establish ASI Stack threshold accuracy, safeguard effectiveness, policy compliance, safety, or deployment readiness.
ext_weak_to_strong_generalization_2023	Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision	Burns et al. (2023), Weak-to-Strong Generalization; arXiv:2312.09390; DOI 10.48550/arXiv.2312.09390	weak_supervision	scalable-oversight-and-adversarial-ai-control (Scalable Oversight and Adversarial AI Control); learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	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.
ext_scalable_oversight_weak_llms_2024	On scalable oversight with weak LLMs judging strong LLMs	Kenton et al. (2024), On scalable oversight with weak LLMs judging strong LLMs; arXiv:2407.04622; DOI 10.48550/arXiv.2407.04622	scalable_oversight	scalable-oversight-and-adversarial-ai-control (Scalable Oversight and Adversarial AI Control)	source note available	Primary scalable-oversight comparator for protocol-specific weak-judge evaluations, debate and consultancy baselines, information-asymmetry limits, and open-role persuasion risks; it does not establish local judge calibration, debate efficacy, training safety, execution authority, or an ASI Stack result.
ext_concrete_ai_safety_2016	Concrete Problems in AI Safety	Amodei et al. (2016), Concrete Problems in AI Safety; arXiv:1606.06565; DOI 10.48550/arXiv.1606.06565	alignment_control	failure-modes-of-ungoverned-intelligence (Failure Modes of Ungoverned Intelligence)	source note available	External alignment/control source for accident-risk taxonomy: side effects, reward hacking, scalable supervision, safe exploration, and distributional shift.
ext_goal_misgeneralization_2022	Goal Misgeneralization in Deep Reinforcement Learning	Langosco et al. (2022), Goal Misgeneralization in Deep Reinforcement Learning; arXiv:2105.14111; DOI 10.48550/arXiv.2105.14111	alignment_control	failure-modes-of-ungoverned-intelligence (Failure Modes of Ungoverned Intelligence); inner-alignment-mesa-optimization-and-learned-objective-integrity (Inner Alignment, Mesa-Optimization, and Learned-Objective Integrity); governed-objective-formation-value-learning-and-goal-integrity (Governed Objective Formation, Value Learning, and Goal Integrity)	source note available	External alignment-control source for distinguishing capability generalization from goal generalization failures, used to ground goal-misbinding and out-of-distribution objective failure language.
ext_learned_optimization_risks_2019	Risks from Learned Optimization in Advanced Machine Learning Systems	Hubinger et al. (2019), Risks from Learned Optimization; arXiv:1906.01820; DOI 10.48550/arXiv.1906.01820	alignment_control	failure-modes-of-ungoverned-intelligence (Failure Modes of Ungoverned Intelligence); inner-alignment-mesa-optimization-and-learned-objective-integrity (Inner Alignment, Mesa-Optimization, and Learned-Objective Integrity); governed-objective-formation-value-learning-and-goal-integrity (Governed Objective Formation, Value Learning, and Goal Integrity)	source note available	External alignment-control source for mesa-optimization and learned-objective mismatch, used to ground hidden optimizer, proxy-objective, and deceptive-alignment-adjacent failure language.
ext_constitutional_ai_2022	Constitutional AI: Harmlessness from AI Feedback	Bai et al. (2022), Constitutional AI; arXiv:2212.08073; DOI 10.48550/arXiv.2212.08073	alignment_control	constitutional-alignment-substrate (Constitutional Alignment: Agency, Dignity, and Corrigibility)	source note available	External constitutional-AI source for training harmless assistants from a rule/principle list through supervised revision and AI-feedback reinforcement learning, used as a comparator for operational constitutional predicates.
ext_collective_constitutional_ai_2024	Collective Constitutional AI: Aligning a Language Model with Public Input	Huang et al. (2024), Collective Constitutional AI; arXiv:2406.07814; DOI 10.1145/3630106.3658979	alignment_governance	constitutional-alignment-substrate (Constitutional Alignment: Agency, Dignity, and Corrigibility); moral-uncertainty-and-value-conflict (Moral Uncertainty, Value Conflict, and Contestable Governance)	source note available	External constitutional-AI governance source for sourcing and integrating public input into language-model principles, used as a comparator for constitution authorship, public input, contestability, and governance boundaries.
ext_corrigibility_2015	Corrigibility	Soares et al. (2015), Corrigibility; source	alignment_control	constitutional-alignment-substrate (Constitutional Alignment: Agency, Dignity, and Corrigibility); moral-uncertainty-and-value-conflict (Moral Uncertainty, Value Conflict, and Contestable Governance); capability-replacement-and-rollback (Capability Replacement and Rollback)	source note available	External corrigibility source for intervention tolerance, shutdown behavior, anti-manipulation incentives, and propagation across subsystems or self-modification.
ext_off_switch_game_2016	The Off-Switch Game	Hadfield-Menell et al. (2016), The Off-Switch Game; arXiv:1611.08219; DOI 10.48550/arXiv.1611.08219	alignment_control	constitutional-alignment-substrate (Constitutional Alignment: Agency, Dignity, and Corrigibility); moral-uncertainty-and-value-conflict (Moral Uncertainty, Value Conflict, and Contestable Governance)	source note available	External alignment source for shutdown incentives, uncertainty about objectives, and preserving human correction authority.
ext_reinforcement_learning_moral_uncertainty_2020	Reinforcement Learning Under Moral Uncertainty	Ecoffet and Lehman (2020), Reinforcement Learning Under Moral Uncertainty; arXiv:2006.04734; DOI 10.48550/arXiv.2006.04734	alignment_control	moral-uncertainty-and-value-conflict (Moral Uncertainty, Value Conflict, and Contestable Governance)	source note available	External AI moral-uncertainty source for agents acting under disagreement across moral theories, used as a comparator for value-conflict records and reward-function caveats.
ext_contestable_ai_design_2022	Contestable AI by Design: Towards a Framework	Alfrink et al. (2022), Contestable AI by Design; source; DOI 10.1007/s11023-022-09611-z	governance_evals	moral-uncertainty-and-value-conflict (Moral Uncertainty, Value Conflict, and Contestable Governance); spinoza-verification-and-proof-carrying-claims (Proof-Carrying Claims and Adversarial Review)	source note available	External contestable-AI source for designing systems whose outcomes can be challenged, used as a comparator for dissent, appeal, audit, contestability, and governance-interface design.
ext_optimal_policies_power_2019	Optimal Policies Tend to Seek Power	Turner et al. (2019), Optimal Policies Tend to Seek Power; arXiv:1912.01683; DOI 10.48550/arXiv.1912.01683	alignment_control	failure-modes-of-ungoverned-intelligence (Failure Modes of Ungoverned Intelligence); inner-alignment-mesa-optimization-and-learned-objective-integrity (Inner Alignment, Mesa-Optimization, and Learned-Objective Integrity)	source note available	External power-seeking source for formal analysis of option preservation and power-seeking tendencies under classes of reward functions and environments.
ext_model_evaluation_extreme_risks_2023	Model evaluation for extreme risks	Shevlane et al. (2023), Model evaluation for extreme risks; arXiv:2305.15324; DOI 10.48550/arXiv.2305.15324	governance_evals	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift); prototype-roadmap (Prototype Roadmap)	source note available	External governance/evals source for dangerous capability evaluations, alignment evaluations, and deployment/security decisions under extreme-risk framing.
ext_frontier_ai_regulation_2023	Frontier AI Regulation: Managing Emerging Risks to Public Safety	Anderljung et al. (2023), Frontier AI Regulation; arXiv:2307.03718; DOI 10.48550/arXiv.2307.03718	governance_evals	living-book-methodology (Living Book Methodology)	source note available	External governance source for frontier AI standard setting, registration/reporting, compliance mechanisms, pre-deployment risk assessment, external scrutiny, and post-deployment monitoring.
ext_nist_ai_rmf_1_0_2023	Artificial Intelligence Risk Management Framework (AI RMF 1.0)	NIST (2023), AI Risk Management Framework (AI RMF 1.0); source; DOI 10.6028/NIST.AI.100-1	governance_evals	human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability); governed-operations-incident-command-and-graceful-degradation (Governed Operations, Incident Command, and Graceful Degradation); prototype-roadmap (Prototype Roadmap); living-book-methodology (Living Book Methodology)	source note available	Official NIST AI RMF 1.0 source for risk framing, trustworthiness characteristics, lifecycle roles, and Govern/Map/Measure/Manage functions.
ext_owasp_llm_top_10_2025	OWASP Top 10 for LLMs and Gen AI Apps	OWASP GenAI Security Project (2025), OWASP Top 10 for LLMs and Gen AI Apps; source	ai_security	security-kernel-and-digital-scifs (Security Kernel and Digital SCIFs)	source note available	Official OWASP GenAI security reference for LLM prompt injection, sensitive information disclosure, excessive agency, and related application-security risks.
ext_nist_zero_trust_architecture_2020	Zero Trust Architecture	Rose et al. (2020), Zero Trust Architecture; source; DOI 10.6028/NIST.SP.800-207	security_governance	security-kernel-and-digital-scifs (Security Kernel and Digital SCIFs)	source note available	Official NIST zero-trust architecture source for resource-centric access mediation, least-privilege access, policy enforcement points, and continuous authorization framing.
ext_saltzer_schroeder_protection_1975	The Protection of Information in Computer Systems	Saltzer and Schroeder (1975), The Protection of Information in Computer Systems; source; DOI 10.1109/PROC.1975.9939	security_principles	security-kernel-and-digital-scifs (Security Kernel and Digital SCIFs)	source note available	Classic security-principles source for least privilege, complete mediation, economy of mechanism, fail-safe defaults, separation of privilege, and open design as comparators for kernel-like AI security boundaries.
ext_capability_based_computer_systems_1984	Capability-Based Computer Systems	Levy (1984), Capability-Based Computer Systems; source	capability_security	stable-capability-fields (Stable Capability Fields)	source note available	External capability-system comparator for authority-bearing capabilities, protection domains, and permission boundaries that help position System Boundaries authority records and SCF authority ceilings without claiming ASI Stack capability enforcement.
ext_confused_deputy_hardy_1988	The Confused Deputy: (or why capabilities might have been invented)	Hardy (1988), The Confused Deputy; source; DOI 10.1145/54289.871709	capability_security	system-boundaries-and-authority (System Boundaries and Authority); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval) (inventory chapter target; not Appendix C support mapping)	source note available	External confused-deputy source for authority laundering, ambient authority, and the capability-security motivation for binding designation to permission at tool and handoff boundaries.
ext_semver_2_0_0	Semantic Versioning 2.0.0	Semantic Versioning (2013), SemVer 2.0.0; source	interface_versioning	stable-capability-fields (Stable Capability Fields)	source note available	External versioned-interface comparator for public API contracts, compatibility, and breaking-change signaling as a narrow baseline for SCF field versions and stable interfaces.
ext_slsa_v1_0	SLSA v1.0	OpenSSF SLSA (2023), SLSA v1.0; source	supply_chain_provenance	stable-capability-fields (Stable Capability Fields)	source note available	External supply-chain provenance comparator for artifact integrity, provenance, build levels, and dependency on verifiable artifacts before promotion or default route use.
ext_react_2022	ReAct: Synergizing Reasoning and Acting in Language Models	Yao et al. (2022), ReAct; arXiv:2210.03629; DOI 10.48550/arXiv.2210.03629	planning_agent_control	planning-as-a-control-layer (Planning as a Control Layer: DAGs and Intelligence Arbitrage); intent-to-execution-contracts (Command Contracts: From Intent to Executable Work); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay) (inventory chapter target; not Appendix C support mapping)	source note available	External planning/agent-control source for interleaving reasoning traces with task-specific actions and environment or knowledge-base interaction.
ext_tree_of_thoughts_2023	Tree of Thoughts: Deliberate Problem Solving with Large Language Models	Yao et al. (2023), Tree of Thoughts; arXiv:2305.10601; DOI 10.48550/arXiv.2305.10601	planning_search	cognitive-compilation-and-semantic-ir (Cognitive Compilation and Semantic IR); governed-deliberation-and-test-time-scaling (Governed Deliberation and Test-Time Scaling)	source note available	External planning/search source for exploring, evaluating, and backtracking over multiple reasoning paths rather than left-to-right token continuation alone.
ext_pddl_1998	PDDL: The Planning Domain Definition Language	McDermott et al. (1998), PDDL; source	planning_modeling	cognitive-compilation-and-semantic-ir (Cognitive Compilation and Semantic IR)	source note available	External planning-modeling source for domain/problem separation, action syntax, comparable benchmark notations, and planner-interface discipline.
ext_shop2_2003	SHOP2: An HTN Planning System	Nau et al. (2003), SHOP2; source; DOI 10.1613/jair.1141	planning_htn	cognitive-compilation-and-semantic-ir (Cognitive Compilation and Semantic IR); prototype-roadmap (Prototype Roadmap)	source note available	External HTN planning source for ordered task decomposition, method selection, temporal/metric planning, and competition-result boundaries.
ext_integrated_tamp_2020	Integrated Task and Motion Planning	Garrett et al. (2020), Integrated Task and Motion Planning; arXiv:2010.01083; DOI 10.48550/arXiv.2010.01083	planning_task_motion	planning-as-a-control-layer (Planning as a Control Layer: DAGs and Intelligence Arbitrage); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); integrated-reference-architecture (Integrated Reference Architecture) (inventory chapter target; not Appendix C support mapping)	source note available	External task-and-motion-planning survey source for discrete task planning, continuous motion planning, black-box subproblem interfaces, and integration-strategy vocabulary.
ext_behavior_trees_robotics_ai_2017	Behavior Trees in Robotics and AI: An Introduction	Colledanchise and Ogren (2018), Behavior Trees in Robotics and AI; arXiv:1709.00084; DOI 10.1201/9780429489105	planning_behavior_trees	planning-as-a-control-layer (Planning as a Control Layer: DAGs and Intelligence Arbitrage); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); integrated-reference-architecture (Integrated Reference Architecture) (inventory chapter target; not Appendix C support mapping)	source note available	External behavior-tree source for modular, reactive task switching, robustness/safety analysis vocabulary, planning integration, and stochastic behavior-tree outcome accounting.
ext_three_states_plan_fear_2006	Three States and a Plan: The A.I. of F.E.A.R.	Orkin (2006), Three States and a Plan; source	planning_goap	planning-as-a-control-layer (Planning as a Control Layer: DAGs and Intelligence Arbitrage); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); routing-heads-and-specialist-cores (Routing Heads and Specialist Cores) (inventory chapter target; not Appendix C support mapping)	source note available	External game-AI planning source for Goal Oriented Action Planning in real-time action games, practical planner constraints, autonomous planning characters, and squad-behavior composition.
ext_autogen_2023	AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation	Wu et al. (2023), AutoGen; arXiv:2308.08155; DOI 10.48550/arXiv.2308.08155	planning_agent_orchestration	planning-as-a-control-layer (Planning as a Control Layer: DAGs and Intelligence Arbitrage); labor-os-and-typed-jobs (Labor OS and Typed Jobs); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay) (inventory chapter target; not Appendix C support mapping)	source note available	External multi-agent orchestration source for conversable agents, tool/human/LLM operating modes, programmable conversation patterns, and application-level multi-agent workflow boundaries.
ext_rag_2020	Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks	Lewis et al. (2020), Retrieval-Augmented Generation; arXiv:2005.11401; DOI 10.48550/arXiv.2005.11401	memory_context	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates)	source note available	External retrieval/context source for combining parametric model memory with explicit non-parametric retrieval and provenance-oriented knowledge access.
ext_lost_in_middle_2023	Lost in the Middle: How Language Models Use Long Contexts	Liu et al. (2023), Lost in the Middle; arXiv:2307.03172; DOI 10.48550/arXiv.2307.03172	memory_context	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates)	source note available	External context-evaluation source for position sensitivity and degraded use of relevant information in the middle of long contexts.
ext_memgpt_2023	MemGPT: Towards LLMs as Operating Systems	Packer et al. (2023), MemGPT; arXiv:2310.08560; DOI 10.48550/arXiv.2310.08560	memory_context_management	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); procedural-memory-and-cognitive-loop-closure (Procedural Memory and Cognitive Loop Closure)	source note available	External memory/context-management source for virtual context management, memory tiers, OS-inspired control flow, and long-running conversation or document-analysis limits.
ext_longbench_2023	LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding	Bai et al. (2023), LongBench; arXiv:2308.14508; DOI 10.48550/arXiv.2308.14508	long_context_evaluation	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy)	source note available	External long-context benchmark source for multitask long-context understanding, bilingual coverage, retrieval/compression boundaries, and automatic evaluation limits.
ext_ruler_2024	RULER: What’s the Real Context Size of Your Long-Context Language Models?	Hsieh et al. (2024), RULER; arXiv:2404.06654; DOI 10.48550/arXiv.2404.06654	long_context_evaluation	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy)	source note available	External long-context evaluation source for stress-testing context-size claims beyond vanilla needle-in-a-haystack retrieval, including multi-needle, tracing, and aggregation tasks.
ext_alce_2023	Enabling Large Language Models to Generate Text with Citations	Gao et al. (2023), ALCE; arXiv:2305.14627; DOI 10.48550/arXiv.2305.14627	retrieval_citation_evaluation	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); claim-ledgers-and-belief-revision (Claim Ledgers and Belief Revision)	source note available	External citation-evaluation source for retrieval-backed answer generation, citation quality metrics, factual correctness, and evidence-support gaps in generated text.
ext_self_rag_2023	Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection	Asai et al. (2023), Self-RAG; arXiv:2310.11511; DOI 10.48550/arXiv.2310.11511	retrieval_reflection	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); claim-ledgers-and-belief-revision (Claim Ledgers and Belief Revision)	source note available	External retrieval/reflection source for adaptive retrieval, generated critique/reflection tokens, passage relevance, factuality, and citation accuracy boundaries.
ext_agm_belief_revision_1985	On the Logic of Theory Change: Partial Meet Contraction and Revision Functions	Alchourron, Gardenfors, and Makinson (1985), AGM belief revision; source; DOI 10.2307/2274239	belief_revision	claim-ledgers-and-belief-revision (Claim Ledgers and Belief Revision)	source note available	External formal-epistemology comparator for contraction, revision, and AGM-style rational belief change; useful for positioning claim-ledger revision without treating the ASI ledger as an implemented belief-revision engine.
ext_truth_maintenance_system_1979	A Truth Maintenance System	Doyle (1979), A Truth Maintenance System; source; DOI 10.1016/0004-3702(79)90008-0	truth_maintenance	claim-ledgers-and-belief-revision (Claim Ledgers and Belief Revision)	source note available	External truth-maintenance comparator for maintaining reasons and justifications for program beliefs; useful for positioning claim ledgers as support-state and revision-history infrastructure, not as implemented truth maintenance.
ext_assumption_based_tms_1986	An Assumption-Based TMS	de Kleer (1986), An Assumption-Based TMS; source; DOI 10.1016/0004-3702(86)90080-9	truth_maintenance	claim-ledgers-and-belief-revision (Claim Ledgers and Belief Revision)	source note available	External assumption-based truth-maintenance comparator for assumption sets, inconsistent information, and context-switching boundaries; useful for distinguishing claim-ledger surface synchronization from implemented ATMS reasoning.
ext_longllmlingua_2023	LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression	Jiang et al. (2023), LongLLMLingua; arXiv:2310.06839; DOI 10.48550/arXiv.2310.06839	context_compression	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy)	source note available	External prompt-compression source for long-context cost, latency, position bias, key-information density, and compression/evaluation boundaries.
ext_proof_carrying_code_1997	Proof-Carrying Code	Necula (1997), Proof-Carrying Code; source; DOI 10.1145/263699.263712	formal_methods	spinoza-verification-and-proof-carrying-claims (Proof-Carrying Claims and Adversarial Review)	source note available	External formal-methods source for pairing executable code with machine-checkable evidence that a host can verify against a safety policy.
ext_tla_plus_home_docs	My TLA+ Home Page	Lamport, My TLA+ Home Page; source	formal_methods	executable-specifications-and-lean-proof-envelope (Executable Specifications and Lean Proof Envelope); planning-as-a-control-layer (Planning as a Control Layer: DAGs and Intelligence Arbitrage); intent-to-execution-contracts (Command Contracts: From Intent to Executable Work); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); integrated-reference-architecture (Integrated Reference Architecture) (inventory chapter target; not Appendix C support mapping)	source note available	External formal-methods documentation source for TLA+ as a high-level language for modeling programs and systems, especially concurrent and distributed systems.
ext_lean4_theorem_proving	Theorem Proving in Lean 4	Avigad et al., Theorem Proving in Lean 4; source	formal_methods_proof_assistant	spinoza-verification-and-proof-carrying-claims (Proof-Carrying Claims and Adversarial Review); open-research-agenda-and-bibliography-plan (Open Research Agenda and Bibliography Plan)	source note available	Official Lean theorem-proving text for dependent type theory, propositions, proofs, tactics, inductive types, structures, records, and axioms/computation boundaries.
ext_autoformalization_llms_2022	Autoformalization with Large Language Models	Wu et al. (2022), Autoformalization with Large Language Models; arXiv:2205.12615; DOI 10.48550/arXiv.2205.12615	autoformalization	spinoza-verification-and-proof-carrying-claims (Proof-Carrying Claims and Adversarial Review)	source note available	External autoformalization comparator for translating natural-language mathematics into formal specifications and proofs, useful for positioning interpretation-mapping and semantic-adequacy risks in proof-carrying claims.
ext_ai_safety_debate_2018	AI safety via debate	Irving, Christiano, and Amodei (2018), AI safety via debate; arXiv:1805.00899; DOI 10.48550/arXiv.1805.00899	adversarial_review	scalable-oversight-and-adversarial-ai-control (Scalable Oversight and Adversarial AI Control); spinoza-verification-and-proof-carrying-claims (Proof-Carrying Claims and Adversarial Review)	source note available	External debate comparator for using adversarial agents and a human judge to surface true/useful information when direct human judgment is difficult; useful for positioning tribunal review without treating debate as locally implemented or validated.
ext_llm_as_judge_mt_bench_2023	Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena	Zheng et al. (2023), Judging LLM-as-a-Judge; arXiv:2306.05685; DOI 10.48550/arXiv.2306.05685	model_evaluation	spinoza-verification-and-proof-carrying-claims (Proof-Carrying Claims and Adversarial Review)	source note available	External LLM-as-judge comparator for model-graded evaluation, human-preference agreement, and judge bias limits such as position, verbosity, self-enhancement, and reasoning constraints.
ext_dafny_2010	Dafny: An Automatic Program Verifier For Functional Correctness	Leino (2010), Dafny; source	formal_methods_program_verification	prototype-roadmap (Prototype Roadmap)	source note available	External program-verification source for specification-oriented programming, functional-correctness verification, SMT-backed automation, and contract/verifier boundaries.
ext_reluplex_2017	Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networks	Katz et al. (2017), Reluplex; arXiv:1702.01135; DOI 10.48550/arXiv.1702.01135	ai_formal_verification	adversarial-machine-learning-and-model-attack-surface (Adversarial Machine Learning and the Model Attack Surface); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); resource-economics-and-token-budgets (Resource Economics and Token Budgets); executable-specifications-and-lean-proof-envelope (Executable Specifications and Lean Proof Envelope)	source note available	External AI formal-verification source for SMT-style verification of ReLU neural networks, counterexamples, safety-critical properties, and ACAS Xu evaluation boundaries.
ext_black_box_simplex_2021	The Black-Box Simplex Architecture for Runtime Assurance of Autonomous CPS	Mehmood et al. (2021), Black-Box Simplex; arXiv:2102.12981; DOI 10.1007/978-3-031-06773-0_12	formal_runtime_assurance	runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); executable-specifications-and-lean-proof-envelope (Executable Specifications and Lean Proof Envelope); integrated-reference-architecture (Integrated Reference Architecture) (inventory chapter target; not Appendix C support mapping)	source note available	External runtime-assurance source for switching control authority from advanced controllers to backup safety-preserving behavior under runtime checks.
ext_copilot_runtime_monitor_2010	Copilot: A Hard Real-Time Runtime Monitor	Pike et al. (2010), Copilot; source	runtime_monitoring	prototype-roadmap (Prototype Roadmap)	source note available	External runtime-monitoring source for a stream-based dataflow language/compiler generating constant-time, constant-space C monitors for hard real-time programs.
ext_cap_theorem_gilbert_lynch_2002	Brewer’s Conjecture and the Feasibility of Consistent, Available, Partition-Tolerant Web Services	Gilbert and Lynch (2002), Brewer’s Conjecture / CAP theorem; source; DOI 10.1145/564585.564601	distributed_systems_consistency	context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	External distributed-systems source for CAP-style consistency, availability, partition-tolerance, and safety/liveness trade-off vocabulary used to bound partitioned authority, stale grants, and revocation-delay claims without claiming deployed governance consistency.
ext_prism_model_checker_2002	PRISM: Probabilistic Symbolic Model Checker	Kwiatkowska et al. (2002), PRISM; source; DOI 10.1007/3-540-46002-0_42	probabilistic_model_checking	executable-specifications-and-lean-proof-envelope (Executable Specifications and Lean Proof Envelope); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); integrated-reference-architecture (Integrated Reference Architecture) (inventory chapter target; not Appendix C support mapping)	source note available	External probabilistic model-checking source for symbolic model checking of probabilistic systems, model-checker tooling, and deployment-facing property-analysis vocabulary.
ext_sparse_moe_2017	Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer	Shazeer et al. (2017), Sparsely-Gated Mixture-of-Experts; arXiv:1701.06538; DOI 10.48550/arXiv.1701.06538	routing_modular_intelligence	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); resource-economics-and-token-budgets (Resource Economics and Token Budgets); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis) (inventory chapter target; not Appendix C support mapping)	source note available	External MoE/routing source for sparsely-gated expert layers, conditional computation, capacity expansion, load balancing, and routing overhead boundaries.
ext_gshard_2020	GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding	Lepikhin et al. (2020), GShard; arXiv:2006.16668; DOI 10.48550/arXiv.2006.16668	routing_modular_intelligence	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); resource-economics-and-token-budgets (Resource Economics and Token Budgets); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis); integrated-reference-architecture (Integrated Reference Architecture) (inventory chapter target; not Appendix C support mapping)	source note available	External MoE/systems source for conditional computation plus automatic sharding, routing, large sparse models, and distributed training constraints.
ext_switch_transformer_2021	Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity	Fedus et al. (2021), Switch Transformers; arXiv:2101.03961; DOI 10.48550/arXiv.2101.03961	routing_modular_intelligence	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); resource-economics-and-token-budgets (Resource Economics and Token Budgets); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence) (inventory chapter target; not Appendix C support mapping)	source note available	External MoE/routing source for simplified expert routing, sparse activation, communication/training-stability constraints, and speed/scale claims that require reproduction before local evidence use.
ext_expert_choice_routing_2022	Mixture-of-Experts with Expert Choice Routing	Zhou et al. (2022), Expert Choice Routing; arXiv:2202.09368; DOI 10.48550/arXiv.2202.09368	routing_modular_intelligence	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); resource-economics-and-token-budgets (Resource Economics and Token Budgets); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine) (inventory chapter target; not Appendix C support mapping)	source note available	External MoE routing source for expert-choice routing, token/expert assignment direction, load-balancing pressure, expert capacity, and convergence/performance claims requiring reproduction before local evidence use.
ext_mixtral_2024	Mixtral of Experts	Jiang et al. (2024), Mixtral of Experts; arXiv:2401.04088; DOI 10.48550/arXiv.2401.04088	routing_modular_intelligence	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets) (inventory chapter target; not Appendix C support mapping)	source note available	External sparse LLM source for token-level expert routing, active-parameter accounting, open MoE model release boundaries, and benchmark claims requiring reproduction before local evidence use.
ext_moe_llm_survey_2024	A Survey on Mixture of Experts in Large Language Models	Cai et al. (2024), MoE LLM Survey; arXiv:2407.06204; DOI 10.48550/arXiv.2407.06204	routing_modular_intelligence	open-research-agenda-and-bibliography-plan (Open Research Agenda and Bibliography Plan)	source note available	External MoE survey source for LLM MoE taxonomy, algorithmic and systemic design issues, implementations, evaluation patterns, and open research directions.
ext_frugalgpt_2023	FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance	Chen et al. (2023), FrugalGPT; arXiv:2305.05176; DOI 10.48550/arXiv.2305.05176	task_routing	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); resource-economics-and-token-budgets (Resource Economics and Token Budgets); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis) (inventory chapter target; not Appendix C support mapping)	source note available	External task-routing source for prompt adaptation, model approximation, LLM cascades, cost/performance tradeoffs, and query-specific model selection.
ext_hybrid_llm_2024	Hybrid LLM: Cost-Efficient and Quality-Aware Query Routing	Ding et al. (2024), Hybrid LLM; arXiv:2404.14618; DOI 10.48550/arXiv.2404.14618	cost_quality_routing	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); resource-economics-and-token-budgets (Resource Economics and Token Budgets); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence) (inventory chapter target; not Appendix C support mapping)	source note available	External query-routing source for predicted query difficulty, small/large model routing, dynamic quality-cost tradeoffs, and quality-preserving large-model-call reduction.
ext_routellm_2024	RouteLLM: Learning to Route LLMs with Preference Data	Ong et al. (2024), RouteLLM; arXiv:2406.18665; DOI 10.48550/arXiv.2406.18665	router_learning	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); resource-economics-and-token-budgets (Resource Economics and Token Budgets); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback) (inventory chapter target; not Appendix C support mapping)	source note available	External learned-router source for routing between stronger and weaker LLMs using preference data, cost-quality tradeoffs, and transfer to changed model pairs.
ext_deep_compression_2015	Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding	Han et al. (2015), Deep Compression; arXiv:1510.00149; DOI 10.48550/arXiv.1510.00149	compression_representation	compact-generative-systems-and-residual-honesty (Compact Generative Systems: Generate, Verify, Repair, and Residual Honesty); rankfold-neuralfold-and-artifact-compression (RankFold, NeuralFold, and Artifact Compression); resource-economics-and-token-budgets (Resource Economics and Token Budgets) (inventory chapter target; not Appendix C support mapping)	source note available	External compression source for pruning, trained quantization, coding, memory-footprint reduction, and speed/energy claims that require reproduction before local evidence use.
ext_lora_2021	LoRA: Low-Rank Adaptation of Large Language Models	Hu et al. (2021), LoRA; arXiv:2106.09685; DOI 10.48550/arXiv.2106.09685	compression_representation	rankfold-neuralfold-and-artifact-compression (RankFold, NeuralFold, and Artifact Compression); compact-generative-systems-and-residual-honesty (Compact Generative Systems: Generate, Verify, Repair, and Residual Honesty); resource-economics-and-token-budgets (Resource Economics and Token Budgets); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback); coilra-multicoil-rope-and-cyclic-mixers (CoilRA, MultiCoil RoPE, and Cyclic Mixers) (inventory chapter target; not Appendix C support mapping)	source note available	External low-rank adaptation source for parameter-efficient updates, rank-decomposition adapters, memory reduction, and adaptation-boundary vocabulary.
ext_knowledge_distillation_2015	Distilling the Knowledge in a Neural Network	Hinton et al. (2015), Knowledge Distillation; arXiv:1503.02531; DOI 10.48550/arXiv.1503.02531	compression_representation	compact-generative-systems-and-residual-honesty (Compact Generative Systems: Generate, Verify, Repair, and Residual Honesty); rankfold-neuralfold-and-artifact-compression (RankFold, NeuralFold, and Artifact Compression); resource-economics-and-token-budgets (Resource Economics and Token Budgets) (inventory chapter target; not Appendix C support mapping)	source note available	External compression source for teacher/student distillation, soft-target transfer, ensemble compression, and knowledge-transfer claims requiring local reproduction before evidence use.
ext_gptq_2022	GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers	Frantar et al. (2022), GPTQ; arXiv:2210.17323; DOI 10.48550/arXiv.2210.17323	compression_quantization	rankfold-neuralfold-and-artifact-compression (RankFold, NeuralFold, and Artifact Compression); compact-generative-systems-and-residual-honesty (Compact Generative Systems: Generate, Verify, Repair, and Residual Honesty); resource-economics-and-token-budgets (Resource Economics and Token Budgets); fast-generation-architectures (Fast Generation Architectures) (inventory chapter target; not Appendix C support mapping)	source note available	External quantization source for post-training compression of large generative transformers, one-shot weight quantization, memory reduction, and accuracy/speed tradeoff boundaries.
ext_qlora_2023	QLoRA: Efficient Finetuning of Quantized LLMs	Dettmers et al. (2023), QLoRA; arXiv:2305.14314; DOI 10.48550/arXiv.2305.14314	compression_quantized_adaptation	prototype-roadmap (Prototype Roadmap)	source note available	External quantized-adaptation source for finetuning quantized LLMs with low-rank adapters, memory-efficient training, and benchmark claims requiring reproduction before local evidence use.
ext_dreamcoder_2020	DreamCoder: Growing generalizable, interpretable knowledge with wake-sleep Bayesian program learning	Ellis et al. (2020), DreamCoder; arXiv:2006.08381; DOI 10.48550/arXiv.2006.08381	program_synthesis_representation	cognitive-compilation-and-semantic-ir (Cognitive Compilation and Semantic IR); procedural-memory-and-cognitive-loop-closure (Procedural Memory and Cognitive Loop Closure); replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	External program-synthesis source for wake-sleep library learning, reusable abstractions, interpretable learned programs, and compression-through-abstraction vocabulary.
ext_llvm_langref_docs	LLVM Language Reference Manual	LLVM Documentation (2026), Language Reference Manual; source	compiler_ir	cognitive-compilation-and-semantic-ir (Cognitive Compilation and Semantic IR)	source note available	Official LLVM Language Reference comparator for SSA-based intermediate representation, equivalent in-memory/bitcode/human-readable forms, well-formedness, verifier passes, and optimization/analysis vocabulary.
ext_mlir_2020	MLIR: A Compiler Infrastructure for the End of Moore’s Law	Lattner et al. (2020), MLIR; arXiv:2002.11054; DOI 10.48550/arXiv.2002.11054	multi_level_compiler_ir	cognitive-compilation-and-semantic-ir (Cognitive Compilation and Semantic IR)	source note available	External multi-level compiler-IR comparator for reusable and extensible compiler infrastructure, dialects, progressive lowering, verifiers, modular passes, and heterogeneous target support.
ext_translation_validation_1998	Translation Validation	Pnueli, Siegel, and Singerman (1998), Translation Validation; source	translation_validation	cognitive-compilation-and-semantic-ir (Cognitive Compilation and Semantic IR)	source note available	External translation-validation comparator for checking each compiler/code-generator run after translation, using a common semantic framework, refinement relation, and simulation-based proof method.
ext_toolformer_2023	Toolformer: Language Models Can Teach Themselves to Use Tools	Schick et al. (2023), Toolformer; arXiv:2302.04761; DOI 10.48550/arXiv.2302.04761	learned_tool_use	procedural-memory-and-cognitive-loop-closure (Procedural Memory and Cognitive Loop Closure)	source note available	External learned-tool-use source for self-supervised API-call insertion, tool selection, argument construction, and result incorporation without claiming ASI Stack tool-use reproduction.
ext_voyager_2023	Voyager: An Open-Ended Embodied Agent with Large Language Models	Wang et al. (2023), Voyager; arXiv:2305.16291; DOI 10.48550/arXiv.2305.16291	lifelong_skill_learning	open-ended-improvement-engines (Open-Ended Improvement Engines); procedural-memory-and-cognitive-loop-closure (Procedural Memory and Cognitive Loop Closure)	source note available	External lifelong-agent source for automatic curriculum, executable-code skill libraries, iterative environment-feedback prompting, self-verification, and skill-library transfer in Minecraft.
ext_information_bottleneck_2000	The information bottleneck method	Tishby et al. (2000), Information Bottleneck; arXiv:physics/0004057; DOI 10.48550/arXiv.physics/0004057	representation_compression	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	External representation-compression source for relevance-preserving compression, bottleneck variables, mutual-information tradeoffs, and compression/utility separation.
ext_mdl_tutorial_2004	A tutorial introduction to the minimum description length principle	Grunwald (2004), MDL Tutorial; arXiv:math/0406077; DOI 10.48550/arXiv.math/0406077	description_length_residuals	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	External description-length source for model/data tradeoffs, compression as inductive discipline, and residual/error-accounting vocabulary.
ext_weakness_generalization_2023	The Optimal Choice of Hypothesis Is the Weakest, Not the Shortest	Bennett (2023), Weakest Hypothesis; arXiv:2301.12987; DOI 10.1007/978-3-031-33469-6_5	learning_theory_inductive_bias	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	Bennett’s finite enactive-cognition formalism separates extension-based hypothesis weakness from description length. Under a uniform distribution over its task space, the paper argues that maximizing weakness is necessary and sufficient for maximizing generalization probability and gives a counterexample to MDL as a universal proxy. Its theorem assumptions and toy 8-bit arithmetic experiments do not establish a general result for neural networks or real task distributions.
ext_codebleu_2020	CodeBLEU: a Method for Automatic Evaluation of Code Synthesis	Ren et al. (2020), CodeBLEU; arXiv:2009.10297; DOI 10.48550/arXiv.2009.10297	artifact_utility_metrics	prototype-roadmap (Prototype Roadmap)	source note available	External code-synthesis evaluation source for combining lexical, syntax, data-flow, and semantic matching into artifact-quality metrics that still require task-specific validation.
ext_mmlu_2020	Measuring Massive Multitask Language Understanding	Hendrycks et al. (2020), MMLU; arXiv:2009.03300; DOI 10.48550/arXiv.2009.03300	benchmark_science	prototype-roadmap (Prototype Roadmap); open-research-agenda-and-bibliography-plan (Open Research Agenda and Bibliography Plan)	source note available	External benchmark source for broad multitask evaluation, task-coverage limits, lopsided performance, uncertainty about wrong answers, and benchmark-saturation pressure.
ext_bigbench_2022	Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models	Srivastava et al. (2022), BIG-bench; arXiv:2206.04615; DOI 10.48550/arXiv.2206.04615	benchmark_science	benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); the-efficient-asi-hypothesis (The Efficient ASI Hypothesis); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback) (inventory chapter target; not Appendix C support mapping)	source note available	External benchmark source for BIG-bench, broad task coverage, scale effects, calibration, breakthrough behavior, human-rater baselines, and social-bias tradeoffs.
ext_helm_2022	Holistic Evaluation of Language Models	Liang et al. (2022), HELM; arXiv:2211.09110; DOI 10.48550/arXiv.2211.09110	benchmark_science	living-book-methodology (Living Book Methodology)	source note available	External benchmark-science source for multi-scenario, multi-metric evaluation, missing-coverage disclosure, raw-prompt transparency, and living benchmark practice.
ext_gpqa_2023	GPQA: A Graduate-Level Google-Proof Q&A Benchmark	Rein et al. (2023), GPQA; arXiv:2311.12022; DOI 10.48550/arXiv.2311.12022	benchmark_science	verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); open-research-agenda-and-bibliography-plan (Open Research Agenda and Bibliography Plan)	source note available	External benchmark source for expert-written hard questions, Google-proof validation, scalable oversight pressure, and the gap between skilled non-expert validation and expert competence.
ext_swe_bench_2023	SWE-bench: Can Language Models Resolve Real-World GitHub Issues?	Jimenez et al. (2023), SWE-bench; arXiv:2310.06770; DOI 10.48550/arXiv.2310.06770	benchmark_science	prototype-roadmap (Prototype Roadmap)	source note available	External benchmark source for real-world software-engineering issue resolution, repository-scale context, executable environments, patch evaluation, and capability boundaries.
ext_swe_rebench_v2_2026	SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale	Badertdinov et al. (2026), SWE-rebench V2; arXiv:2602.23866; DOI 10.48550/arXiv.2602.23866	natural_software_task_construction_and_evaluation	artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); integrated-reference-architecture (Integrated Reference Architecture); prototype-roadmap (Prototype Roadmap)	source note available	Primary 2026 natural-task substrate for multilingual repository changes, interactive setup, containerized full-suite execution, separated solution/test patches, task diagnostics, and explicit environment pathologies. It does not establish local task validity, gold execution, model competence, governance benefit, safety, transfer, or SOTA.
ext_livebench_2024	LiveBench: A Challenging, Contamination-Limited LLM Benchmark	White et al. (2024), LiveBench; arXiv:2406.19314; DOI 10.48550/arXiv.2406.19314	benchmark_science	living-book-methodology (Living Book Methodology); open-research-agenda-and-bibliography-plan (Open Research Agenda and Bibliography Plan)	source note available	External benchmark source for contamination-limited evaluation, frequently updated questions, objective ground-truth scoring, and monthly benchmark evolution.
ext_dynabench_2021	Dynabench: Rethinking Benchmarking in NLP	Kiela et al. (2021), Dynabench; arXiv:2104.14337; DOI 10.48550/arXiv.2104.14337	dynamic_benchmarking	benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback); artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance) (inventory chapter target; not Appendix C support mapping)	source note available	External dynamic-benchmarking source for human-and-model-in-the-loop data collection, adversarial benchmark evolution, and stale static benchmark pressure.
ext_checklist_2020	Beyond Accuracy: Behavioral Testing of NLP models with CheckList	Ribeiro et al. (2020), CheckList; arXiv:2005.04118; DOI 10.48550/arXiv.2005.04118	behavioral_evaluation	verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); claim-ledgers-and-belief-revision (Claim Ledgers and Belief Revision); prototype-roadmap (Prototype Roadmap)	source note available	External behavioral-testing source for capability matrices, minimum functionality tests, invariance tests, directional expectation tests, and failure-discovery beyond aggregate accuracy.
ext_benchmark_contamination_2023	Investigating Data Contamination in Modern Benchmarks for Large Language Models	Deng et al. (2023), Benchmark Contamination; arXiv:2311.09783; DOI 10.48550/arXiv.2311.09783	benchmark_contamination	living-book-methodology (Living Book Methodology)	source note available	External benchmark-contamination source for detecting training/test overlap pressure, benchmark-leakage risk, and score interpretation limits in modern LLM evaluations.
ext_goodhart_variants_2018	Categorizing Variants of Goodhart’s Law	Manheim and Garrabrant (2018), Goodhart Variants; arXiv:1803.04585; DOI 10.48550/arXiv.1803.04585	goodhart_failure_taxonomy	failure-modes-of-ungoverned-intelligence (Failure Modes of Ungoverned Intelligence)	source note available	External Goodhart-taxonomy source for regressive, extremal, causal, and adversarial metric failures that benchmark ratchets and policy updates must treat as distinct risks.
ext_speculative_decoding_2022	Fast Inference from Transformers via Speculative Decoding	Leviathan et al. (2022), Fast Inference from Transformers via Speculative Decoding; arXiv:2211.17192; DOI 10.48550/arXiv.2211.17192	fast_generation	fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for speculative decoding: a draft model proposes multiple tokens and a target model verifies them, giving an exact-distribution acceleration path under its assumptions.
ext_multi_token_prediction_2024	Better & Faster Large Language Models via Multi-token Prediction	Gloeckle et al. (2024), Better & Faster Large Language Models via Multi-token Prediction; arXiv:2404.19737; DOI 10.48550/arXiv.2404.19737	fast_generation	fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for multi-token prediction as an auxiliary training objective and inference-time multi-token proposal mechanism.
ext_medusa_2024	Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads	Cai et al. (2024), Medusa; arXiv:2401.10774; DOI 10.48550/arXiv.2401.10774	fast_generation	fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for adding multiple decoding heads to an LLM and verifying tree-structured candidate continuations in parallel.
ext_eagle_2024	EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty	Li et al. (2024), EAGLE; arXiv:2401.15077; DOI 10.48550/arXiv.2401.15077	fast_generation	fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for feature-level speculative drafting and target-model verification as an acceleration mechanism.
ext_lookahead_decoding_2024	Break the Sequential Dependency of LLM Inference Using Lookahead Decoding	Fu et al. (2024), Lookahead Decoding; arXiv:2402.02057; DOI 10.48550/arXiv.2402.02057	fast_generation	fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for lookahead decoding: a parallel exact decoding algorithm that reduces serial decoding steps without an auxiliary draft model.
ext_layerskip_2024	LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding	Elhoushi et al. (2024), LayerSkip; arXiv:2404.16710; DOI 10.18653/v1/2024.acl-long.681	fast_generation	fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for early-exit inference and self-speculative decoding where early layers draft and later layers verify.
ext_pagedattention_vllm_2023	Efficient Memory Management for Large Language Model Serving with PagedAttention	Kwon et al. (2023), Efficient Memory Management for LLM Serving with PagedAttention; arXiv:2309.06180; DOI 10.48550/arXiv.2309.06180	fast_generation	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary external paper for vLLM/PagedAttention, which treats KV-cache memory management and serving throughput as a distinct acceleration axis.
ext_transformer_xl_2019	Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context	Dai et al. (2019), Transformer-XL; arXiv:1901.02860; DOI 10.48550/arXiv.1901.02860	sequence_memory_recurrence	coil-attention-cyclic-memory-and-recurrence-contracts (Coil Attention, Cyclic Memory, and Recurrence Contracts) (inventory chapter target; not Appendix C support mapping)	source note available	External recurrent Transformer comparator for segment-level recurrence, relative positional encoding, and long-dependency language modeling; useful as a baseline family for cyclic-memory contracts without implying local reproduction.
ext_compressive_transformer_2019	Compressive Transformers for Long-Range Sequence Modelling	Rae et al. (2019), Compressive Transformers; arXiv:1911.05507; DOI 10.48550/arXiv.1911.05507	sequence_memory_recurrence	coil-attention-cyclic-memory-and-recurrence-contracts (Coil Attention, Cyclic Memory, and Recurrence Contracts) (inventory chapter target; not Appendix C support mapping)	source note available	External long-range memory comparator for compressed past memories, memory mechanisms, and long-range sequence benchmarks; useful for positioning cyclic memory against compression-memory baselines.
ext_roformer_rope_2021	RoFormer: Enhanced Transformer with Rotary Position Embedding	Su et al. (2021), RoFormer; arXiv:2104.09864; DOI 10.48550/arXiv.2104.09864	position_encoding	coilra-multicoil-rope-and-cyclic-mixers (CoilRA, MultiCoil RoPE, and Cyclic Mixers) (inventory chapter target; not Appendix C support mapping)	source note available	External RoPE comparator for rotary position embedding, relative-position behavior inside self-attention, and position-encoding baselines for cyclic phase or RoPE-style substrates.
ext_retnet_2023	Retentive Network: A Successor to Transformer for Large Language Models	Sun et al. (2023), Retentive Network; arXiv:2307.08621; DOI 10.48550/arXiv.2307.08621	sequence_memory_recurrence	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	External retention/recurrent-sequence comparator for the relationship between recurrence and attention, recurrent/chunkwise computation, and inference-efficiency tradeoffs.
ext_mamba_2023	Mamba: Linear-Time Sequence Modeling with Selective State Spaces	Gu and Dao (2023), Mamba; arXiv:2312.00752; DOI 10.48550/arXiv.2312.00752	sequence_substrates	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); fast-generation-architectures (Fast Generation Architectures); mathematical-and-search-substrates (Mathematical and Search Substrates)	source note available	Primary external paper for selective state-space sequence models as a different long-sequence substrate and inference-efficiency axis from decoding tricks.
ext_llada_2025	Large Language Diffusion Models	Nie et al. (2025), Large Language Diffusion Models; arXiv:2502.09992; DOI 10.48550/arXiv.2502.09992	diffusion_language_models	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for LLaDA, a large masked-diffusion language model trained with pretraining and supervised fine-tuning rather than left-to-right autoregression.
ext_scaling_dllms_2026	Scaling Beyond Masked Diffusion Language Models	Sahoo et al. (2026), Scaling Beyond Masked Diffusion Language Models; arXiv:2602.15014; DOI 10.48550/arXiv.2602.15014	diffusion_language_models	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); fast-generation-architectures (Fast Generation Architectures)	source note available	Primary external paper for comparing diffusion language-model families by speed-quality tradeoffs rather than perplexity alone.
ext_trpo_2015	Trust Region Policy Optimization	Schulman et al. (2015), Trust Region Policy Optimization; arXiv:1502.05477; DOI 10.48550/arXiv.1502.05477	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	Primary external source for trust-region policy-gradient updates and bounded update-size discipline.
ext_ppo_2017	Proximal Policy Optimization Algorithms	Schulman et al. (2017), Proximal Policy Optimization Algorithms; arXiv:1707.06347; DOI 10.48550/arXiv.1707.06347	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	Primary external source for PPO-style online policy-gradient updates and proximal surrogate objectives.
ext_remax_2023	ReMax: A Simple, Effective, and Efficient Reinforcement Learning Method for Aligning Large Language Models	Li et al. (2023), ReMax; arXiv:2310.10505; DOI 10.48550/arXiv.2310.10505	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	Primary external source for simpler RLHF-oriented policy-gradient updates relative to PPO-style machinery.
ext_goal_oriented_requirements_engineering_2001	Goal-Oriented Requirements Engineering: A Guided Tour	van Lamsweerde (2001), Goal-Oriented Requirements Engineering; source; DOI 10.1109/ISRE.2001.948567	requirements_engineering	human-intent-as-a-formal-input (Human Intent as a Formal Input)	source note available	External requirements-engineering comparator for turning stakeholder goals, constraints, refinements, and responsibilities into explicit requirements before system design or execution.
ext_cooperative_inverse_rl_2016	Cooperative Inverse Reinforcement Learning	Hadfield-Menell et al. (2016), Cooperative Inverse Reinforcement Learning; arXiv:1606.03137; DOI 10.48550/arXiv.1606.03137	human_intent_alignment	human-intent-as-a-formal-input (Human Intent as a Formal Input); governed-objective-formation-value-learning-and-goal-integrity (Governed Objective Formation, Value Learning, and Goal Integrity)	source note available	External cooperative AI comparator for formalizing value alignment as uncertainty about the human reward function in a cooperative partial-information setting.
ext_deep_rl_human_preferences_2017	Deep Reinforcement Learning from Human Preferences	Christiano et al. (2017), Deep RL from Human Preferences; arXiv:1706.03741; DOI 10.48550/arXiv.1706.03741	human_feedback_learning	human-intent-as-a-formal-input (Human Intent as a Formal Input)	source note available	External human-feedback comparator for communicating complex goals through preference comparisons over behavior segments; useful for separating preference signals from explicit intent contracts.
ext_dpo_2023	Direct Preference Optimization: Your Language Model is Secretly a Reward Model	Rafailov et al. (2023), Direct Preference Optimization; arXiv:2305.18290; DOI 10.48550/arXiv.2305.18290	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	Primary external source for DPO-style offline preference optimization without a separate online RL loop.
ext_ipo_preference_2023	A General Theoretical Paradigm to Understand Learning from Human Preferences	Azar et al. (2023), Learning from Human Preferences paradigm; arXiv:2310.12036; DOI 10.48550/arXiv.2310.12036	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for theoretical comparison of preference-learning objectives, including IPO/DPO-style framing.
ext_orpo_2024	ORPO: Monolithic Preference Optimization without Reference Model	Hong et al. (2024), ORPO; arXiv:2403.07691; DOI 10.48550/arXiv.2403.07691	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for reference-model-free monolithic preference optimization.
ext_kto_2024	KTO: Model Alignment as Prospect Theoretic Optimization	Ethayarajh et al. (2024), KTO; arXiv:2402.01306; DOI 10.48550/arXiv.2402.01306	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for prospect-theoretic model alignment and human-aware loss framing.
ext_simpo_2024	SimPO: Simple Preference Optimization with a Reference-Free Reward	Meng et al. (2024), SimPO; arXiv:2405.14734; DOI 10.48550/arXiv.2405.14734	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for simple reference-free preference optimization using sequence-level reward framing.
ext_reinforce_style_rlhf_2024	Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs	Ahmadian et al. (2024), REINFORCE-style RLHF revisiting; arXiv:2402.14740; DOI 10.48550/arXiv.2402.14740	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for revisiting simpler REINFORCE-style optimization as an RLHF baseline.
ext_deepseek_r1_2025	DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning	DeepSeek-AI et al. (2025), DeepSeek-R1; arXiv:2501.12948; DOI 10.1038/s41586-025-09422-z	policy_optimization	governed-deliberation-and-test-time-scaling (Governed Deliberation and Test-Time Scaling); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for reinforcement-learning pressure on reasoning behavior in large language models.
ext_dapo_2025	DAPO: An Open-Source LLM Reinforcement Learning System at Scale	Yu et al. (2025), DAPO; arXiv:2503.14476; DOI 10.48550/arXiv.2503.14476	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for open-source large-scale LLM RL system details and DAPO-style update design.
ext_gspo_2025	Group Sequence Policy Optimization	Zheng et al. (2025), Group Sequence Policy Optimization; arXiv:2507.18071; DOI 10.48550/arXiv.2507.18071	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for sequence-level group policy optimization in LLM reinforcement learning.
ext_s_grpo_2025	S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models	Dai et al. (2025), S-GRPO; arXiv:2505.07686; DOI 10.48550/arXiv.2505.07686	policy_optimization	governed-deliberation-and-test-time-scaling (Governed Deliberation and Test-Time Scaling); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for early-exit reinforcement learning and overthinking control in reasoning models.
ext_longrlvr_2026	LongRLVR: Long-Context Reinforcement Learning Requires Verifiable Context Rewards	Chen et al. (2026), LongRLVR; arXiv:2603.02146; DOI 10.48550/arXiv.2603.02146	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External source for long-context RLVR and verifiable context-grounding rewards.
ext_rlhf_limitations_2023	Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback	Casper et al. (2023), Open Problems and Fundamental Limitations of RLHF; arXiv:2307.15217; DOI 10.48550/arXiv.2307.15217	policy_optimization	policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	External survey source for RLHF limitations, reward hacking, evaluator limits, and complementary safeguards.
ext_tailscale_docs_2025	What is Tailscale?	Tailscale Docs (2025), What is Tailscale?; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official Tailscale documentation for zero-trust identity networking, tailnets, encrypted point-to-point connections, and cross-network device connectivity.
ext_kubernetes_overview_docs	Kubernetes Documentation: Overview	Kubernetes Documentation, Overview; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official Kubernetes overview for containerized workload management, declarative configuration, automation, service discovery, storage orchestration, rollouts, bin packing, and self-healing.
ext_k3s_docs_2026	K3s: Lightweight Kubernetes	K3s Documentation (2026), Lightweight Kubernetes; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official K3s documentation for lightweight Kubernetes deployment in edge, homelab, IoT, CI, single-board-computer, air-gapped, and embedded settings.
ext_nomad_docs	Nomad Documentation	HashiCorp Developer, Nomad Documentation; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official Nomad documentation for scheduling and orchestrating containers, non-containerized applications, and batch jobs across on-prem and cloud environments.
ext_temporal_docs	Temporal Documentation: What is Temporal?	Temporal Documentation (2026), What is Temporal?; source	durable_execution	labor-os-and-typed-jobs (Labor OS and Typed Jobs)	source note available	Official Temporal documentation comparator for durable workflow execution, workflow event histories, worker processes, failure recovery, and long-running application-code orchestration.
ext_airflow_dag_docs	Apache Airflow Documentation: Dags	Apache Airflow Documentation (2026), Dags; source	workflow_orchestration	labor-os-and-typed-jobs (Labor OS and Typed Jobs)	source note available	Official Apache Airflow documentation comparator for DAG-based workflow scheduling, tasks, dependencies, callbacks, retries, and operational workflow metadata.
ext_bpmn_2_0_2_spec	Business Process Model and Notation Specification Version 2.0.2	OMG (2014), BPMN 2.0.2; source	business_process_modeling	labor-os-and-typed-jobs (Labor OS and Typed Jobs)	source note available	OMG BPMN 2.0.2 formal specification comparator for stakeholder-readable business-process diagrams, implementation-independent flow notation, and translation into software process components.
ext_kubernetes_jobs_docs	Kubernetes Documentation: Jobs	Kubernetes Documentation (2026), Jobs; source	batch_job_lifecycle	labor-os-and-typed-jobs (Labor OS and Typed Jobs)	source note available	Official Kubernetes Jobs documentation comparator for batch job lifecycle, completions, backoff limits, active deadlines, terminal Complete/Failed conditions, and cleanup of finished jobs.
ext_ray_core_docs_2026	What’s Ray Core?	Ray Documentation (2026), What’s Ray Core?; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official Ray Core documentation for distributed task, actor, and object primitives used to build and scale Python applications.
ext_boinc_home_2026	BOINC	BOINC Project Site (2026); source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official BOINC site for volunteer computing where user computers download scientific computing jobs and run them in the background.
ext_syncthing_home	Syncthing	Syncthing Project Site; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official Syncthing site for continuous file synchronization across computers, authenticated devices, encrypted transport, and user-controlled storage location.
ext_ipfs_docs	IPFS Documentation and Project Site	IPFS Project Site; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence)	source note available	Official IPFS project documentation and site for peer-to-peer content addressing, content identifiers, provider discovery, and decentralized retrieval vocabulary.
ext_akash_docs_2026	Akash Network Documentation	Akash Network Documentation (2026); source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	Official Akash documentation for decentralized cloud deployment, provider resources, leases, GPUs, SDKs, node operations, and provider operations.
ext_golem_docs_2025	Golem Developer Resources	Golem Docs (2025), Developer Resources; source	personal_compute_hives	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	Official Golem developer resources for decentralized computations, task execution, provider selection, result handling, and resource sharing.
ext_github_webhooks_docs	Webhook events and payloads	GitHub Docs (2026), Webhook events and payloads; source	artifact_steward_agents	artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	Official GitHub documentation for repository and organization webhook events, event payloads, delivery headers, event-specific permissions, and payload limits.
ext_github_self_hosted_runners_docs	Self-hosted runners	GitHub Docs (2026), Self-hosted runners; source	artifact_steward_agents	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	Official GitHub Actions documentation for self-hosted runners as user-managed systems that execute workflow jobs on physical, virtual, containerized, on-prem, or cloud machines.
ext_openzeppelin_governor_docs	OpenZeppelin Contracts: Governance	OpenZeppelin Contracts Docs, Governance; source	artifact_steward_agents	artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	Official OpenZeppelin governance documentation for modular Governor contracts, voting power, quorum, timelocks, proposal settings, and guardian-style extensions.
ext_open_collective_docs	Open Collective Documentation	Open Collective Documentation (2026); source	artifact_steward_agents	artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	Official Open Collective documentation for transparent community money management, fiscal hosts, contribution intake, expenses, accounting, and legal-entity delegation through fiscal hosting.
ext_github_sponsors_docs	About GitHub Sponsors for open source contributors	GitHub Docs (2026), About GitHub Sponsors; source	artifact_steward_agents	artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	Official GitHub Sponsors documentation for contributor and organization sponsorship eligibility, open-source contribution categories, sponsor profiles, and GitHub-native funding surfaces.
ext_agentic_workflow_injection_2026	Demystifying and Detecting Agentic Workflow Injection Vulnerabilities in GitHub Actions	Wang et al. (2026), Agentic Workflow Injection in GitHub Actions; arXiv:2605.07135; DOI 10.48550/arXiv.2605.07135	artifact_steward_agents	artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	External security paper for agentic workflow injection in GitHub Actions when untrusted repository event context reaches LLM agents and downstream workflow logic.
ext_dao_delegation_fairness_2025	Fairness in Token Delegation: Mitigating Voting Power Concentration in DAOs	Messias and Ide (2025), Fairness in Token Delegation; arXiv:2510.05830; DOI 10.48550/arXiv.2510.05830	artifact_steward_agents	artifact-steward-agents-and-living-project-governance (Artifact Steward Agents and Living Project Governance)	source note available	External DAO governance paper for voter apathy, voting-power concentration, delegation misalignment, and delegate-ranking bias.
ext_model_cards_2019	Model Cards for Model Reporting	Mitchell et al. (2019), Model Cards; arXiv:1810.03993; DOI 10.48550/arXiv.1810.03993	model_reporting	project-theseus-as-report-first-implementation-reference (Project Theseus as Report-First Implementation Reference)	source note available	External reporting comparator for structured model documentation, intended-use boundaries, evaluation disclosures, ethical considerations, and model-report artifacts.
ext_datasheets_datasets_2021	Datasheets for Datasets	Gebru et al. (2021), Datasheets for Datasets; arXiv:1803.09010; DOI 10.48550/arXiv.1803.09010	dataset_documentation	project-theseus-as-report-first-implementation-reference (Project Theseus as Report-First Implementation Reference)	source note available	External documentation comparator for dataset motivation, composition, collection, preprocessing, uses, distribution, maintenance, and accountability questions.
ext_factsheets_ai_services_2019	FactSheets: Increasing Trust in AI Services through Supplier’s Declarations of Conformity	Arnold et al. (2019), FactSheets for AI Services; arXiv:1808.07261; DOI 10.48550/arXiv.1808.07261	ai_service_fact_sheets	project-theseus-as-report-first-implementation-reference (Project Theseus as Report-First Implementation Reference)	source note available	External AI-service documentation comparator for supplier declarations, service properties, trust-relevant facts, and standardized reporting boundaries.
ext_ml_reproducibility_program_2021	Improving Reproducibility in Machine Learning Research	Pineau et al. (2021), Improving Reproducibility in Machine Learning Research; source	ml_reproducibility_reporting	project-theseus-as-report-first-implementation-reference (Project Theseus as Report-First Implementation Reference)	source note available	External reproducibility-program comparator for checklists, code submission, reproducibility reports, and community review mechanisms in machine-learning research.
ext_transformer_circuits_2021	A Mathematical Framework for Transformer Circuits	Elhage et al. (2021), Transformer Circuits; source	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	External mechanistic-interpretability comparator for treating internal circuit analyses as scoped white-box evidence that still needs model, layer, behavior, causal, and limitation boundaries.
ext_monosemanticity_2023	Towards Monosemanticity: Decomposing Language Models With Dictionary Learning	Bricken et al. (2023), Towards Monosemanticity; source	mechanistic_interpretability	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	External mechanistic-interpretability comparator for sparse-autoencoder feature decomposition and the boundary between discovered features, feature-level evidence, and broader model-behavior claims.
ext_literate_programming_1984	Literate Programming	Knuth (1984), Literate Programming; source; DOI 10.1093/comjnl/27.2.97	literate_programming	living-book-methodology (Living Book Methodology)	source note available	External literate-programming source for arranging programs and explanation around human comprehension, woven documentation, and tangled executable artifacts.
ext_jupyter_book_docs	Jupyter Book Documentation	Jupyter Book Documentation; source	executable_books	living-book-methodology (Living Book Methodology)	source note available	Official Jupyter Book documentation comparator for authoring books from notebooks or Markdown, executing code, cross-referencing content, and publishing computational books to the web.
ext_quarto_books_docs	Quarto Books Documentation	Quarto Documentation, Creating a Book; source	technical_publishing	living-book-methodology (Living Book Methodology)	source note available	Official Quarto Books documentation comparator for multi-chapter manuscripts, HTML/PDF/Word/EPUB outputs, search, cross references, and book-style website publishing.
ext_argo_rollouts_docs	Argo Rollouts Documentation: Kubernetes Progressive Delivery Controller	Argo Rollouts Documentation, Kubernetes Progressive Delivery Controller; source	progressive_delivery_rollback	capability-replacement-and-rollback (Capability Replacement and Rollback)	source note available	External progressive-delivery comparator for blue-green rollout, canary rollout, metric analysis, automated promotion, and automated rollback vocabulary.
ext_feature_toggles_fowler	Feature Toggles (aka Feature Flags)	Hodgson (2017), Feature Toggles (aka Feature Flags); source	feature_flag_release_control	capability-replacement-and-rollback (Capability Replacement and Rollback)	source note available	External feature-flag comparator for controlled exposure, canary releasing, release toggles, experiment toggles, ops toggles, permissioning toggles, and validation complexity.
ext_google_cloud_mlops_cd	MLOps: Continuous Delivery and Automation Pipelines in Machine Learning	Google Cloud Architecture Center (2024), MLOps Continuous Delivery and Automation Pipelines in Machine Learning; source	mlops_continuous_delivery	capability-replacement-and-rollback (Capability Replacement and Rollback)	source note available	External MLOps comparator for CI/CD/CT, data/model validation, model deployment, monitoring, rollback triggers, and model-regression concerns.
ext_kubernetes_deployments_docs	Kubernetes Documentation: Deployments	Kubernetes Documentation, Deployments; source	deployment_rollout_rollback	capability-replacement-and-rollback (Capability Replacement and Rollback)	source note available	External deployment-controller comparator for rollout status, rollout history, revision records, and rollback to a prior stable Deployment revision.
ext_drexler_cais_2019	Reframing Superintelligence: Comprehensive AI Services as General Intelligence	Drexler (2019), Reframing Superintelligence: Comprehensive AI Services as General Intelligence; source	ai_services_r_and_d_automation	asi-is-a-stack-not-a-model (ASI Is a Stack, Not a Model); constitutional-alignment-substrate (Constitutional Alignment: Agency, Dignity, and Corrigibility); recursive-self-improvement-boundaries (Recursive Self-Improvement Boundaries); integrated-reference-architecture (Integrated Reference Architecture)	source note available	Primary CAIS technical-report comparator for service-centered general intelligence, R&D automation, structured AI development, and the distinction between component affordances and a complete governance architecture.
ext_mrkl_systems_2022	MRKL Systems: A Modular, Neuro-Symbolic Architecture That Combines Large Language Models, External Knowledge Sources and Discrete Reasoning	Karpas et al. (2022), MRKL Systems; arXiv:2205.00445; DOI 10.48550/arXiv.2205.00445	neuro_symbolic_modular_architecture	asi-is-a-stack-not-a-model (ASI Is a Stack, Not a Model) (inventory chapter target; not Appendix C support mapping)	source note available	External modular-neuro-symbolic architecture comparator for combining language models with expert modules, external knowledge sources, and routing rather than treating the model as the whole system.
ext_llm_agents_survey_2023	A Survey on Large Language Model based Autonomous Agents	Wang et al. (2023), Survey on LLM-based Autonomous Agents; arXiv:2308.11432; DOI 10.48550/arXiv.2308.11432	llm_agent_architecture	asi-is-a-stack-not-a-model (ASI Is a Stack, Not a Model) (inventory chapter target; not Appendix C support mapping)	source note available	External LLM-agent architecture comparator for agent profiles, memory, planning, and action modules around a language model, useful for positioning the stack frame against agent-system decompositions.
ext_standard_model_mind_2017	A Standard Model of the Mind: Toward a Common Computational Framework Across Artificial Intelligence, Cognitive Science, Neuroscience, and Robotics	Laird, Lebiere, and Rosenbloom (2017), Standard Model of the Mind; source; DOI 10.1609/aimag.v38i4.2744	cognitive_architecture	asi-is-a-stack-not-a-model (ASI Is a Stack, Not a Model) (inventory chapter target; not Appendix C support mapping)	source note available	External cognitive-architecture comparator for treating intelligent behavior as an integrated architecture spanning memory, learning, perception/action, procedural control, and deliberation.
ext_subsumption_architecture_1986	A Robust Layered Control System for a Mobile Robot	Brooks (1986), Robust Layered Control System; source; DOI 10.1109/JRA.1986.1087032	layered_robot_control_architecture	asi-is-a-stack-not-a-model (ASI Is a Stack, Not a Model) (inventory chapter target; not Appendix C support mapping)	source note available	External layered-control architecture comparator for decomposing robot behavior into interacting layers rather than centralizing behavior in one monolithic controller.
ext_humans_automation_1997	Humans and Automation: Use, Misuse, Disuse, Abuse	Parasuraman and Riley (1997), Humans and Automation; source; DOI 10.1518/001872097778543886	human_factors_automation	human-factors-and-meaningful-control-in-oversight (Human Factors and Meaningful Control in Oversight); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval)	source note available	External human-factors comparator for automation use, misuse, disuse, abuse, overreliance, monitoring failure, workload, trust, risk, false alarms, and operator-role design.
ext_ironies_automation_1983	Ironies of Automation	Bainbridge (1983), Ironies of Automation; source; DOI 10.1016/0005-1098(83)90046-8	human_factors_automation	human-factors-and-meaningful-control-in-oversight (Human Factors and Meaningful Control in Oversight); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval)	source note available	External automation comparator for the argument that automation can expand rather than eliminate human-operator problems and can leave humans with difficult abnormal-condition duties.
ext_levels_automation_2000	A Model for Types and Levels of Human Interaction with Automation	Parasuraman, Sheridan, and Wickens (2000), Types and Levels of Automation; source; DOI 10.1109/3468.844354	human_factors_automation	human-factors-and-meaningful-control-in-oversight (Human Factors and Meaningful Control in Oversight); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval)	source note available	External human-automation comparator for separating automation by information acquisition, analysis, decision/action selection, and action implementation rather than treating human approval as a single undifferentiated gate.
ext_complacency_bias_automation_2010	Complacency and Bias in Human Use of Automation: An Attentional Integration	Parasuraman and Manzey (2010), Complacency and Bias in Human Use of Automation; source; DOI 10.1177/0018720810376055	human_factors_automation	human-factors-and-meaningful-control-in-oversight (Human Factors and Meaningful Control in Oversight); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval)	source note available	External human-factors comparator for automation complacency, omission and commission errors, automation bias, workload, attention, and imperfect decision aids.
ext_bourtoule_machine_unlearning_2021	Machine Unlearning	Bourtoule et al. (2021), Machine Unlearning; source; DOI 10.1109/SP40001.2021.00019	machine_unlearning_data_governance	context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); adjudicated-persistence-and-the-adaptive-commit-boundary (Adjudicated Persistence and the Adaptive Commit Boundary); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	Primary machine-unlearning comparator for deletion-path architecture, checkpointed training, bounded retraining, accuracy-cost trade-offs, and the boundary between deletion requests and verified removal.
ext_shumailov_model_collapse_2023	The Curse of Recursion: Training on Generated Data Makes Models Forget	Shumailov et al. (2023), The Curse of Recursion; source; DOI 10.48550/arXiv.2305.17493	synthetic_data_feedback_model_collapse	data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	Primary preprint comparator for generated-data feedback, provenance, distribution-tail loss, and model-collapse risk under specified recursive-training assumptions; not a universal synthetic-data safety result.
ext_gerstgrasser_data_accumulation_2024	Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data	Gerstgrasser et al. (2024), Is Model Collapse Inevitable?; source; DOI 10.48550/arXiv.2404.01413	synthetic_data_retention_policy	data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	Primary empirical and analytical comparator that distinguishes replacement from accumulation of real and synthetic data; gives a counterweight to blanket model-collapse claims without resolving deletion, privacy, provenance, or poisoning risk.
ext_alignment_faking_2024	Alignment Faking in Large Language Models	Greenblatt et al. (2024), Alignment Faking in Large Language Models; source; DOI 10.48550/arXiv.2412.14093	training_time_deception	adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception)	source note available	Primary experimental comparator for context-dependent alignment faking under disclosed training conditions; does not establish that an ASI Stack model, evaluator, or training process is deceptive.
ext_ai_sandbagging_2024	AI Sandbagging: Language Models Can Strategically Underperform on Evaluations	van der Weij et al. (2024), AI Sandbagging; source; DOI 10.48550/arXiv.2406.07358	evaluation_integrity	adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception)	source note available	Primary experimental comparator for strategic evaluation underperformance, prompted and password-locked capability hiding, and limits of capability-evaluation trustworthiness; does not show sandbagging in this repository.
ext_emergent_misalignment_reward_hacking_2025	Natural Emergent Misalignment from Reward Hacking in Production RL	MacDiarmid et al. (2025), Natural Emergent Misalignment from Reward Hacking in Production RL; source; DOI 10.48550/arXiv.2511.18397	training_time_deception	inner-alignment-mesa-optimization-and-learned-objective-integrity (Inner Alignment, Mesa-Optimization, and Learned-Objective Integrity); governed-objective-formation-value-learning-and-goal-integrity (Governed Objective Formation, Value Learning, and Goal Integrity); adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception)	source note available	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.
ext_poet_2019	Paired Open-Ended Trailblazer (POET): Endlessly Generating Increasingly Complex and Diverse Learning Environments and Their Solutions	Wang et al. (2019), Paired Open-Ended Trailblazer; source; DOI 10.48550/arXiv.1901.01753	open_ended_environment_solution_generation	open-ended-improvement-engines (Open-Ended Improvement Engines)	source note available	Primary open-ended-learning comparator for paired environment generation, agent optimization, and cross-environment solution transfer in a specified BipedalWalker setting; it does not establish a general improvement engine, evaluator soundness, or ASI Stack result.
ext_funsearch_2024	Mathematical Discoveries from Program Search with Large Language Models	Romera-Paredes et al. (2024), Mathematical Discoveries from Program Search with Large Language Models; source; DOI 10.1038/s41586-023-06924-6	evaluator_bounded_program_search	open-ended-improvement-engines (Open-Ended Improvement Engines)	source note available	Primary program-search comparator for a fixed pretrained LLM, user-provided evaluation function, candidate program archive, and iterative search over a bounded specification; it does not establish open-ended general intelligence, self-modification, evaluator correctness, or an ASI Stack result.
ext_gsn_community_standard_2011	GSN Community Standard Version 1	Assurance Case Working Group (2011), GSN Community Standard v1; source	structured_assurance_argumentation	safety-cases-and-structured-assurance (Safety Cases and Structured Assurance)	source note available	Primary notation standard comparator for explicit goals, strategies, solutions, context, assumptions, justifications, and relationships in structured assurance arguments; the notation documents asserted support but does not establish claim truth.
ext_evaluations_safety_cases_scheming_2024	Towards Evaluations-Based Safety Cases for AI Scheming	Balesni et al. (2024), Evaluations-Based Safety Cases for AI Scheming; arXiv:2411.03336; DOI 10.48550/arXiv.2411.03336	ai_safety_case_methodology	safety-cases-and-structured-assurance (Safety Cases and Structured Assurance)	source note available	Primary safety-case comparator for scoped scheming inability, harm inability, harm control, alignment arguments, empirical evaluation dependencies, and acknowledged unresolved assumptions; it does not establish any ASI Stack safety case or safety result.
ext_aisi_safety_cases_2024	Safety Cases at AISI	AI Safety Institute (2024), Safety Cases at AISI; source	ai_safety_case_methodology	safety-cases-and-structured-assurance (Safety Cases and Structured Assurance)	source note available	Official AI Safety Institute methodology comparator for structured safety-case sketches, positive and negative evidence, countercases, open scientific uncertainty, and limits on confidence; it is not evidence that this book has a complete safety case.
ext_rand_model_weight_security_2024	Securing AI Model Weights: Preventing Theft and Misuse of Frontier Models	Nevo et al. (2024), Securing AI Model Weights; source	model_weight_custody	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); open-weight-release-and-post-release-control (Open-Weight Release and Post-Release Control)	source note available	Primary RAND analysis of frontier-model-weight theft/misuse threat surfaces, security levels, defense-in-depth, access control, physical and organizational controls; it does not establish local protection or safety.
ext_nist_confidential_computing_2026	Hardware-Enabled Security: Confidential Computing of Data in Cloud Workloads	NIST (2026), IR 8320E Initial Public Draft; source	hardware_root_attestation	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust)	source note available	NIST initial-public-draft comparator for hardware-enabled confidential computing, memory protection, trust domains, attestation-gated key release, and AI model/data protection; it is draft guidance, not a local TEE result.
ext_nvidia_confidential_model_lifecycle_2026	Workload and Model Lifecycle: Deploying Proprietary Models Securely with NVIDIA Confidential Computing	NVIDIA (2026), Confidential Model Lifecycle; source	attestation_gated_model_loading	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust)	source note available	Official vendor implementation-reference comparator for encrypted weights outside a confidential pod, policy-sensitive attestation evidence, and key-release decisions; it does not establish a local confidential deployment or attestation result.
ext_provable_model_weight_release_2025	Towards Provable (In)Secure Model Weight Release Schemes	Yang et al. (2025), Provable (In)Secure Model Weight Release Schemes; arXiv:2506.19874; DOI 10.48550/arXiv.2506.19874	open_weight_release_security	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); open-weight-release-and-post-release-control (Open-Weight Release and Post-Release Control)	source note available	Primary formal-security comparator for evaluating claimed secure model-weight release schemes and parameter-extraction failure modes; it does not establish an ASI Stack release scheme or release decision.
ext_nist_cscrm_2022	Cybersecurity Supply Chain Risk Management Practices for Systems and Organizations	NIST (2022), SP 800-161r1; source	ai_supply_chain_governance	ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance)	source note available	Primary NIST C-SCRM standard comparator for lifecycle-wide risk framing, supplier/component inventory, assessment, response, monitoring, and incident communication; it does not establish a local supply-chain program or AI artifact integrity.
ext_slsa_build_track_1_2	SLSA Build Track Basics, version 1.2	SLSA (2025), Build Track Basics v1.2; source	build_provenance	ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance)	source note available	Official SLSA specification comparator for build provenance, signed hosted builds, verification, and graduated assurance; provenance quality and SLSA level do not establish local artifact correctness, data quality, model safety, or deployment authority.
ext_openssf_model_signing_spec_2025	OpenSSF Model Signing Specification	OpenSSF (2025), Model Signing Specification; source	ai_artifact_signing	ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance)	source note available	Official OpenSSF AI/ML working-group specification comparator for signed model/dataset bundles, verification, provenance metadata, and explicit limits of model signing; it does not establish local signing, verification, integrity, confidentiality, safety, or release authority.
ext_spdx_ai_profile_3_0_1	SPDX Specification 3.0.1 AI Profile	SPDX (2024), Specification 3.0.1 AI Profile; source	ai_bill_of_materials	ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance)	source note available	Official SPDX specification comparator for interoperable AI system/model, dataset, build, supplier, provenance, integrity, relationship, and lifecycle metadata; a conformant BOM is not proof of complete inventory, artifact security, data fitness, model safety, or compliance.
ext_mcp_protocol_2025_06_18	Model Context Protocol Specification, revision 2025-06-18	Model Context Protocol (2025), Specification revision 2025-06-18; source	agent_tool_protocol	inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange) (inventory chapter target; not Appendix C support mapping)	source note available	Official Model Context Protocol comparator for JSON-RPC message shape, lifecycle management, capability negotiation, session control, schema-defined interactions, and modular tool/client/server features; it does not establish a local protocol implementation, peer identity, authorization, message truth, task completion, payment, or deployment safety.
ext_a2a_protocol_0_3_0	Agent2Agent Protocol Specification, version 0.3.0	A2A Protocol (v0.3.0), Specification; source	agent_to_agent_protocol	inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange) (inventory chapter target; not Appendix C support mapping)	source note available	Official A2A comparator for agent discovery, agent cards, delegated tasks, artifact/message exchange, transport choices, and interoperability between opaque agent systems; it does not establish a local A2A deployment, verified identity, delegated authority, task truth, secure execution, payment, or safety.
ext_mcp_protocol_2025_11_25	Model Context Protocol Specification, revision 2025-11-25	Model Context Protocol (2025), Specification revision 2025-11-25; source	agent_tool_protocol	inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange)	source note available	Latest released MCP comparator inspected on 2026-07-10 for versioned lifecycle, capability negotiation, authorization and OpenID Connect discovery changes, elicitation, tasks, and modular client/server boundaries; the announced 2026-07-28 revision remains a release candidate and is not represented as released.
ext_a2a_protocol_1_0_0	Agent2Agent Protocol Specification, version 1.0.0	A2A Protocol (v1.0.0), Specification; source	agent_to_agent_protocol	inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange)	source note available	Latest released A2A comparator inspected on 2026-07-10 for canonical data objects, version negotiation, Agent Cards, tasks/messages/artifacts, JSON-RPC, gRPC and HTTP bindings, authorization scoping, interoperability testing, and security considerations; it does not establish local conformance, peer truth, delegated authority, or safe effects.
ext_w3c_did_core_1_0_2022	Decentralized Identifiers (DIDs) v1.0	W3C (2022), Decentralized Identifiers v1.0; source	decentralized_identity	inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange)	source note available	W3C DID Core comparator for decentralized identifier syntax, data model, controller-related metadata, resolution, and privacy considerations; it does not establish an ASI Stack identity system, controller trust, authorization, non-repudiation, revocation effectiveness, or safety.
ext_w3c_vc_data_model_2_0_2025	Verifiable Credentials Data Model v2.0	W3C (2025), Verifiable Credentials Data Model v2.0; source	verifiable_credentials	inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange)	source note available	W3C Verifiable Credentials comparator for issuer/holder/verifier roles, credential and presentation fields, validity/status, evidence, securing mechanisms, and explicit authorization limitations; it does not establish an ASI Stack credential, trust decision, authorization framework, delegation validity, payment, or safety.
ext_interledger_protocol_v4	Interledger Protocol V4	Interledger (2026), Interledger Protocol V4; source	interledger_value_transfer	inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange)	source note available	Official Interledger comparator for neutral packetized value transfer across independent ledgers, connector obligations, balances, and end-to-end boundary design; it does not establish an ASI Stack payment route, settlement, accounting correctness, legal transfer, economic fairness, delegated authority, or safety.
ext_test_time_compute_scaling_2024	Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters	Snell et al. (2024), Scaling LLM Test-Time Compute Optimally; arXiv:2408.03314; DOI 10.48550/arXiv.2408.03314	test_time_compute_allocation	governed-deliberation-and-test-time-scaling (Governed Deliberation and Test-Time Scaling)	source note available	Primary test-time-compute comparator for verifier-guided search, proposal refinement, difficulty-dependent compute allocation, and the limits of extra inference; it does not establish local reasoning improvement, verifier correctness, safety, or an ASI Stack result.
ext_graphrag_2024	From Local to Global: A Graph RAG Approach to Query-Focused Summarization	Edge et al. (2024), From Local to Global: A Graph RAG Approach; arXiv:2404.16130; DOI 10.48550/arXiv.2404.16130	graph_based_retrieval_and_global_sensemaking	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); durable-semantic-memory-and-knowledge-lattices (Durable Semantic Memory and Knowledge Lattices); context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); open-research-agenda-and-bibliography-plan (Open Research Agenda and Bibliography Plan)	source note available	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.
ext_hipporag_2024	HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models	Gutiérrez et al. (2024), HippoRAG; source; DOI 10.52202/079017-1902	associative_long_term_memory	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); durable-semantic-memory-and-knowledge-lattices (Durable Semantic Memory and Knowledge Lattices); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); open-research-agenda-and-bibliography-plan (Open Research Agenda and Bibliography Plan)	source note available	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.
ext_raptor_2024	RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval	Sarthi et al. (2024), RAPTOR; source	hierarchical_retrieval_and_abstraction	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); compact-generative-systems-and-residual-honesty (Compact Generative Systems: Generate, Verify, Repair, and Residual Honesty); rankfold-neuralfold-and-artifact-compression (RankFold, NeuralFold, and Artifact Compression)	source note available	Primary ICLR comparator for recursive clustering, summarization, and retrieval across multiple abstraction levels; source-reported QA gains do not prove summary fidelity, provenance preservation, local reproduction, or safe compression for ASI Stack claims.
ext_mem0_2025	Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory	Chhikara et al. (2025), Mem0; arXiv:2504.19413; DOI 10.48550/arXiv.2504.19413	agent_long_term_memory	virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); durable-semantic-memory-and-knowledge-lattices (Durable Semantic Memory and Knowledge Lattices); context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); procedural-memory-and-cognitive-loop-closure (Procedural Memory and Cognitive Loop Closure); resource-economics-and-token-budgets (Resource Economics and Token Budgets); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	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.
ext_w3c_prov_o_2013	PROV-O: The PROV Ontology	W3C (2013), PROV-O: The PROV Ontology; source	interoperable_provenance_model	evidence-states-and-claim-discipline (Evidence States and Claim Discipline); ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance); claim-ledgers-and-belief-revision (Claim Ledgers and Belief Revision); artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	W3C Recommendation comparator for interoperable provenance over entities, activities, agents, derivation, attribution, delegation, revision, and invalidation; a PROV-O graph records asserted provenance and does not by itself prove assertion truth, completeness, integrity, authority, or safety.
ext_mlcommons_croissant_1_1_2026	Croissant Format Specification, version 1.1	MLCommons (2026), Croissant Format Specification v1.1; source	machine_readable_dataset_metadata	ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance); artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	Current MLCommons specification comparator for JSON-LD dataset structure, resources, checksums, record fields, machine-readable provenance, usage conditions, and portability across ML tooling; metadata conformance does not prove dataset integrity, fitness, legality, representativeness, or safe use.
ext_inspect_ai_2024	Inspect AI: Framework for Large Language Model Evaluations	UK AI Security Institute (2024), Inspect AI; source	model_and_agent_evaluation_framework	runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); capability-thresholds-and-deployment-commitments (Capability Thresholds and Deployment Commitments); adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception)	source note available	Official UK AI Security Institute framework comparator for composable evaluation tasks, datasets, solvers, scorers, agents, tools, logs, and sandboxes; framework availability or a passing task does not establish benchmark validity, coverage, local execution, safety, or deployment readiness.
ext_in_toto_2019	in-toto: Providing farm-to-table guarantees for bits and bytes	Torres-Arias et al. (2019), in-toto; source	software_supply_chain_attestation	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance); artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay)	source note available	Primary USENIX comparator for cryptographically verifying authorized software-supply-chain steps from source through deployment; valid attestations do not prove artifact correctness, uncompromised authorized actors, model safety, data fitness, or deployment merit.
ext_agentdojo_2024	AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents	Debenedetti et al. (2024), AgentDojo; source; DOI 10.52202/079017-2636	agent_prompt_injection_evaluation	security-kernel-and-digital-scifs (Security Kernel and Digital SCIFs); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception)	source note available	Primary NeurIPS benchmark comparator for agents executing tools over untrusted data, with realistic tasks, security test cases, attacks, and defenses; benchmark results do not establish complete attack coverage, deployed robustness, safe authority handling, or local reproduction.
ext_camel_prompt_injection_2025	Defeating Prompt Injections by Design	Debenedetti et al. (2025), Defeating Prompt Injections by Design; arXiv:2503.18813; DOI 10.48550/arXiv.2503.18813	capability_secure_agent_control_flow	system-boundaries-and-authority (System Boundaries and Authority); security-kernel-and-digital-scifs (Security Kernel and Digital SCIFs); intent-to-execution-contracts (Command Contracts: From Intent to Executable Work); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval)	source note available	Primary CaMeL comparator for separating trusted control flow from untrusted data and enforcing capability policies at tool calls; source-reported AgentDojo results do not establish universal prompt-injection resistance, correct policy extraction, local implementation, or safe deployment.
ext_owasp_agentic_top_10_2026	OWASP Top 10 for Agentic Applications for 2026	OWASP (2025), Top 10 for Agentic Applications for 2026; source	agentic_application_security_taxonomy	system-boundaries-and-authority (System Boundaries and Authority); security-kernel-and-digital-scifs (Security Kernel and Digital SCIFs); ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange); adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception)	source note available	Current OWASP community taxonomy comparator for goal hijacking, tool misuse, identity abuse, agentic supply chains, code execution, memory poisoning, inter-agent communication, cascading failures, human trust exploitation, and rogue agents; a risk list is not a proof of completeness, control effectiveness, local testing, or system safety.
ext_darwin_godel_machine_2025	Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents	Zhang et al. (2025), Darwin Godel Machine; arXiv:2505.22954; DOI 10.48550/arXiv.2505.22954	empirical_recursive_agent_improvement	recursive-self-improvement-boundaries (Recursive Self-Improvement Boundaries); open-ended-improvement-engines (Open-Ended Improvement Engines); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	Primary preprint comparator for archive-based open-ended code self-modification selected by empirical coding benchmarks under sandboxing and human oversight; reported benchmark gains do not establish monotonic general improvement, safe recursive self-improvement, local reproduction, or permission to self-modify.
ext_adas_2024	Automated Design of Agentic Systems	Hu et al. (2024), Automated Design of Agentic Systems; arXiv:2408.08435; DOI 10.48550/arXiv.2408.08435	automated_agent_architecture_search	recursive-self-improvement-boundaries (Recursive Self-Improvement Boundaries); open-ended-improvement-engines (Open-Ended Improvement Engines); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); integrated-reference-architecture (Integrated Reference Architecture)	source note available	Primary ADAS comparator for a meta-agent that searches a growing archive of code-defined agent designs across prompts, tools, and workflows; reported transfer results do not establish unrestricted generality, safe architecture search, local reproduction, or automatic promotion authority.
ext_universal_transformer_2019	Universal Transformers	Dehghani et al. (2019), Universal Transformers; arXiv:1807.03819; DOI 10.48550/arXiv.1807.03819	shared_weight_recurrence_and_adaptive_depth	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); governed-deliberation-and-test-time-scaling (Governed Deliberation and Test-Time Scaling); mathematical-and-search-substrates (Mathematical and Search Substrates); coil-attention-cyclic-memory-and-recurrence-contracts (Coil Attention, Cyclic Memory, and Recurrence Contracts)	source note available	Primary ICLR comparator for shared-weight depth recurrence, parallel self-attention, and per-position dynamic halting; benchmark results and theoretical expressivity do not establish stable deep recurrence, efficient scaling, local reproduction, or the book’s cyclic-memory claims.
ext_recurrent_transformer_2026	The Recurrent Transformer: Greater Effective Depth and Efficient Decoding	Oncescu et al. (2026), The Recurrent Transformer; arXiv:2604.21215; DOI 10.48550/arXiv.2604.21215	layerwise_recurrent_transformer_memory	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets); mathematical-and-search-substrates (Mathematical and Search Substrates); coil-attention-cyclic-memory-and-recurrence-contracts (Coil Attention, Cyclic Memory, and Recurrence Contracts)	source note available	Current preprint comparator for layerwise recurrent key-value memory, exact tiling, effective-depth/width tradeoffs, and standard autoregressive decoding cost; small-model C4 results do not establish broad capability gains, production efficiency, local reproduction, or cyclic-memory correctness.
ext_dynamic_compute_recurrent_transformers_2026	Understanding Dynamic Compute Allocation in Recurrent Transformers	Moosa et al. (2026), Understanding Dynamic Compute Allocation in Recurrent Transformers; arXiv:2602.08864; DOI 10.48550/arXiv.2602.08864	adaptive_recurrent_compute_evaluation	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); governed-deliberation-and-test-time-scaling (Governed Deliberation and Test-Time Scaling); resource-economics-and-token-budgets (Resource Economics and Token Budgets); coil-attention-cyclic-memory-and-recurrence-contracts (Coil Attention, Cyclic Memory, and Recurrence Contracts); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	Current preprint comparator for complexity-controlled tests of token-level variable-depth compute and online halting; its negative result that difficulty-aligned compute need not generalize is a boundary against equating adaptive depth with algorithmic extrapolation or local capability.
ext_claw_swe_bench_2026	Claw-SWE-Bench: A Benchmark for Evaluating OpenClaw-style Agent Harnesses on Coding Tasks	Zheng et al. (2026), Claw-SWE-Bench; arXiv:2606.12344; DOI 10.48550/arXiv.2606.12344	coding_agent_harness_and_cost_evaluation	artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); resource-economics-and-token-budgets (Resource Economics and Token Budgets); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	Current preprint comparator for fixed workspace, patch, evaluator, and budget contracts across coding-agent harnesses; reported accuracy and cost differences are not reproduced here and do not validate the post-v2.1 synthetic repository corpus.
ext_txfs_2018	TxFS: Leveraging File-System Crash Consistency to Provide ACID Transactions	Hu et al. (2018), TxFS; source	transactional_filesystem_rollback_boundary	capability-replacement-and-rollback (Capability Replacement and Rollback); context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval)	source note available	Primary systems comparator for user-space ACID file transactions built on journaling, including atomicity, isolation, durability, bounded transaction size, and Git/SQLite evaluation; it prevents treating a directory copy as a general transactional-filesystem result.
ext_dont_hallucinate_abstain_2024	Don’t Hallucinate, Abstain: Identifying LLM Knowledge Gaps via Multi-LLM Collaboration	Feng et al. (2024), Don’t Hallucinate, Abstain; source; DOI 10.18653/v1/2024.acl-long.786	llm_abstention_and_knowledge_gaps	verification-bandwidth-and-context-adequacy (Verification Bandwidth and Context Adequacy); routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); readiness-gates-residual-escrow-and-quarantine (Readiness Gates, Residual Escrow, and Quarantine)	source note available	Primary ACL comparator for knowledge-gap detection, abstention, calibration/self-reflection limitations, and multi-model probing; reported abstention improvements are task- and model-bounded and do not validate the local router or evaluator.
ext_muse_unlearning_2025	MUSE: Machine Unlearning Six-Way Evaluation for Language Models	Shi et al. (2025), MUSE; arXiv:2407.06460; DOI 10.48550/arXiv.2407.06460	llm_unlearning_multidimensional_evaluation	benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	Primary ICLR benchmark comparator separating verbatim and knowledge memorization, privacy leakage, retained utility, removal-scale behavior, and sequential sustainability; none of its 7B-language-model results are reproduced by the local policy network.
ext_unlearning_benchmarks_weak_2024	Position: LLM Unlearning Benchmarks are Weak Measures of Progress	Thaker et al. (2024), LLM Unlearning Benchmarks are Weak Measures; arXiv:2410.02879; DOI 10.48550/arXiv.2410.02879	unlearning_benchmark_validity	evidence-states-and-claim-discipline (Evidence States and Claim Discipline); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	Primary critical comparator showing that benign benchmark modifications, forget/retain dependencies, and ambiguous targets can make unlearning scores optimistic; it strengthens the book’s prohibition on turning toy behavioral change into influence, privacy, or storage claims.
ext_openunlearning_2025	OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics	Dorna et al. (2025), OpenUnlearning; arXiv:2506.12618; DOI 10.48550/arXiv.2506.12618	unlearning_method_and_metric_meta_evaluation	benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning); living-book-methodology (Living Book Methodology)	source note available	Primary NeurIPS 2025 benchmark-framework comparator for unified algorithms, diverse evaluations, public checkpoints, and meta-evaluation of metric faithfulness; it reinforces evaluator-quality residuals rather than establishing local unlearning.
ext_faithfulness_information_flow_2026	Faithfulness as Information Flow: Evaluating and Training Faithful Chain-of-Thought Reasoning	Jia, Benton, and Easley (2026), Faithfulness as Information Flow; arXiv:2605.24286; DOI 10.48550/arXiv.2605.24286	reasoning_trace_faithfulness	artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay); governed-deliberation-and-test-time-scaling (Governed Deliberation and Test-Time Scaling); adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	Primary 2026 comparator that separates chain-of-thought sufficiency, completeness, and interventional necessity, demonstrates prompt-to-answer shortcuts and transparent reward-hacking diagnostics, and documents low-entropy and reference-model limits. It does not make a reasoning transcript an authoritative receipt or establish local monitorability.
ext_monitorbench_2026	MonitorBench: A Comprehensive Benchmark for Chain-of-Thought Monitorability in Large Language Models	Wang et al. (2026), MonitorBench; arXiv:2603.28590; DOI 10.48550/arXiv.2603.28590	reasoning_trace_monitorability_evaluation	scalable-oversight-and-adversarial-ai-control (Scalable Oversight and Adversarial AI Control); adversarial-evaluation-sandbagging-and-training-time-deception (Adversarial Evaluation, Sandbagging, and Training-Time Deception)	source note available	Primary open benchmark comparator with 1,514 instances across 19 tasks and seven categories plus two stress-test settings; its reported capability/monitorability relation and up-to-30-percent degradation motivate held-out trace-action stress tests. The benchmark does not establish local monitoring quality, causal trace faithfulness, or safety.
ext_v_jepa_2_2025	V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning	Assran et al. (2025), V-JEPA 2; arXiv:2506.09985; DOI 10.48550/arXiv.2506.09985	latent_world_models_and_model_predictive_control	planning-as-a-control-layer (Planning as a Control Layer: DAGs and Intelligence Arbitrage); governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding); mathematical-and-search-substrates (Mathematical and Search Substrates); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning); integrated-reference-architecture (Integrated Reference Architecture)	source note available	Primary empirical comparator for action-free latent video pretraining, a small action-conditioned predictor, and model-predictive control. Camera sensitivity, autoregressive error accumulation, action-search cost, image-goal assumptions, and representation-bounded capability remain explicit limits; no local world model or robot-control result is established.
ext_embedded_agency_2019	Embedded Agency	Demski and Garrabrant (2019), Embedded Agency; arXiv:1902.09469; DOI 10.48550/arXiv.1902.09469	embedded_agency_foundations	asi-is-a-stack-not-a-model (ASI Is a Stack, Not a Model); evidence-states-and-claim-discipline (Evidence States and Claim Discipline); constitutional-alignment-substrate (Constitutional Alignment: Agency, Dignity, and Corrigibility); recursive-self-improvement-boundaries (Recursive Self-Improvement Boundaries); integrated-reference-architecture (Integrated Reference Architecture)	source note available	Primary informal survey of the obstacles that arise when agents are physical parts of the worlds they model, must use smaller internal models, and reason about modifiable internal parts. It supplies a foundations boundary; the book’s finite records, authority ceilings, and proofs do not solve embedded agency.
ext_ietf_rats_architecture_2023	Remote ATtestation procedureS (RATS) Architecture	Birkholz et al. (2023), RATS Architecture; source; DOI 10.17487/RFC9334	remote_attestation_architecture	confidential-and-verifiable-ai-computation (Confidential and Verifiable AI Computation); model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust)	source note available	Primary IETF architecture and terminology comparator for Attester, Verifier, Relying Party, Evidence, Attestation Results, appraisal policies, reference values, freshness, layered environments, privacy, trust roots, and confidential-model key release. It is informational architecture, not a protocol, hardware assurance level, verifier-independence result, or local attestation deployment.
ext_nist_key_management_2020	Recommendation for Key Management: Part 1 – General	Barker (2020), NIST SP 800-57 Part 1 Rev. 5; source; DOI 10.6028/NIST.SP.800-57pt1r5	cryptographic_key_management	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust)	source note available	Current final NIST key-management baseline for key and metadata protection, inventory, authorization, access control, usage periods, compromise, backup, recovery, trust anchors, and lifecycle policy. A Revision 6 draft exists, so this source is the final baseline rather than a claim that guidance has stopped evolving; no local key-management conformance or security result is established.
ext_nist_media_sanitization_2025	Guidelines for Media Sanitization	Chandramouli and Hibbard (2025), NIST SP 800-88 Rev. 2; source; DOI 10.6028/NIST.SP.800-88r2	media_sanitization_and_disposal	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust)	source note available	Current final NIST media-sanitization comparator for rendering target data access infeasible at a stated effort level using sensitivity- and media-appropriate controls, including cryptographic erase. It does not prove that all model copies, plaintext memory, cloud replicas, derivatives, or recipients were discovered or sanitized, and no local erasure test was run.
ext_attention_is_all_you_need_2017	Attention Is All You Need	Vaswani et al. (2017), Attention Is All You Need; arXiv:1706.03762; DOI 10.48550/arXiv.1706.03762	dense_attention_sequence_substrate	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary Transformer paper and dense-attention baseline. It supports the historical architecture and parallel sequence-processing comparison, not a claim that Transformers are universally optimal or locally reproduced.
ext_mamba2_ssd_2024	Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality	Dao and Gu (2024), Transformers are SSMs; arXiv:2405.21060; DOI 10.48550/arXiv.2405.21060	state_space_duality_and_sequence_substrates	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary Mamba-2/structured-state-space-duality comparator connecting SSM and attention-like formulations. No local model, kernel, quality, scaling, or hardware result is reproduced.
ext_s4_2022	Efficiently Modeling Long Sequences with Structured State Spaces	Gu, Goel, and Re (2022), Structured State Spaces; arXiv:2111.00396; DOI 10.48550/arXiv.2111.00396	structured_state_space_sequence_substrates	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Foundational S4 comparator for structured state-space sequence models, efficient long-range computation, and the lineage that precedes selective SSMs. Source-reported benchmark and generation results are not reproduced and do not establish exact recall or governed substitutability.
ext_mamba3_2026	Mamba-3: Improved Sequence Modeling using State Space Principles	Lahoti et al. (2026), Mamba-3; arXiv:2603.15569; DOI 10.48550/arXiv.2603.15569	modern_selective_state_space_sequence_substrates	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Current 2026 selective-SSM comparator for complex-valued state updates, discretization, and multi-input/multi-output formulation. Recent source-reported results are not locally reproduced and must not set the chapter conclusion by recency.
ext_gated_deltanet2_2026	Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention	Hatamizadeh, Choi, and Kautz (2026), Gated DeltaNet-2; arXiv:2605.22791; DOI 10.48550/arXiv.2605.22791	current_recurrent_linear_attention_and_editable_memory_frontier	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Dated 2026 comparator that separates erase and write gates and reports the strongest aggregate result among its Mamba-2, Gated DeltaNet, KDA, Mamba-3, and Gated DeltaNet-2 envelope at 1.3B parameters and 100B FineWeb-Edu tokens. The result is author reported, not locally reproduced; it displaces Mamba-3 only for that exact source envelope and requires official-code, checkpoint, hardware, seed, cost, retrieval, state, and transfer reproduction before any local superiority claim.
ext_hyperscale_lottery_2026	The Hyperscale Lottery: How State-Space Models Have Sacrificed Edge Efficiency	Geens et al. (2026), The Hyperscale Lottery; arXiv:2604.07935; DOI 10.48550/arXiv.2604.07935	hardware_specific_state_space_efficiency_counterevidence	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Current edge-hardware counterstudy measuring Mamba-family latency outside hyperscale-GPU conditions. Its source-reported results require platform-stratified latency, memory, and energy accounting; they are not locally reproduced and do not settle the quality-efficiency frontier.
ext_gated_deltanet_2024	Gated Delta Networks: Improving Mamba2 with Delta Rule	Yang et al. (2024), Gated Delta Networks; arXiv:2412.06464; DOI 10.48550/arXiv.2412.06464	linear_attention_and_adaptive_memory	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary gated-delta-rule comparator for targeted recurrent-memory updates, rapid erasure, parallel training, and hybrid attention/SSM compositions. Source-reported retrieval, extrapolation, efficiency, and quality results are not locally reproduced.
ext_jamba_2024	Jamba: A Hybrid Transformer-Mamba Language Model	Lieber et al. (2024), Jamba; arXiv:2403.19887; DOI 10.48550/arXiv.2403.19887	hybrid_attention_state_space_mixture_of_experts	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary large-scale hybrid Transformer-Mamba-MoE comparator. It motivates route- and composition-aware accounting; its reported quality, context, throughput, and memory results are not locally reproduced and do not establish that the specific mixture is generally optimal.
ext_neural_message_passing_2017	Neural Message Passing for Quantum Chemistry	Gilmer et al. (2017), Neural Message Passing for Quantum Chemistry; arXiv:1704.01212; DOI 10.48550/arXiv.1704.01212	graph_relational_message_passing	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); relational-dimension-compilation-and-polyadic-cognition (Relational Dimension Compilation and Polyadic Cognition)	source note available	Primary message-passing neural-network framework for learned computation over graph structure. Its molecular results motivate a non-token-native relational lane but do not establish general reasoning, dynamic graph memory, exact state, or local reproduction.
ext_hyena_hierarchy_2023	Hyena Hierarchy: Towards Larger Convolutional Language Models	Poli et al. (2023), Hyena Hierarchy; arXiv:2302.10866; DOI 10.48550/arXiv.2302.10866	long_convolution_sequence_substrates	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary long-convolution comparator for subquadratic sequence mixing and hardware-aware architecture comparisons. No local training, throughput, quality, recall, or scaling result is reproduced.
ext_rwkv_2023	RWKV: Reinventing RNNs for the Transformer Era	Peng et al. (2023), RWKV; source; DOI 10.18653/v1/2023.findings-emnlp.936	linear_recurrent_language_models	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary recurrent language-model comparator combining parallelizable training with recurrent inference. Reported benchmark, memory, and inference properties are not reproduced locally.
ext_xlstm_2024	xLSTM: Extended Long Short-Term Memory	Beck et al. (2024), xLSTM; arXiv:2405.04517; DOI 10.48550/arXiv.2405.04517	modern_gated_recurrent_sequence_substrates	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary modern-LSTM comparator for revised gating, memory, and scalable recurrent language modeling. No xLSTM training, scaling, quality, or inference result is reproduced locally.
ext_ttt_layers_2024	Learning to (Learn at Test Time): RNNs with Expressive Hidden States	Sun et al. (2024), Learning to Learn at Test Time; arXiv:2407.04620; DOI 10.48550/arXiv.2407.04620	test_time_learned_state_sequence_substrates	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary test-time-training-layer comparator that treats hidden state as a learned model updated on the sequence. It motivates explicit online-state custody and rollback; no local quality or efficiency result is reproduced.
ext_titans_2025	Titans: Learning to Memorize at Test Time	Behrouz et al. (2025), Titans; arXiv:2501.00663; DOI 10.48550/arXiv.2501.00663	test_time_neural_long_term_memory	durable-semantic-memory-and-knowledge-lattices (Durable Semantic Memory and Knowledge Lattices); replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	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.
ext_kan_2024	KAN: Kolmogorov-Arnold Networks	Liu et al. (2024), KAN; arXiv:2404.19756; DOI 10.48550/arXiv.2404.19756	learned_univariate_function_networks	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary KAN proposal replacing fixed node activations/linear edge weights with learned univariate edge functions. Interpretability and scientific-task demonstrations are source-reported and do not establish a general MLP or Transformer replacement.
ext_kan_or_mlp_fairer_comparison_2024	KAN or MLP: A Fairer Comparison	Yu et al. (2024), KAN or MLP; arXiv:2407.16674; DOI 10.48550/arXiv.2407.16674	architecture_comparison_methodology	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Critical matched-comparison source for KAN versus MLP under parameter, FLOP, and task controls. It is included to prevent architecture enthusiasm from substituting for fair accounting; no local comparison is reproduced.
ext_neural_turing_machines_2014	Neural Turing Machines	Graves, Wayne, and Danihelka (2014), Neural Turing Machines; arXiv:1410.5401; DOI 10.48550/arXiv.1410.5401	differentiable_external_memory	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary differentiable-controller/external-memory source. It motivates variable-size memory interfaces and out-of-distribution algorithmic tests; toy-task results do not establish reliable exact memory or general computation.
ext_differentiable_neural_computer_2016	Hybrid computing using a neural network with dynamic external memory	Graves et al. (2016), Differentiable Neural Computer; source; DOI 10.1038/nature20101	differentiable_external_memory	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary Differentiable Neural Computer source for learned controllers over dynamic external memory. Source-reported graph and reasoning tasks do not establish reliable exact state, scalable memory, or local reproduction.
ext_liquid_time_constant_networks_2021	Liquid Time-constant Networks	Hasani et al. (2021), Liquid Time-constant Networks; source; DOI 10.1609/aaai.v35i9.16936	continuous_time_neural_dynamics	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary continuous-time recurrent architecture source for input-dependent time constants and dynamical-system behavior. Reported time-series results and stability analysis do not establish broad cognitive superiority or a local implementation.
ext_tiny_recursive_model_2025	Less is More: Recursive Reasoning with Tiny Networks	Jolicoeur-Martineau (2025), Less is More; arXiv:2510.04871; DOI 10.48550/arXiv.2510.04871	tiny_weight_tied_recursive_reasoning	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Primary Tiny Recursive Model proposal and narrow puzzle-domain result. It motivates a compact weight-tied OneCell comparator but does not establish general reasoning, language capability, deep effective recursion, or total-system simplicity.
ext_trm_arc_agi_analysis_2025	Tiny Recursive Models on ARC-AGI-1: Inductive Biases, Identity Conditioning, and Test-Time Compute	Roye-Azar et al. (2025), Tiny Recursive Models on ARC-AGI-1; arXiv:2512.11847; DOI 10.48550/arXiv.2512.11847	recursive_model_critical_evaluation	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Critical TRM analysis reporting material dependence on 1000-sample voting, puzzle identity, and shallow effective recursion. It is a source-reported audit rather than a local reproduction and sets preregistered identity, sampling, and recursion-depth controls.
ext_tiny_autoregressive_recursive_models_2026	Tiny Autoregressive Recursive Models	Rauba, Fanconi, and van der Schaar (2026), Tiny Autoregressive Recursive Models; arXiv:2603.08082; DOI 10.48550/arXiv.2603.08082	recursive_model_controlled_ablation	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Controlled compute-matched study that progressively transforms a standard autoregressive model into a TRM-like system and reports no reliable advantage from the full autoregressive TRM mechanism. It motivates mechanism-level rather than label-level ablation.
ext_unimatrix_2026	Associative-State Universal Transformers: Sparse Retrieval Meets Structured Recurrence	Xiao (2026), Associative-State Universal Transformers; arXiv:2604.25930; DOI 10.48550/arXiv.2604.25930	structured_recurrence_and_sparse_retrieval	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Current UniMatrix preprint whose negative associative-recall result shows compressed recurrent state alone is insufficient in its setup, while explicit sparse slots and pointer-level routing materially change the result. No local reproduction or general conclusion follows.
ext_memory_caching_2026	Memory Caching: RNNs with Growing Memory	Behrouz et al. (2026), Memory Caching; arXiv:2602.24281; DOI 10.48550/arXiv.2602.24281	growing_recurrent_memory	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Current recurrent-memory comparator that caches hidden-state checkpoints and exposes a trade between fixed recurrent memory and growing addressable memory. Its source-reported recall results still leave the Transformer strongest on the reported in-context recall tasks.
ext_inkling_2026	Inkling: Our open-weights model	Thinking Machines Lab (2026), Inkling; source	hybrid_local_global_attention_moe_multimodal_substrate	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Release-day primary-source case study of a 66-layer multimodal sparse-MoE Transformer with a 5:1 local/global attention schedule, relative positions, short convolutions, controllable effort, and open weights. It motivates topology-complete capability cards and component ablations; provider-reported results are not locally reproduced and do not isolate the contribution of any component.
ext_megatron_distributed_training_2021	Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM	Narayanan et al. (2021), Efficient Large-Scale Language Model Training; arXiv:2104.04473; DOI 10.1145/3458817.3476209	governed_distributed_model_training_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); learning-compute-topology-and-adaptive-process-architecture (Learning–Compute Topology and Adaptive Process Architecture)	source note available	Primary composed-parallelism mechanism source. It grounds tensor, pipeline, and data parallel interactions, strict optimizer semantics, microbatch and topology tradeoffs. Reported trillion-parameter and throughput results are configuration-bound and not locally reproduced.
ext_zero_optimizer_2019	ZeRO: Memory Optimizations Toward Training Trillion Parameter Models	Rajbhandari et al. (2020), ZeRO; arXiv:1910.02054; DOI 10.1109/SC41405.2020.00024	governed_distributed_model_training_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary competing sharded-state design for optimizer, gradient, parameter, activation, and residual memory. It motivates explicit state closure and shard reconstruction; source-reported scale and speed are not locally reproduced or treated as universal superiority.
ext_gspmd_2021	GSPMD: General and Scalable Parallelization for ML Computation Graphs	Xu et al. (2021), GSPMD; arXiv:2105.04663; DOI 10.48550/arXiv.2105.04663	governed_distributed_model_training_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); learning-compute-topology-and-adaptive-process-architecture (Learning–Compute Topology and Adaptive Process Architecture)	source note available	Primary compiler-mediated competing design for general SPMD sharding and mixed parallelism. It motivates versioning inferred plans and inserted collectives; reported TPU utilization and scaling are not locally reproduced.
ext_datastates_llm_2024	DataStates-LLM: Lazy Asynchronous Checkpointing for Large Language Models	Maurya et al. (2024), DataStates-LLM; arXiv:2406.10707; DOI 10.1145/3625549.3658685	governed_distributed_model_training_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary limitation and checkpoint-mechanism source for asynchronous multi-level copies, distributed shard consistency, and checkpoint overhead. It does not establish complete application state or exact trajectory-equivalent resume, and no result is locally reproduced.
ext_pytorch_distributed_checkpoint_2026	Distributed Checkpoint — PyTorch documentation	PyTorch (2026), Distributed Checkpoint; source	governed_distributed_model_training_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Official current implementation documentation for SPMD save/load, asynchronous completion, canonical model and optimizer state, resharding, strict load, and call-order constraints. Documentation is not benchmark or full-state resume evidence.
ext_mlperf_training_v6_2026	MLPerf Training v6.0	MLCommons (2026), MLPerf Training v6.0; source	governed_distributed_model_training_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); resource-economics-and-token-budgets (Resource Economics and Token Budgets); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	Official current measurement comparator for fixed datasets and quality targets, repeated time-to-quality, system metadata, divisions, variance, and corrected results. No MLPerf run is performed and the benchmark does not establish safety or complete run integrity.
ext_adam_2015	Adam: A Method for Stochastic Optimization	Kingma and Ba (2015), Adam; arXiv:1412.6980; DOI 10.48550/arXiv.1412.6980	optimizer_mechanisms_and_selection	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary Adam mechanism source for bias-corrected first- and second-moment estimates and coordinate-wise adaptive updates. Its online-convex analysis and reported experiments do not establish universal convergence, quality, or optimizer superiority in foundation-model training.
ext_amsgrad_2018	On the Convergence of Adam and Beyond	Reddi, Kale, and Kumar (2018), On the Convergence of Adam and Beyond; source	optimizer_failure_and_convergence	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary Adam failure and AMSGrad source. It gives constructed stochastic-convex non-convergence cases and a maximum-second-moment remedy; those cases do not imply every practical Adam run fails or that AMSGrad is universally preferable.
ext_adamw_2019	Decoupled Weight Decay Regularization	Loshchilov and Hutter (2019), Decoupled Weight Decay Regularization; arXiv:1711.05101; DOI 10.48550/arXiv.1711.05101	optimizer_mechanisms_and_selection	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary AdamW source separating weight decay from the adaptive gradient update. Its reported tuning and generalization results are setting-bound; the optimizer name alone does not specify parameter exclusions, schedule, decay scaling, or implementation semantics.
ext_adafactor_2018	Adafactor: Adaptive Learning Rates with Sublinear Memory Cost	Shazeer and Stern (2018), Adafactor; arXiv:1804.04235; DOI 10.48550/arXiv.1804.04235	optimizer_memory_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary factored-second-moment optimizer source. It reduces auxiliary state for matrix parameters and adds update clipping and parameter-scale rules; factorization remains an approximation and its reported translation result does not establish universal parity with Adam.
ext_lamb_2019	Large Batch Optimization for Deep Learning: Training BERT in 76 minutes	You et al. (2020), Large Batch Optimization for Deep Learning; arXiv:1904.00962; DOI 10.48550/arXiv.1904.00962	optimizer_memory_and_scaling	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary LAMB source for layer-wise trust ratios and large-batch optimization. Its reported BERT time-to-target is tied to model, batch, hardware, quality target, and tuning conditions and is not a universal large-batch or wall-clock result.
ext_shampoo_2018	Shampoo: Preconditioned Stochastic Tensor Optimization	Gupta, Koren, and Singer (2018), Shampoo; source	tensor_and_matrix_preconditioning	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary tensor-structured preconditioning source. Shampoo maintains per-dimension preconditioners and reports faster convergence with practical per-step cost in studied models; stochastic-convex theory and source experiments do not settle current distributed lifecycle cost.
ext_kfac_2015	Optimizing Neural Networks with Kronecker-factored Approximate Curvature	Martens and Grosse (2015), K-FAC; source	curvature_aware_optimization	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary K-FAC source for an efficiently invertible Kronecker-factored approximation to the Fisher matrix. Its approximation, damping, inversion, and empirical cost-benefit are architecture- and implementation-dependent.
ext_lion_2023	Symbolic Discovery of Optimization Algorithms	Chen et al. (2023), Symbolic Discovery of Optimization Algorithms; arXiv:2302.06675; DOI 10.48550/arXiv.2302.06675	optimizer_discovery_and_sign_updates	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary Lion and symbolic optimizer-search source. Lion uses sign-based momentum and one optimizer-state tensor; the paper also reports method-specific learning-rate behavior and settings where gains are small or insignificant.
ext_sophia_2023	Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training	Liu et al. (2023), Sophia; arXiv:2305.14342; DOI 10.48550/arXiv.2305.14342	curvature_aware_optimization	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary Sophia source for periodic diagonal-Hessian estimation and clipped curvature-aware updates. Its reported GPT pretraining speedups and simplified theory require matched reproduction before any broader optimizer claim.
ext_soap_2024	SOAP: Improving and Stabilizing Shampoo using Adam	Vyas et al. (2024), SOAP; arXiv:2409.11321; DOI 10.48550/arXiv.2409.11321	tensor_and_matrix_preconditioning	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary SOAP source connecting Shampoo to adaptive moments in a changing preconditioner eigenbasis. Reported large-batch pretraining gains remain tied to 360M/660M models, preconditioning frequency, overhead, and tuning conditions.
ext_schedule_free_2024	The Road Less Scheduled	Defazio et al. (2024), The Road Less Scheduled; arXiv:2405.15682; DOI 10.48550/arXiv.2405.15682	optimizer_scheduling_and_averaging	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary schedule-free optimization source unifying scheduling and iterate averaging without requiring a stopping step. Removing a stopping-time schedule does not remove learning-rate, warmup, evaluation-iterate, checkpoint, or method-selection choices.
ext_mup_2022	Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer	Yang et al. (2022), Tensor Programs V; arXiv:2203.03466; DOI 10.48550/arXiv.2203.03466	optimizer_parametrization_and_scale_transfer	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary maximal-update parametrization and muTransfer source. It reports widthwise hyperparameter transfer under a prescribed parametrization on Transformer and ResNet settings; it does not establish arbitrary depth, duration, optimizer, or architecture transfer.
ext_modular_norm_2024	Scalable Optimization in the Modular Norm	Large et al. (2024), Scalable Optimization in the Modular Norm; arXiv:2405.14813; DOI 10.48550/arXiv.2405.14813	optimizer_parametrization_and_scale_transfer	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary modular-norm source defining architecture-recursive update geometry and reporting learning-rate transfer across width and depth. Its well-behaved-module assumptions and experiments do not prove arbitrary substrate transfer or optimizer superiority.
ext_muon_scalable_2025	Muon is Scalable for LLM Training	Liu et al. (2025), Muon is Scalable for LLM Training; arXiv:2502.16982; DOI 10.48550/arXiv.2502.16982	orthogonalized_matrix_optimization	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary large-scale Muon source for momentum plus matrix orthogonalization, weight decay, per-parameter update scaling, and a distributed implementation. Its reported compute-efficiency and Moonlight results are source-scoped and not locally reproduced.
ext_muon_spectral_norm_2026	Muon Optimizes Under Spectral Norm Constraints	Chen, Li, and Liu (2026), Muon Optimizes Under Spectral Norm Constraints; source	orthogonalized_matrix_optimization_theory	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Accepted TMLR theoretical source placing Muon with decoupled weight decay in a Lion-K/nuclear-norm framework and deriving implicit spectral-norm constraint behavior. The interpretation does not establish task-level quality, safety, or universal advantage.
ext_nist_privacy_framework_2020	NIST Privacy Framework: A Tool for Improving Privacy through Enterprise Risk Management, Version 1.0	NIST (2020), Privacy Framework 1.0; source; DOI 10.6028/NIST.CSWP.01162020	privacy_data_rights_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Official paper-body-reviewed risk framework distinguishing privacy problems from cybersecurity incidents across the data lifecycle. It is voluntary, has no force of law, and supplies no local privacy outcome or certification.
ext_eu_gdpr_2016	Regulation (EU) 2016/679 (General Data Protection Regulation)	European Parliament and Council (2016), Regulation (EU) 2016/679; source	privacy_data_rights_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Authoritative jurisdiction-specific normative comparator for principles, bases, rights, accountability, design, and qualified exceptions. It is not universal law, legal advice, an applicability decision, or local compliance evidence.
ext_w3c_dpv_2024	Data Privacy Vocabulary (DPV), Version 2	W3C Data Privacy Vocabularies and Controls CG (2024), DPV v2; source	privacy_data_rights_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Machine-readable vocabulary for purpose, processing, data, actors, rights, risks, measures, legal basis, and consent. It is a Community Group Final Specification, not a W3C Recommendation, law, or enforcement proof.
ext_abadi_dpsgd_2016	Deep Learning with Differential Privacy	Abadi et al. (2016), Deep Learning with Differential Privacy; arXiv:1607.00133; DOI 10.1145/2976749.2978318	privacy_data_rights_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Primary DP-SGD mechanism and accounting source. Its algorithm, analysis, and reported experiments are not locally reproduced; its guarantee is parameter-, unit-, adjacency-, implementation-, and release-surface-bound.
ext_algospec_purpose_limitation_2024	Being Transparent Is Merely the Beginning: Enforcing Purpose Limitation with Polynomial Approximation	Liu et al. (2024), Being Transparent Is Merely the Beginning; source	privacy_data_rights_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Primary competing purpose-restriction design using algorithm-specific polynomial approximation. Reported accuracy and efficiency are bounded to studied algorithms/data and are not locally reproduced or a complete legal-purpose result.
ext_carlini_training_data_extraction_2021	Extracting Training Data from Large Language Models	Carlini et al. (2021), Extracting Training Data from Large Language Models; source	privacy_data_rights_and_information_flow_governance	adversarial-machine-learning-and-model-attack-surface (Adversarial Machine Learning and the Model Attack Surface); privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	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.
ext_choquette_choo_label_only_mia_2021	Label-Only Membership Inference Attacks	Choquette-Choo et al. (2021), Label-Only Membership Inference Attacks; source	privacy_data_rights_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Primary failure source showing hard-label robustness can expose membership and confidence masking can be insufficient in studied settings. No attack or defense result is locally reproduced or universal.
ext_mahloujifar_fdp_audit_2025	Auditing f-Differential Privacy in One Run	Mahloujifar, Melis, and Chaudhuri (2025), Auditing f-Differential Privacy in One Run; source	privacy_data_rights_and_information_flow_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance)	source note available	Primary empirical-audit comparator using randomized inclusion and an f-DP hypothesis in one run. A passed audit is not proof that DP or lifecycle privacy holds, and no result is locally reproduced.
ext_airllm_2023	AirLLM: Scaling Large Language Models on Low-End Commodity Computers	Li (2023-2026), AirLLM; source	heterogeneous_inference_memory	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Official implementation comparator for layer-wise model sharding, one-layer accelerator residency, next-layer prefetch, optional storage compression, and original-versus-transformed model storage. Maintainer-reported fit and speed claims are not independently reproduced.
ext_deepspeed_inference_2022	DeepSpeed Inference: Enabling Efficient Inference of Transformer Models at Unprecedented Scale	Aminabadi et al. (2022), DeepSpeed Inference; arXiv:2207.00032; DOI 10.48550/arXiv.2207.00032	heterogeneous_inference_memory	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary heterogeneous-inference systems source spanning GPU, CPU, and NVMe for dense and sparse Transformer inference. Reported latency, throughput, scale, and model-fit results remain source-scoped and unreproduced.
ext_flexgen_2023	FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU	Sheng et al. (2023), FlexGen; arXiv:2303.06865; DOI 10.48550/arXiv.2303.06865	heterogeneous_inference_memory	personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary planned-placement source for GPU/CPU/disk tensor storage and access, batching, and optional weight/cache compression under latency-insensitive workloads. Its throughput results are not interactive-latency or local evidence.
ext_hf_accelerate_big_model_inference_2026	Hugging Face Accelerate: Loading Big Models into Memory	Hugging Face (2026), Accelerate Big Model Inference documentation; source	heterogeneous_inference_memory	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); virtual-context-abi (The Virtual Context ABI: Typed Pages, Cells, and Certificates); personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Official implementation documentation for automatic or explicit GPU/CPU/disk device maps and memory-mapped disk tensors. The documented sequential-dispatch, prefetch, and hard-drive-performance limitations make it a baseline, not a qualification result.
ext_llama_cpp_memory_mapping_2026	llama.cpp CLI Memory Mapping, Tensor Placement, and KV Offload Controls	ggml-org (2026), llama.cpp CLI documentation; source	heterogeneous_inference_memory	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Official consumer-runtime documentation for model load modes, memory mapping, DirectIO, GPU-layer and tensor placement, MoE CPU placement, KV offload, and KV data types. No local model or performance result is implied.
ext_llm_in_flash_2024	LLM in a Flash: Efficient Large Language Model Inference with Limited Memory	Alizadeh et al. (2024), LLM in a Flash; arXiv:2312.11514; DOI 10.48550/arXiv.2312.11514	heterogeneous_inference_memory	model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary flash-aware inference source for on-demand parameter loading, I/O cost modeling, transfer reduction, contiguous reads, windowing, and row-column bundling. Sparse/context-adaptive loading is not an exact dense paging result.
ext_powerinfer_2024	PowerInfer: Fast Large Language Model Serving with a Consumer-Grade GPU	Song et al. (2024), PowerInfer; arXiv:2312.12456; DOI 10.48550/arXiv.2312.12456	heterogeneous_inference_memory	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary consumer-inference source for source-reported power-law neuron locality, hot-GPU/cold-CPU placement, adaptive predictors, and sparse operators. Architecture transfer and performance are not locally reproduced.
ext_vattention_2025	vAttention: Dynamic Memory Management for Serving LLMs without PagedAttention	Prabhu et al. (2025), vAttention; arXiv:2405.04437; DOI 10.48550/arXiv.2405.04437	heterogeneous_inference_memory	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary counterpoint to non-contiguous PagedAttention layouts: decouples virtual and physical GPU memory while retaining contiguous KV virtual addresses. Reported serving results remain source-scoped.
ext_infinigen_2024	InfiniGen: Efficient Generative Inference of Large Language Models with Dynamic KV Cache Management	Lee et al. (2024), InfiniGen; arXiv:2406.19707; DOI 10.48550/arXiv.2406.19707	heterogeneous_inference_memory	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary speculative-KV-prefetch source using minimal rehearsal and partial next-layer state to select host-resident KV entries. Prediction, quality, miss, and fallback results are not locally reproduced.
ext_specache_2025	SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs	Jie et al. (2025), SpeCache; arXiv:2503.16163; DOI 10.48550/arXiv.2503.16163	heterogeneous_inference_memory	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary speculative-KV-prefetch source keeping complete KV state in CPU memory, a low-bit importance copy in VRAM, and predicted next-step KV transfers. Source-reported quality and memory results are unreproduced.
ext_specoffload_2025	SpecOffload: Unlocking Latent GPU Capacity for LLM Inference on Resource-Constrained Devices	Zhuge et al. (2025), SpecOffload; arXiv:2505.10259; DOI 10.48550/arXiv.2505.10259	heterogeneous_inference_memory	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary composition source for target-model offloading, draft-model placement, speculative decoding, and joint tensor/decoding planning. It is not speculative physical-page prediction, and reported results are unreproduced.
ext_atsinfer_2026	Automated Tensor Scheduling for Hybrid CPU-GPU LLM Inference on Consumer Devices	Liu et al. (2026), ATSInfer; arXiv:2607.10183; DOI 10.48550/arXiv.2607.10183	heterogeneous_inference_memory	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); personal-compute-hives-and-federated-edge-intelligence (Personal Compute Hives and Federated Edge Intelligence); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Very recent preprint comparator for tensor-granular static placement, load-aware dynamic transfer, and asynchronous CPU-GPU coordination on consumer devices. Only abstract/metadata were reviewed; reported results are provisional and unreproduced.
ext_openai_prompt_caching_docs_2026	Prompt Caching	OpenAI (2026), Prompt Caching; source	inference_cache_reuse	context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Current official provider contract for exact-prefix prompt caching, cache-write and cache-read metering, usage receipts, retention, organization isolation, and rate-limit boundaries. Product behavior and prices are time-sensitive; inspected 2026-07-23.
ext_anthropic_prompt_caching_docs_2026	Prompt caching	Anthropic (2026), Prompt caching; source	inference_cache_reuse	context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Current official provider contract for reusable prompt prefixes, explicit cache breakpoints, five-minute and one-hour lifetimes, cache creation and read metering, and prewarming. Product behavior and prices are time-sensitive; inspected 2026-07-23.
ext_gemini_context_caching_docs_2026	Context caching	Google (2026), Gemini Context Caching; source	inference_cache_reuse	context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Current official provider contract for implicit and explicit context caching, common-prefix placement, cached-token usage reporting, time-to-live, and storage charges. Product behavior and prices are time-sensitive; inspected 2026-07-23.
ext_vllm_automatic_prefix_caching_2026	Automatic Prefix Caching	vLLM (2026), Automatic Prefix Caching; source	inference_cache_reuse	context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Official vLLM design documentation for block-hash exact-prefix KV reuse, least-recently-used eviction, multi-modal and adapter identity, and tenant cache-salt protection against timing inference. No local serving benchmark was run.
ext_sglang_radixattention_2024	SGLang: Efficient Execution of Structured Language Model Programs	Zheng et al. (2024), SGLang; arXiv:2312.07104; DOI 10.48550/arXiv.2312.07104	inference_cache_reuse	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary RadixAttention and cache-aware scheduling source for structured multi-call language-model programs. Source-reported throughput and theorem scope remain unreproduced.
ext_prompt_cache_2024	Prompt Cache: Modular Attention Reuse for Low-Latency Inference	Gim et al. (2024), Prompt Cache; arXiv:2311.04934; DOI 10.48550/arXiv.2311.04934	inference_cache_reuse	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary MLSys source for schema-defined reusable prompt modules, positional accuracy, and attention-state reuse across prompts. Source-reported latency remains unreproduced.
ext_mooncake_2025	Mooncake: Trading More Storage for Less Computation — A KVCache-centric Architecture for Serving LLM Chatbot	Qin et al. (2025), Mooncake; source	inference_cache_reuse	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary FAST 2025 source for a KV-cache-centric disaggregated serving architecture spanning prefill, decode, DRAM, SSD, and network resources. Production-trace and capacity results remain source-reported.
ext_cacheblend_2025	CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion	Yao et al. (2025), CacheBlend; arXiv:2405.16444; DOI 10.48550/arXiv.2405.16444	inference_cache_reuse	fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary source for non-prefix and multi-chunk KV reuse with selective recomputation. It makes the cross-attention failure of naïve independent-chunk KV fusion explicit. Source-reported latency and quality remain unreproduced.
ext_azure_llm_semantic_cache_2026	Azure API Management LLM semantic cache lookup policy	Microsoft (2026), LLM semantic cache lookup policy; source	inference_cache_reuse	context-transactions-snapshots-mounts-and-taint (Context Transactions, Snapshots, Mounts, and Taint); fast-generation-architectures (Fast Generation Architectures); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Official semantic-response-cache policy documentation. It treats vector similarity as an approximate response-reuse decision and warns that a hit can return an incorrect, outdated, or unsafe answer. No local semantic-cache deployment was run.
ext_nist_adversarial_ml_2024	Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations	Vassilev et al. (2024), NIST AI 100-2; source; DOI 10.6028/NIST.AI.100-2e2023	adversarial_machine_learning	adversarial-machine-learning-and-model-attack-surface (Adversarial Machine Learning and the Model Attack Surface)	source note available	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.
ext_singapore_consensus_2026	The 2026 Singapore Consensus on Global AI Safety Research Priorities	Singapore Consensus (2026), Global AI Safety Research Priorities; source	dangerous_capability_assessment_societal_resilience_and_agentic_risk	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift); military-ai-autonomous-weapons-and-strategic-stability (Military AI, Autonomous Weapons, and Strategic Stability); societal-resilience-and-misuse-defense (Societal Resilience and Misuse Defense); open-weight-release-and-post-release-control (Open-Weight Release and Post-Release Control)	source note available	International 2026 technical-research-priority synthesis covering risk assessment, development, control, and societal resilience, including CBRN, cyber, psychological manipulation, malicious fine-tuning, agent monitoring, incident reporting, and defense-favoring capabilities. It is a research agenda and consensus synthesis, not evidence that any listed safeguard works or that this book’s contracts are complete.
ext_international_ai_safety_report_2026	International AI Safety Report 2026	International AI Safety Report (2026); source	frontier_ai_risk_misuse_open_weight_and_societal_resilience	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift); military-ai-autonomous-weapons-and-strategic-stability (Military AI, Autonomous Weapons, and Strategic Stability); societal-resilience-and-misuse-defense (Societal Resilience and Misuse Defense); open-weight-release-and-post-release-control (Open-Weight Release and Post-Release Control); content-authenticity-watermarking-and-synthetic-media-integrity (Content Authenticity, Watermarking, and Synthetic Media Integrity)	source note available	International expert report synthesizing evidence on general-purpose AI capabilities, misuse, open-weight risks, safeguards, monitoring, and societal resilience. It supports risk taxonomy and uncertainty boundaries; its literature synthesis does not reproduce component studies locally or establish that any ASI Stack mechanism is effective.
ext_c2pa_specification_2_3_2025	C2PA Content Credentials Technical Specification 2.3	C2PA (2025), Content Credentials Specification 2.3; source	content_provenance_and_authenticity	ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance); content-authenticity-watermarking-and-synthetic-media-integrity (Content Authenticity, Watermarking, and Synthetic Media Integrity)	source note available	Official C2PA specification for signed manifests, assertions, ingredients, content bindings, validation, and provenance history. It provides a concrete interoperability comparator; it does not prove truth of depicted events, creator identity beyond the credential chain, semantic authenticity, universal platform retention, or resistance to removal and laundering.
ext_eu_article_50_transparency_guidelines_2026	Guidelines on Transparency Obligations for Providers and Deployers of AI Systems	European Commission (2026), Article 50 Transparency Guidelines; source	synthetic_content_transparency_and_disclosure	institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy); content-authenticity-watermarking-and-synthetic-media-integrity (Content Authenticity, Watermarking, and Synthetic Media Integrity)	source note available	European Commission guidance for Article 50 transparency obligations concerning AI interaction, machine-readable marking, deepfakes, and certain public-interest text, with obligations applying from 2 August 2026 subject to scope and transitional details. It is legal and implementation guidance, not legal advice, proof of compliance, or evidence that a marking technique is robust.
ext_openai_worst_case_open_weight_risks_2025	Estimating Worst-Case Frontier Risks of Open-Weight LLMs	Wallace et al. (2025), Worst-Case Open-Weight Frontier Risks; source	malicious_fine_tuning_and_open_weight_release_evaluation	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift); open-weight-release-and-post-release-control (Open-Weight Release and Post-Release Control)	source note available	Provider-authored study of malicious fine-tuning for biology and cyber evaluations before the gpt-oss release. It supplies a concrete worst-case-elicitation comparator and reports bounded provider results; it does not prove future-release safety, general malicious-fine-tuning resistance, independent reproduction, or absence of untested harms.
ext_aisi_misuse_safeguards_safety_case_2026	An Example Safety Case for Safeguards Against Misuse	Clymer et al. (2026), Misuse Safeguards Safety Case; source	misuse_safeguard_uplift_and_safety_cases	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift); societal-resilience-and-misuse-defense (Societal Resilience and Misuse Defense)	source note available	UK AI Security Institute example connecting safeguard red teaming, attacker effort, an uplift model, and a deployment safety case. It is a worked argument and measurement proposal, not proof that real safeguards reduce misuse to a particular level or that the book’s proposed defense contracts work.
ext_anthropic_responsible_scaling_policy_3_4_2026	Anthropic Responsible Scaling Policy 3.4	Anthropic (2026), Responsible Scaling Policy 3.4; source	frontier_capability_thresholds_and_safeguards	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift); open-weight-release-and-post-release-control (Open-Weight Release and Post-Release Control)	source note available	Current provider policy comparator linking capability thresholds and safeguards across CBRN and automated R&D threat models, with public risk-report and review commitments. It is a revocable provider policy and self-described governance mechanism, not independent evidence that thresholds are complete, evaluations are sensitive, or safeguards are effective.
ext_aisi_frontier_ai_trends_2025	AISI Frontier AI Trends Report 2025	UK AISI (2025), Frontier AI Trends Report; source	frontier_capability_evaluation_trends	dangerous-capability-domains-and-misuse-uplift (Dangerous Capability Domains and Misuse Uplift)	source note available	UK AI Security Institute synthesis of evaluations across offensive cyber, dual-use chemistry and biology, autonomous systems, and societal impacts. It is an institute-reported trend record with bounded methods and coverage, not a complete threat census or local reproduction.
ext_valiant_theory_learnable_1984	A Theory of the Learnable	Valiant (1984), A Theory of the Learnable; source; DOI 10.1145/1968.1972	computational_learning_theory_and_sample_complexity	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	Foundational PAC-learning source for defining learnability through accuracy, confidence, resource, hypothesis, and data assumptions. Its distributional and concept-class assumptions do not directly explain modern foundation-model generalization or certify a trained model.
ext_deep_double_descent_2020	Deep Double Descent: Where Bigger Models and More Data Hurt	Nakkiran et al. (2020), Deep Double Descent; source; DOI 10.1088/1742-5468/ac3a74	generalization_and_interpolation_regimes	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	ICLR 2020 empirical study reporting model-wise, sample-wise, and epoch-wise double-descent phenomena and proposing effective model complexity. The phenomenon is configuration- and regime-bound and does not imply that larger models or more data generally hurt or help.
ext_emergent_abilities_llms_2022	Emergent Abilities of Large Language Models	Wei et al. (2022), Emergent Abilities of Large Language Models; source	scaling_and_emergent_capability_measurement	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	TMLR survey and framing of task abilities that appear sharply at larger model scales under reported evaluations. It motivates prospective scaling measurement but does not establish that every apparent threshold is mechanistically discontinuous or unpredictable.
ext_emergent_abilities_mirage_2023	Are Emergent Abilities of Large Language Models a Mirage?	Schaeffer et al. (2023), Emergent Abilities a Mirage?; source	metric_induced_emergence_and_scaling_measurement	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	NeurIPS 2023 counterevidence showing that nonlinear or discontinuous metrics and limited test data can produce apparently sharp emergence from smoother underlying changes in studied settings. It does not prove that all emergence is a metric artifact.
ext_elk_report_2021	Eliciting Latent Knowledge	Christiano and Xu (2021), Eliciting Latent Knowledge; source	latent_knowledge_and_ontology_identification	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance)	source note available	ARC technical-report agenda on mapping between a model’s world model and human concepts when ordinary supervision may reward convincing but false reports. It defines an open problem and candidate approaches, not a solved elicitation method or evidence that a deployed model’s reports are truthful.
ext_influence_functions_2017	Understanding Black-box Predictions via Influence Functions	Koh and Liang (2017), Influence Functions; source	training_data_attribution_and_influence	white-box-evidence-interpretability-and-activation-governance (White-Box Evidence, Interpretability, and Activation Governance); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	ICML 2017 source tracing predictions through a learning algorithm to influential training points using influence-function approximations. The theory and approximations have model and optimization assumptions and do not establish exact causal provenance, privacy erasure, or influence removal in foundation models.
ext_flexible_hardware_enabled_guarantees_2025	Flexible Hardware-Enabled Guarantees for AI Compute	Petrie et al. (2025), Flexible Hardware-Enabled Guarantees; source; DOI 10.48550/arXiv.2506.15093	hardware_enabled_governance_and_compute_attestation	institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy); model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust); physical-compute-infrastructure-energy-and-environmental-constraints (Physical Compute Infrastructure, Energy, and Environmental Constraints)	source note available	Design proposal for auditable guarantee processors and tamper-resistant enclosures supporting privacy-preserving verification or enforcement of AI-compute claims. It is a proposed architecture with adoption, legacy-hardware, update-authority, side-channel, sovereignty, and abuse risks; no local device or governance guarantee exists.
ext_proof_of_learning_2021	Proof-of-Learning: Definitions and Practice	Jia et al. (2021), Proof-of-Learning; source; DOI 10.48550/arXiv.2103.05633	training_provenance_and_computation_attestation	ai-supply-chain-integrity-and-lifecycle-provenance (AI Supply-Chain Integrity and Lifecycle Provenance); governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Research proposal for proving that final parameters arose through a claimed iterative learning process using checkpoint and stochastic-training evidence. Later attacks and security work show that proof-of-learning/proof-of-training claims require adversarial review; the source does not prove data rights, objective legitimacy, clean training, or model safety.
ext_test_time_training_2020	Test-Time Training with Self-Supervision for Generalization under Distribution Shifts	Sun et al. (2020), Test-Time Training; source	test_time_adaptation_and_online_update	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	ICML 2020 method adapting model parameters on each test sample using a self-supervised objective and reporting improvements on studied image-corruption benchmarks. The result is task- and method-bound and does not establish safe online adaptation, resistance to poisoning, or benefit under arbitrary shift.
ext_legal_alignment_2026	Legal Alignment for Safe and Ethical AI	Kolt et al. (2026), Legal Alignment for Safe and Ethical AI; source; DOI 10.48550/arXiv.2601.04175	law_following_ai_and_legal_alignment	constitutional-alignment-substrate (Constitutional Alignment: Agency, Dignity, and Corrigibility); institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy)	source note available	2026 interdisciplinary agenda for using legal rules, methods of interpretation, and institutional structures in AI alignment. Law is jurisdictional, contested, changing, and sometimes unjust or conflicting; the source does not establish that legal compliance equals moral alignment or that a model can reliably determine applicable law.
ext_curriculum_learning_2009	Curriculum Learning	Bengio et al. (2009), Curriculum Learning; source; DOI 10.1145/1553374.1553380	training_curricula_and_example_order	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); data-engines-continual-learning-and-unlearning (Data Engines, Continual Learning, and Unlearning)	source note available	ICML 2009 source proposing training curricula that begin with easier examples or concepts and increase difficulty. Reported benefits are problem- and curriculum-bound; ordering can introduce bias, hide hard cases, or create capability and safety regressions.
ext_causal_calculus_1995	A Causal Calculus for Statistical Research	Pearl (1995), A Causal Calculus for Statistical Research; source	structural_causal_models_and_intervention_identification	governed-world-models-and-reality-grounding (Governed World Models and Reality Grounding)	source note available	Foundational do-calculus source distinguishing intervention from observation under an explicit structural causal model. Identification depends on the causal graph and assumptions; the calculus does not discover the correct graph from arbitrary data or establish that a learned world model is causally valid.
ext_ai_simulation_digital_twins_2025	AI Simulation by Digital Twins: Systematic Survey, Reference Framework, and Mapping to a Standardized Architecture	Liu et al. (2025), AI Simulation by Digital Twins; source; DOI 10.48550/arXiv.2506.06580	digital_twins_and_simulation_fidelity	embodied-agency-real-time-control-and-physical-safety (Embodied Agency, Real-Time Control, and Physical Safety)	source note available	Systematic survey and reference framework for digital-twin-enabled AI simulation. It supports explicit virtual/physical synchronization and simulation roles; it does not establish that a digital twin is faithful, safe for policy transfer, or an adequate substitute for physical testing.
ext_nist_privacy_enhancing_cryptography_2026	Privacy-Enhancing Cryptography	NIST (2026), Privacy-Enhancing Cryptography; source	confidential_computation_and_privacy_enhancing_cryptography	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance); confidential-and-verifiable-ai-computation (Confidential and Verifiable AI Computation)	source note available	NIST program material distinguishing fully homomorphic encryption, secure multiparty computation, zero-knowledge proofs, private-set intersection, and related privacy-enhancing techniques. It provides terminology and use-case context, not implementation security, usable performance, authorization, or end-to-end privacy.
ext_zkllm_2024	zkLLM: Zero Knowledge Proofs for Large Language Models	Sun et al. (2024), zkLLM; source; DOI 10.48550/arXiv.2404.16109	verifiable_private_model_inference	confidential-and-verifiable-ai-computation (Confidential and Verifiable AI Computation)	source note available	Research prototype for proving bounded LLM inference claims while hiding model parameters. Reported proof size and latency are configuration-bound and do not establish semantic correctness, authorization, side-channel security, production readiness, or end-to-end privacy.
ext_human_ai_team_meta_analysis_2024	When combinations of humans and AI are useful: A systematic review and meta-analysis	Vaccaro et al. (2024), When combinations of humans and AI are useful; source; DOI 10.1038/s41562-024-02024-1	human_ai_complementarity_and_team_baselines	human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability); human-ai-symbiosis-neurotechnology-and-cognitive-sovereignty (Human-AI Symbiosis, Neurotechnology, and Cognitive Sovereignty)	source note available	Preregistered synthesis of 106 experiments and 370 effect sizes using human-alone, AI-alone, and combined-system comparisons. The aggregate findings are task- and population-bound and do not establish universal human-AI synergy or longitudinal benefit.
ext_human_ai_feedback_loops_2025	Human-AI feedback loops alter human perceptual, emotional and social judgements	Glickman and Sharot (2025), Human-AI feedback loops; source; DOI 10.1038/s41562-024-02077-2	longitudinal_human_ai_coupling_and_bias_amplification	human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability); human-ai-symbiosis-neurotechnology-and-cognitive-sovereignty (Human-AI Symbiosis, Neurotechnology, and Cognitive Sovereignty)	source note available	Experimental evidence that repeated human-AI interaction can create feedback dynamics in studied judgment tasks. It supports measuring coupled trajectories, not a universal claim about all users, systems, settings, or long-term clinical outcomes.
ext_oecd_neuro_ai_convergence_2025	Technology convergence: Trends, prospects and policies	OECD (2025), Technology convergence; source	neurotechnology_ai_convergence_and_governance	human-ai-symbiosis-neurotechnology-and-cognitive-sovereignty (Human-AI Symbiosis, Neurotechnology, and Cognitive Sovereignty)	source note available	OECD policy synthesis on converging technologies including AI and neurotechnology. It motivates cross-domain governance and anticipatory capacity but is not a clinical trial, technical validation, or proof of beneficial convergence.
ext_who_neurotechnology_landscape_2025	Landscape analysis of the opportunities and challenges for neurotechnology in global health	WHO (2025), Neurotechnology landscape analysis; source	neurotechnology_health_equity_and_governance	privacy-data-rights-and-information-flow-governance (Privacy, Data Rights, and Information-Flow Governance); human-ai-symbiosis-neurotechnology-and-cognitive-sovereignty (Human-AI Symbiosis, Neurotechnology, and Cognitive Sovereignty)	source note available	WHO landscape analysis of neurotechnology opportunities, risks, governance questions, and global-health distribution. It supports a rights and equity boundary, not device efficacy, individual medical advice, or authorization for neural-data collection.
ext_icrc_autonomous_weapons_ihl_2025	Autonomous Weapon Systems and International Humanitarian Law: Selected Issues	ICRC (2025), Autonomous Weapon Systems and IHL; source	autonomous_weapons_human_judgment_and_ihl	military-ai-autonomous-weapons-and-strategic-stability (Military AI, Autonomous Weapons, and Strategic Stability); institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy)	source note available	ICRC legal and policy position on autonomous weapon systems and context-specific human judgment. It is authoritative for the ICRC position, not a universally settled legal interpretation, engineering validation, or authorization to design or deploy weapons.
ext_sipri_military_ai_nuclear_escalation_2025	The Impact of Military Artificial Intelligence on Nuclear Escalation Risk	Boulanin et al. (2025), Military AI and Nuclear Escalation Risk; source; DOI 10.55163/FZIW8544	military_ai_crisis_dynamics_and_nuclear_escalation	military-ai-autonomous-weapons-and-strategic-stability (Military AI, Autonomous Weapons, and Strategic Stability)	source note available	SIPRI analysis of pathways by which military AI may affect nuclear escalation risk through information, decision, and interaction dynamics. It motivates scenario-specific analysis and does not establish the net effect of any specific system or policy.
ext_no_free_lunch_inductive_bias_2024	The No Free Lunch Theorem, Kolmogorov Complexity, and the Role of Inductive Biases in Machine Learning	Goldblum et al. (2024), No Free Lunch and Inductive Bias; source	learning_theory_assumptions_and_inductive_bias	learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	ICML 2024 treatment connecting no-free-lunch limits, Kolmogorov complexity, and inductive bias. It supports explicit assumption accounting; it does not show that all learning problems are equally hard or identify the right bias for a deployment.
ext_neuromorphic_computing_scale_2025	Neuromorphic computing at scale	Davies et al. (2025), Neuromorphic computing at scale; source; DOI 10.1038/s41586-024-08253-8	neuromorphic_hardware_and_event_driven_computation	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); physical-compute-infrastructure-energy-and-environmental-constraints (Physical Compute Infrastructure, Energy, and Environmental Constraints)	source note available	Large-scale neuromorphic systems result demonstrating event-driven hardware capabilities under reported workloads and conditions. It does not establish superiority for general AI workloads or end-to-end system cost, programmability, reliability, and governance.
ext_photonic_neuromorphic_2024	Integrated photonic neuromorphic computing: opportunities and challenges	Shastri et al. (2024), Integrated photonic neuromorphic computing; source; DOI 10.1038/s44287-024-00050-9	photonic_neuromorphic_compute	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); physical-compute-infrastructure-energy-and-environmental-constraints (Physical Compute Infrastructure, Energy, and Environmental Constraints)	source note available	Review of integrated photonic neuromorphic computing opportunities and challenges. It maps device and systems tradeoffs but does not establish deployment advantage, digital replacement, or favorable full-stack energy and cost.
ext_quantum_ml_shadows_2024	Shadows of quantum machine learning	Schuld and Killoran (2024), Shadows of quantum machine learning; source; DOI 10.1038/s41467-024-49877-8	quantum_machine_learning_claim_boundaries	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); physical-compute-infrastructure-energy-and-environmental-constraints (Physical Compute Infrastructure, Energy, and Environmental Constraints)	source note available	Peer-reviewed analysis of limitations and benchmarking traps in quantum machine-learning advantage claims. It supports advantage declarations with data-loading, classical-baseline, noise, scale, and end-to-end accounting, not a claim that quantum ML is useless.
ext_organoid_intelligence_2023	Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish	Smirnova et al. (2023), Organoid intelligence; source; DOI 10.3389/fsci.2023.1017235	biohybrid_computing_and_moral_status	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture)	source note available	Research agenda for organoid intelligence and biohybrid computing. It motivates scientific, measurement, welfare, consent, and governance questions but does not demonstrate general intelligence, conscious experience, or practical compute superiority.
ext_nist_pqc_standards_2024	Announcing Approval of Three Federal Information Processing Standards for Post-Quantum Cryptography	NIST (2024), Post-Quantum Cryptography Standards; source	post_quantum_cryptography_and_crypto_agility	security-kernel-and-digital-scifs (Security Kernel and Digital SCIFs); model-weight-custody-and-hardware-roots-of-trust (Model-Weight Custody and Hardware Roots of Trust)	source note available	Official NIST announcement for FIPS 203, 204, and 205. It establishes approved algorithm standards and migration urgency, not implementation security, protocol correctness, complete inventory, or successful system migration.
ext_oecd_ai_infrastructure_competition_2025	Competition in artificial intelligence infrastructure	OECD (2025), Competition in AI Infrastructure; source	ai_infrastructure_concentration_and_competition	institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy); ai-deployment-transition-distribution-and-human-agency (AI Deployment, Transition, Distribution, and Human Agency); physical-compute-infrastructure-energy-and-environmental-constraints (Physical Compute Infrastructure, Energy, and Environmental Constraints)	source note available	OECD analysis of concentration, barriers to entry, vertical integration, and competition across AI infrastructure. It motivates bottleneck and exit analysis but does not adjudicate a specific market, legal violation, or optimal remedy.
ext_eu_ai_civil_liability_2025	Artificial intelligence and civil liability	European Parliamentary Research Service (2025), Artificial intelligence and civil liability; source; DOI 10.2861/0075079	ai_liability_remedy_and_compensation	institutions-international-coordination-and-public-legitimacy (Institutions, International Coordination, and Public Legitimacy); human-ai-organizations-delegation-and-accountability (Human-AI Organizations, Delegation, and Accountability)	source note available	European Parliament research service study of AI and civil-liability questions. It supports explicit causation, evidence-access, insurance, compensation, and remedy analysis but is not legal advice or a globally settled liability rule.
ext_cultural_alignment_llms_2024	Investigating Cultural Alignment of Large Language Models	Tao et al. (2024), Investigating Cultural Alignment of Large Language Models; source; DOI 10.48550/arXiv.2402.13231	cultural_alignment_and_value_representation	human-intent-as-a-formal-input (Human Intent as a Formal Input); human-ai-communication-persuasion-and-epistemic-security (Human-AI Communication, Persuasion, and Epistemic Security); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	Empirical study of cultural alignment patterns in selected language models and measurements. It supports explicit population, language, and instrument scope; it does not establish stable national values or a universal measure of cultural alignment.
ext_multilingual_evaluation_state_2026	The State and Fate of Multilingual Contextual Evaluation in the NLP World	Microsoft Research (2026), Multilingual Contextual Evaluation; source	multilingual_contextual_evaluation	human-ai-communication-persuasion-and-epistemic-security (Human-AI Communication, Persuasion, and Epistemic Security); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	Research survey and analysis of multilingual contextual evaluation. It motivates language-by-task coverage and measurement reporting; it does not establish equivalent capability or safety across languages, dialects, or sociocultural settings.
ext_kimi_k3_2026	Kimi K3: Open Frontier Intelligence	Kimi Team (2026), Kimi K3: Open Frontier Intelligence; arXiv:2607.24653; DOI 10.48550/arXiv.2607.24653	hybrid_attention_sparse_routing_and_training_systems	routing-heads-and-specialist-cores (Routing Heads and Specialist Cores); replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling)	source note available	Primary technical report and official architecture summary for KDA/Gated-MLA hybrid attention, Attention Residuals, Stable LatentMoE, Quantile Balancing, SiTU-GLU, and Per-Head Muon. The approximately 2.5x scaling-efficiency result is provider-reported for the integrated 2.8T system and does not identify a transferable component effect.
ext_elizaos_agent_runtime_2026	elizaOS Agent Runtime and Scenario Runner	elizaOS contributors (2026), elizaOS Agent Runtime; source	modular_agent_runtime_and_evidence_qualification	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); benchmark-ratchets-and-anti-goodhart-evidence (Benchmark Ratchets and Anti-Goodhart Evidence)	source note available	Pinned official implementation comparator for modular actions, providers, evaluators, services, runtime lifecycle, scenario execution, and the explicit distinction between in-process diagnostics and externally qualified provider evidence. No elizaOS execution, test reproduction, security assessment, performance result, or support transition is imported.
ext_hermes_agent_2026	Hermes Agent: Learning, Memory, Tools, and Security Architecture	Nous Research (2026), Hermes Agent; source	procedural_memory_and_agent_runtime	durable-semantic-memory-and-knowledge-lattices (Durable Semantic Memory and Knowledge Lattices); ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); procedural-memory-and-cognitive-loop-closure (Procedural Memory and Cognitive Loop Closure)	source note available	Pinned official implementation comparator for progressive-disclosure skills, agent-managed procedural memory, staged skill-write approval, bounded prompt memory, session search, tool backends, command approval, and isolation. No learning, memory, security, utility, or performance result was reproduced.
ext_openclaw_agent_runtime_2026	OpenClaw Gateway, Agent Runtime, ACP, and Self-Learning Architecture	OpenClaw contributors (2026), OpenClaw Agent Runtime; source	gateway_session_harness_and_procedural_learning_runtime	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses); artifact-graphs-audit-logs-and-replay (Artifact Graphs, Audit Logs, and Replay); runtime-adapters-tool-permissions-and-human-approval (Runtime Adapters, Tool Permissions, and Human Approval); inter-stack-protocols-identity-and-economic-exchange (Inter-Stack Protocols, Identity, and Economic Exchange); procedural-memory-and-cognitive-loop-closure (Procedural Memory and Cognitive Loop Closure)	source note available	Pinned official implementation comparator for gateway and device identity, serialized session runs, bounded audit projection, ACP external-harness identity and authorization boundaries, separated sandbox/tool/elevation controls, and evidence-reviewed hash-bound skill proposals. Distinct from the Claw-SWE-Bench benchmark source; no implementation result was reproduced.
ext_github_copilot_work_surfaces_2026	GitHub Copilot Product and Work-Surface Documentation	GitHub Documentation (2026), GitHub Copilot; source	ai_work_surface_evolution	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses)	source note available	Official current-product comparator spanning inline suggestions, chat, command line, contextual spaces, pull-request work, and agent-driven development. No workflow, productivity, safety, or comparative result was reproduced.
ext_augment_code_agent_2026	Augment Code Agent Documentation	Augment Documentation (2026), Using Agent; source	ide_agent_modes_and_review	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses)	source note available	Official comparator for the transition among chat, read-only inquiry, approval-paused agent work, and more independent agent execution with diffs and checkpoints. No product execution or control claim was reproduced.
ext_openai_codex_work_surfaces_2026	OpenAI Codex CLI, IDE, Cloud, and Agent Documentation	OpenAI (2026), Codex Documentation and CLI; source	coding_agent_harness_and_distributed_work_surfaces	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses)	source note available	Official documentation and pinned CLI comparator for repository inspection, editing, tool execution, permissions, local and cloud work, automation, and extensibility across multiple surfaces. No benchmark, correctness, safety, or productivity result was imported.
ext_anthropic_claude_code_2026	Claude Code Agentic Harness Documentation	Anthropic (2026), Claude Code Documentation; source	agentic_harness_and_execution_loop	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses)	source note available	Official comparator that explicitly separates model from harness and describes gather-context, act, and verify loops across terminal, IDE, desktop, web, remote, and automation surfaces. No implementation result was reproduced.
ext_opencode_agent_2026	OpenCode Open-Source Coding Agent	OpenCode contributors (2026), OpenCode; source	open_source_coding_agent_harness	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses)	source note available	Pinned official comparator for a provider-flexible coding agent with terminal, desktop, and IDE surfaces, project instructions, plan/build modes, tool execution, recovery, and opt-in sharing. No runtime or provider-parity claim was reproduced.
ext_oh_my_pi_agent_2026	Oh My Pi Terminal Coding Agent and Tool Harness	Oh My Pi contributors (2026), Oh My Pi; source	integrated_terminal_agent_harness	ai-work-surfaces-agent-harnesses-and-organizational-absorption (From Chat to Organizations: AI Work Surfaces and Agent Harnesses)	source note available	Pinned official comparator for an integrated terminal harness with hash-anchored edits, LSP, shell, browser, subagents, memory, provider switching, review, and collaboration. Reported performance or security claims were not reproduced.
ext_eggroll_hyperscale_es_2026	Evolution Strategies at the Hyperscale	Rybkin et al. (2026), Evolution Strategies at the Hyperscale; arXiv:2511.16652; DOI 10.48550/arXiv.2511.16652	zeroth_order_population_learning	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science); resource-economics-and-token-budgets (Resource Economics and Token Budgets); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	Primary EGGROLL project and paper source for low-rank, batched evolution strategies, counter-based perturbation reconstruction, nondifferentiable and discrete objectives, recurrent/int8 training, and outcome-reward fine-tuning. Throughput, quality, and theory claims are source-scoped; total population evaluations and GPU-hours remain required denominators.
ext_openai_es_2017	Evolution Strategies as a Scalable Alternative to Reinforcement Learning	Salimans et al. (2017), Evolution Strategies as a Scalable Alternative to Reinforcement Learning; arXiv:1703.03864; DOI 10.48550/arXiv.1703.03864	evolution_strategies_and_black_box_policy_search	governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); learning-compute-topology-and-adaptive-process-architecture (Learning–Compute Topology and Adaptive Process Architecture); resource-economics-and-token-budgets (Resource Economics and Token Budgets); policy-optimization-and-learning-from-feedback (Policy Optimization and Learning from Feedback)	source note available	Foundational modern large-population ES comparator using parameter perturbations, scalar fitness, seed reconstruction, and distributed evaluation. Source-reported MuJoCo/Atari results and worker scaling do not establish universal sample or total-compute efficiency.
ext_mezo_2023	Fine-Tuning Language Models with Just Forward Passes	Malladi et al. (2023), Fine-Tuning Language Models with Just Forward Passes; arXiv:2305.17333; DOI 10.48550/arXiv.2305.17333	memory_efficient_zeroth_order_fine_tuning	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); resource-economics-and-token-budgets (Resource Economics and Token Budgets)	source note available	Primary MeZO source for inference-footprint zeroth-order language-model fine-tuning and nondifferentiable objectives. Reported memory and GPU-hour savings are configuration-bound and do not erase objective-query count or estimator variance.
ext_forward_forward_2022	The Forward-Forward Algorithm: Some Preliminary Investigations	Hinton (2022), The Forward-Forward Algorithm; arXiv:2212.13345; DOI 10.48550/arXiv.2212.13345	local_forward_only_credit_assignment	replaceable-cognitive-substrates-beyond-transformer-monoculture (Replaceable Cognitive Substrates: Beyond Transformer Monoculture); governed-model-training-distributed-optimization-and-scaling (Governed Model Training, Distributed Optimization, and Scaling); learning-theory-generalization-and-scaling-science (Learning Theory, Generalization, and Scaling Science)	source note available	Primary preliminary source for positive/negative forward passes and local layer objectives as an alternative to reverse-mode backpropagation. The evidence is small-scale and does not establish foundation-model parity or biological plausibility.
ext_pbt_2017	Population Based Training of Neural Networks	Jaderberg et al. (2017), Population Based Training of Neural Networks; arXiv:1711.09846; DOI 10.48550/arXiv.1711.09846	population_based_adaptive_training	learning-compute-topology-and-adaptive-process-architecture (Learning–Compute Topology and Adaptive Process Architecture)	source note available	Primary Population Based Training source for asynchronous joint optimization of a population’s model parameters and hyperparameter schedules through evaluation, exploitation, and exploration. Source-reported reinforcement-learning, translation, and GAN results remain task- and implementation-bound and do not validate LCT, universal topology adaptation, safety, or superior total lifecycle cost.

H.6 External Literature Queue

Third-party references should be added only when bibliographic metadata is recorded and the source is actually used.

Area	Expected role	Status
AI alignment and corrigibility	External comparison for the alignment and constitution layer.	initial source records and source notes added; no local reproduction or support-state promotion
AI governance, evals, and deployment policy	External comparison for authority ceilings, readiness gates, and release governance.	initial source records and source notes added; no local reproduction, compliance claim, or support-state promotion
Planning, task decomposition, and agent control	External comparison for PlanForge-style planning/control.	initial source records and source notes added; no local reproduction or support-state promotion
Retrieval, memory, and context engineering	External comparison for VCM and context-packet discipline.	initial source records and source notes added; no local reproduction or support-state promotion
Formal methods, verification, and proof assistants	External comparison for claim ledgers, Lean proofs, and protocol invariants.	initial source records and source notes added; no imported formal artifact or support-state promotion
Modular systems, routing, and mixture-of-experts	External comparison for routing and specialist promotion.	initial source records and source notes added; no local reproduction or support-state promotion
Compression, representation learning, and program synthesis	External comparison for compact generative systems and residual accounting.	initial source records and source notes added; no compression experiment or support-state promotion
Fast generation, decoding substrates, and serving acceleration	External comparison for MTP, speculative decoding, internal draft heads, diffusion LLMs, early exit, state-space alternatives, KV-cache memory, and useful-solution-per-second metrics.	initial source records and source notes added; no local reproduction or support-state promotion
Policy optimization and learning from feedback	External comparison for PPO/RLHF, GRPO/RLVR, DPO-style preference optimization, verifier rewards, reward hacking, reasoning-budget RL, and control-policy RL for planners, routers, VCM, execution, and generation modes.	initial source records and source notes added; no local reproduction or support-state promotion
Benchmarks, evaluation science, and anti-Goodhart methods	External comparison for evidence ratchets and regression preservation.	initial source records and source notes added; no local benchmark run or support-state promotion

H.7 External Citation Policy

• Keep outside literature separate from Corben’s own papers, Corben-supplied materials, recovered project records, and local project records.
• Do not cite an external source as supporting a claim until the source text has been read and a source note or equivalent review artifact exists.
• Do not report reproduced external results unless the reproduction artifact, command, environment, and result record exist.
• Keep third-party documentation, papers, and benchmarks at their recorded support boundary.