Selfhost Readiness Scorecard v0.1
Purpose
Evaluate whether GenesisCode’s strict selfhost command-routing profile, exact per-decision H2 semantic authority, per-decision H3 bootstrap closure, parity isolation, bootstrap-mode configuration, selected runtime quality checks, and active P0/P1 defect closure satisfy this scorecard, and define the deterministic machine-readable report produced on explicit request. This report does not establish H2 semantic authority, an H3 cross-host bootstrap fixpoint, L5 release evidence, or v1 release readiness.
Runner
- Read-only check:
scripts/check_selfhost_readiness_scorecard.sh - Explicit producer:
scripts/update_selfhost_readiness_scorecard_report.sh - Optional primary report:
.genesis/perf/selfhost_readiness_report.json - Optional history:
.genesis/perf/selfhost_readiness_history.jsonl
Strict profiles invoke dashboard freshness and this readiness scorecard as separate checks. Neither check regenerates the other’s local E0 report. The readiness check renders into private temporary outputs while evaluating p95 against the configured existing history input. Only the explicit producer appends to the retained history and replaces the retained report.
Default history floor:
GENESIS_SELFHOST_READINESS_P95_MIN_SAMPLES=5
Report Contract
kind = "genesis/selfhost-readiness-v0.1"okbooleanscore_percentelapsed_ms,budget_ms,history_samples,history_p95_ms,history_p95_enforced,history_p95_okp95_min_samplesfail_reasons(string list)- canonical closure token:
unresolved-upgrade-plan-ids - backward-compatible legacy token accepted by cutover gates:
open-upgrade-plan-ids
- canonical closure token:
unresolved_upgrade_plan_ids(string list),closure_okdimensionsobject with scored dimensions:runtime_routing_coveragesemantic_authority_closurebootstrap_fixpoint_closureparity_only_surface_isolationbootstrap_mode_strictnessdeprecated_bootstrap_reference_countcritical_gate_truthruntime_quality_truth
semantic_authority_closure and bootstrap_fixpoint_closure are derived from the validated docs/spec/SEMANTIC_OWNERSHIP_LEDGER_v0.1.json, never from dashboard percentages or broad capability rows. The former passes only when every applicable semantic decision is at least H2; the latter passes only when every applicable decision is at least H3. Declared residual stage0 decisions receive no score credit and cannot be promoted by this report. Each dimension emits the exact ledger identity and blocking decision rows.
runtime_quality_truth is a fail-closed aggregate over machine reports:
.genesis/perf/runtime_microbench_runtime_report.json.genesis/perf/hot_path_runtime_report.json.genesis/perf/task_concurrency_stress_report.json.genesis/perf/host_api_evolution_contract_report.json
critical_gate_truth is a fail-closed aggregate over machine reports:
.genesis/perf/agent_capability_gauntlet_release_confidence_report.json.genesis/perf/agent_generative_workloads_report.json.genesis/perf/agent_workflow_runtime_parity_report.json.genesis/perf/production_cli_help_surface_report.json.genesis/perf/gpu_gfx_headroom_conformance_report.json.genesis/perf/domain_starter_registry_bootstrap_report.json.genesis/perf/gcpm_target_runtime_evidence_report.json
Produce domain-starter registry evidence explicitly with scripts/update_domain_starter_registry_bootstrap_report.sh; the read-only scorecard never refreshes this prerequisite.
Target runtime evidence report contract:
- producer:
scripts/update_gcpm_target_runtime_pipelines_report.sh kind = "genesis/gcpm-target-runtime-evidence-v0.1"- per-target runtime evidence payload includes:
- runtime mode (
synthetic-adapterornon-synthetic) - runtime class (
emulator|device|container|host-runtime|synthetic-adapter) - typed qualification status (
qualified|unsupported-product|setup-required|infrastructure-failure|execution-failure|synthetic-only) - command, SDK/image/device identity, artifact identity, lifecycle, and replay-log hashes
- replay artifact directory + stdout/stderr hashes
- runtime mode (
- strict policy:
GENESIS_GCPM_TARGET_RUNTIME_REQUIRE_NON_SYNTHETIC=1requires a typed authentic lifecycle or fails closed at an explicit readiness blockerGENESIS_GCPM_TARGET_RUNTIME_EXPECT_OUTCOME=unsupported-productaccepts only the current product-matrix blocker and never changesrelease_qualified=false- default strictness follows CI context (
CI=true=> strict)
GPU/GFX headroom conformance must include lane backend metadata consumed by readiness:
require_device_lane_mode,require_device_lane_active,device_runtime_availablelanes.normal.backend_policy = "require-device"andlanes.normal.expected_backend = "device-runtime"whenever device runtime is availablelanes.low-headroom.fallback_policy = "allow-fallback-under-headroom"with observed backend evidence
The scorecard is read-only with respect to prerequisite reports. Produce the gauntlet input separately in strict release-confidence mode before running the scorecard:
GENESIS_AGENT_GAUNTLET_PROFILE=release-fullGENESIS_AGENT_GAUNTLET_REQUIRE_GPU_DEVICE_BACKEND=1require_gpu_device_backend=trueconfidence_lane="release-confidence-device"
Each dimension records at least:
okscoremax_score- dimension-specific evidence fields
Closure Semantics
Readiness is ok=true only when all are true:
- All scored dimensions are
ok=true, including exact H2 semantic authority and H3 bootstrap closure. - No unresolved
upgrade_plan.mdchecklist IDs remain. - Runtime elapsed and history-p95 remain within configured budget.
Routing coverage, a .gc wrapper, a parity match, or a decomposition-table phase can never satisfy either H2 or H3 dimension. ok=true closes only this scorecard contract. Semantic selfhost authority is reported in docs/status/SELFHOST_AUTHORITY_v0.1.md; release-claim eligibility is reported in feature_matrix.md. Neither may be inferred from this mutable local report.
Runtime History Floor Inputs
Runtime/perf reports consumed by readiness are expected to be produced with their own seeded baseline-history + minimum-sample contract. For migrated report sets, use the explicit producer:
scripts/update_runtime_microbench_budgets_report.shscripts/update_hot_path_budgets_report.shscripts/update_perf_budgets_report.shscripts/update_task_concurrency_stress_report.shscripts/check_production_cli_help_surface.shscripts/update_gpu_compute_runtime_profile_report.shscripts/check_gfx_runtime_profile.sh
Default baseline seed files live under policies/perf/*_seed_history.jsonl.
Default mode is non-strict evaluation (always renders an ephemeral report). Set GENESIS_SELFHOST_READINESS_STRICT=1 to make non-ready status fail the gate.