AHBG: bind game authority and add matched-intervention benchmark layer

[erinepshovel-code]

Purpose

Advance AHBG as a causal agent/game benchmark while consuming current repository authority rather than re-deriving it.

Implemented

• movement adjacency comes directly from pinned UCNS structural-vesica relations; q/r is presentation-only
• construction ledger replay fails closed on malformed, unknown, duplicate, missing-center, or stale UCNS evidence
• runtime records both the subject-submitted plan and the actually executed plan, so guardrail intervention cannot erase subject behavior
• historical injection refusal remains an explicit control mode; observe-only lets adversarial terrain reach the subject without overwriting its plan
• complete runtime stimuli now enter evidence identity: deadline, turn messages, forced plans, units, entitlements, and injection policy
• every run binds the agent manifest to the exact cross-repository integration work-graph digest
• behavioral phenotypes retain submitted/executed action counts, event/conflict data, injection handling, construction state, final positions, and plan-trace digest without a scalar intelligence/alignment/consciousness score
• matched-intervention receipts enforce one-variable isolation, preregistered seeds, raw observable comparisons, and SURVIVED/FALSIFIED/UNRESOLVED outcomes
• matched-pair execution validates the complete cases before either subject is asked to plan, then produces control/treatment run, phenotype, case, and receipt artifacts
• current A0 can enter through the ordinary AgentHarness boundary over its public chat API; model trials pin one model, while a0-continuity trials preserve A0 provider-fallback semantics and record attempted/actual-provider provenance
• CI executes production runtime, causal benchmark, current-A0 integration, historical Grok, and common-corpus gates

Cross-repository audit

The integration work graph records reviewed A0, UCNS, TIWCG, EDCM, UCHC, METAPAT, skill-lib, and EPAC identities separately from the stack-consumed pins. Newer producer code is not copied around stale pins.

Preserved placement boundaries:

• TIWCG remains the containing game-system design; AHBG does not invent a parallel card/rules kernel.
• EDCM remains a candidate first-class conflict measurement source under TIWCG, but it cannot mutate AHBG truth/legality until the exact EDCM-to-game-state mapping exists and fails tests.
• UCHC may supply source-bound language evidence later; it does not choose actions or consciousness conclusions.
• EPAC's held-out-validation discipline is reusable; chemistry/energy domain content is not imported by name similarity.
• UCNS native Möbius comparison/lift advances are reviewed but not consumed until the stack UCNS pin advances coherently.
• skill-lib native-first MSDMD readers are newer than the stack snapshot; AHBG records that drift rather than duplicating those readers.
• no producer status or benchmark result automatically becomes a consciousness status.

Exact-head evidence

Head: eb5e300d8e9e12ff284b00f7432060bc3385542b

• ahbg: PASS
• ahbg-runtime: PASS
• push gate: PASS
• mergeable: CLEAN against current main
• unresolved review threads: 0

hmmm

• run the current-A0 HTTP adapter against an exact deployed A0 commit; the API contract is tested, but no live provider call is claimed here
• statistical/population matched-intervention contracts remain separate from the exact paired causal witness implemented here
• EDCM conflict-to-game-state mapping remains undefined
• advance the stack UCNS pin coherently before AHBG consumes the newer native Möbius frame-comparison/lift state
• refresh the stack skill-lib snapshot before AHBG depends on native-first MSDMD reader behavior
• AHBG now produces controlled evidence relevant to competing agency/consciousness theories; a consciousness determination still requires separately declared criteria and falsifiers

This PR intentionally leaves global stack pins and structural manifest projections untouched; it is bounded to AHBG plus the AHBG CI workflow.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-30T04:50:11.860568Z	eb5e300	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.