Damage audit: harden AI-assisted governance/schema/prototype surfaces

[mdheller]

Purpose

Track the superconscious damage-audit and keep-but-harden remediation sequence.

This issue is the work tracker for the audit plan added in PR #38.

Core risk

The risk is not primarily destructive runtime code. The risk is semantic overpromotion:

• schema validation appearing to verify invariants;
• fixture polarity appearing to prove runtime behavior;
• doctrine references importing authority without commit pins;
• theorem-facing language appearing proof-complete without proof-review evidence;
• exploratory prototypes appearing canonical;
• CI success being read as semantic, mathematical, or runtime verification.

Work packages

• H0 — Containment ledger: confirm baseline SHA, open/draft PRs, ahead-of-main branches, and branch dispositions.
• H1 — External doctrine source pinning: commit-pin ProCybernetica / cross-repo doctrine references.
• H2 — Schema-only guarantee banners: add explicit Layer 1 declaration-only disclaimers to binding schemas/docs.
• H3 — Claim-status ledger: record high-risk claim surfaces with normalized status and evidence requirements.
• H4 — Theorem-language review gate: demote or proof-review all theorem/theorem-pattern language.
• H5 — CI polarity and aggregate target audit: verify make certificate-ci and make -f GNUmakefile certificate-ci coverage and polarity.
• H6 — Branch/content reconciliation: classify every ahead-of-main branch as captured, superseded, quarantine, merge-candidate, or delete-after-capture.
• H7 — PR Add Neuronpedia v0.1 schema tranche with fixture naming #37 readiness gate: keep Neuronpedia PR in draft until polarity and Layer 1/Layer 2 boundaries are recorded.

Non-destruction rule

Do not delete or close useful work merely because it was weakly reviewed. Allowed outcomes:

merge after hardening
replay after demotion
capture as exploratory
record supersession
record intentional non-capture
then delete stale branch only after capture/supersession is durable

Immediate state

• Baseline branch: audit/2026-05-19-claude-damage-baseline
• Audit plan PR: Add superconscious damage-audit hardening plan #38
• PR Add Neuronpedia v0.1 schema tranche with fixture naming #37: converted back to draft pending readiness gate

Acceptance criteria

This issue can close only after every high-risk surface has an explicit disposition and all retained surfaces carry correct claim-status boundaries.

[mdheller]

Progress update:

• H0 containment baseline landed via PR Add superconscious damage-audit hardening plan #38 and merged at acd390f9fc52d355c1138e243afd0e590914d489.
• PR Add Neuronpedia v0.1 schema tranche with fixture naming #37 remains draft and blocked pending polarity / Layer 1 vs Layer 2 readiness checks.
• H1/H2 remediation opened as PR Pin Tier 2 doctrine source and harden binding-schema boundaries #40: Pin Tier 2 doctrine source and harden binding-schema boundaries.
• PR Pin Tier 2 doctrine source and harden binding-schema boundaries #40 pins ProCybernetica Tier 2 doctrine to immutable merge commits for PR Capture neuro-symbolic cognition boundary #56 and PR Extend neuro-symbolic visual failure modes #57 and clarifies local Tier 2 binding schemas as Layer 1 declaration-shape artifacts only.

Open next gates:

1. Wait for PR Pin Tier 2 doctrine source and harden binding-schema boundaries #40 CI.
2. Merge PR Pin Tier 2 doctrine source and harden binding-schema boundaries #40 if green.
3. Start H3 claim-status ledger covering Tier 2 bindings, lawful-learning theorem/theorem-pattern docs, interpretability harness, Neuronpedia PR Add Neuronpedia v0.1 schema tranche with fixture naming #37, SPO/Baez prototype branch, and ahead-of-main branch families.
4. Start H4 theorem-language review after ledger lands.