Add superconscious damage-audit hardening plan

[mdheller]

Summary

Adds a repo-local audit and hardening plan for the recent AI-assisted governance/schema/theorem/prototype workstream.

The goal is keep-but-harden, not deletion. The document classifies the risk as semantic overpromotion: schema/fixture/CI/prototype surfaces being mistaken for invariant verification, proof, runtime enforcement, or external authority.

Adds

• docs/audits/2026-05-19-superconscious-hardening-plan.md

Captures

• current main audit baseline SHA: e8e9da945a5178dd7299136d04a982f38ac7c650
• baseline branch: audit/2026-05-19-claude-damage-baseline
• PR Add Neuronpedia v0.1 schema tranche with fixture naming #37 containment state: converted to draft pending polarity / Layer 1 vs Layer 2 review
• hardening taxonomy for claim-status normalization
• H0-H7 work packages
• branch/content reconciliation policy
• non-destruction rule for useful-but-weakly-reviewed artifacts

Boundary

This PR is audit documentation only.

It does not change schemas, checkers, runtime behavior, lawful-learning claims, Neuronpedia schemas, or certificate CI targets.

Next after merge

1. Source-pin and schema-only banner PR for Tier 2 bindings.
2. Claim-status ledger PR.
3. Lawful-learning theorem-language review/demotion gate PR.
4. PR Add Neuronpedia v0.1 schema tranche with fixture naming #37 hardening update before ready-for-review.
5. Branch reconciliation PRs, one branch family at a time.