[daveh-beep]

What observed experience does this improve?

The sharing-readiness review found stale claims that merged receipt work was still proposed, template ownership ambiguity, unbounded corpus ingestion, and a receipt result that callers could mistake for verified evidence.

Plan and executed changes

• Fix stale site claims and link current traceability/validation documentation.
• Add an adoption checklist and explicit status, limits, and next-experiment plan.
• Retain the existing CODEOWNERS owner while removing unused scaffold comments; do not appoint new maintainers.
• Bound reference ingestion: 1,000 files, 1 MiB/file, 16 MiB total, 16 directory levels, 10,000 filesystem entries. Bounded reads use the same bytes for parsing and hashes; errors return no partial JSON or output file.
• Add a successful receipt CLI stderr notice: structure/arithmetic valid, evidence and approval unverified. Preserve stdout JSON shape and exit semantics, including blocked receipts exiting 0.
• Add regression tests for limits, exact-boundary controls, bounded actual reads, hash consistency, and unverified high-score/approval claims.
• Document an optional Jev experiment for fact preservation and retrieval relevance. No SDK, paid calls, hosted service, telemetry, or automated approval added.

Verification

Executed locally with Python 3.9.6 and Node v24.11.0:

• bash checks/adversarial-contract.sh — passed
• python3 -B -m unittest discover -s tests -v — 43 tests passed
• python3 -B tools/corpus.py check --root . / examples/meridian / examples/harbor — counts 0 / 4 / 5, valid
• node --check website/script.js and node website/test.cjs — passed
• local documentation link check and git diff --check — passed
• Independent read-only code review — no blockers

Tool negatives and unchanged controls are synthetic and executed. Agent setup/drafting instructions were not changed; no held-out model behavior or cross-harness parity claim is made. CODEOWNERS comments do not enforce the two-maintainer governance rule.

Remaining work

Cross-harness setup validation (#3), a consenting-team pilot (#6), semantic ingest quality (#5), and keyboard/screen-reader accessibility checks are not executed here. Snapshot projection remains manual, not bundled enforcement. Fixed resource caps bound ingestion, not all resources or concurrent filesystem modification. Receipt evidence and approval remain self-reported.

Safety

Only existing public material and synthetic fixtures are included. No raw agent sessions, private corpora, customer data, credentials, or private links. Commit includes DCO sign-off. Draft pending maintainer review; no merge or deployment requested.