[iHow1]

Summary

Core-generated ingestion metadata could influence lexical retrieval, while empty lock files left by interrupted initialization could block later writers indefinitely. This integrated 0.1.1 candidate combines the bounded FTS projection/migration fix with initialized lock publication and guarded stale-lock recovery. Original Markdown, business dates, governance eligibility, and activation trust boundaries remain intact.

The version chain now includes Core, bundled Hermes plugins, bilingual release notes, and upgrade assertions. The release workflow retains one artifact before the expensive gate and verifies that exact tarball through an offline fresh install, byte checks, CLI/MCP round-trips, isolation, and restart persistence before publication.

Validation

• Frozen candidate: b7dfd68f69d5a03969b07791df8b3034f4930400.
• Local artifact SHA-256: 03b2b679854b4e03557081f039bef091dc785ef2ed522fd07756a2ea1510b952.
• Version contracts, typecheck, build, secret scan, and release-artifact guard tests pass.
• Local macOS full gate passed: 1,184 tests passed, zero failed, and two optional Hermes B3 host-contract checks skipped because the installed checkout lacks agent/transcript_revision.py. Governed proof, secret scan, and the retained tarball verifier passed; source and artifact hashes remained unchanged. The installed-host B3 coverage remains explicitly unavailable; the installed checkout was not modified.
• The same tarball passes 35 isolated AML JavaScript tests, 55 Python unit tests, and 10 tests against the published DSH alpha.5 adapter, plus synthetic cross-process write/recall/forget/restore. All 112 installed Core files match the artifact in both fixtures. AML requires an exact version-guard update; DSH requires an explicit dependency override or a separately released adapter.
• Cross-platform CI run 34137586448 passed all five jobs. Linux Node 22 and 24 each passed 1,181 tests with zero failures and five optional Hermes host skips; Windows Node 22 and 24 each passed 18 targeted smoke tests plus version/connect/doctor checks. Gitleaks working-tree and history scans passed. The CI synthetic merge commit has the exact frozen candidate tree. Three of the five Linux host skips are exercised by the local macOS gate; the two unavailable B3 contracts remain disclosed above.
• No paid model calls or production runtime changes are part of this candidate validation.

Release boundaries

Merged by rebase after a maintainer source review. The final main tree matches the frozen RC exactly. Version 0.1.1 is published, and registry bytes match the accepted artifact. Local and cross-platform CI gates passed. Recovery guards interrupted by a crash fail explicitly for operator recovery; old clients do not participate in the new recovery guard, so writer upgrades must be coordinated. Atomic publication requires same-directory hard-link support; network filesystems are not certified.

All commits carry DCO sign-offs. Tests and examples use synthetic data; detailed local benchmark material and local acceptance outputs are excluded from the repository.

Checklist

• npm run build passes locally
• Full repository runner (npm test) passes locally; two optional installed-host skips are disclosed above
• node scripts/proof.mjs passes
• No customer data, private memory content, real user paths or credentials added
• Synthetic data only in tests and examples
• No private product code or internal strategy added
• DCO sign-off present on all three commits
• License and trademark boundary unchanged
• CHANGELOG.md updated

Publication evidence

• Final main/tag commit: 5794ea7c1bdd554054d89d5cca40c9983de8fda6; tree unchanged from this RC.
• Release workflow and post-merge CI passed.
• v0.1.1 release includes the verified package and evidence. npm latest and next both resolve 0.1.1.
• The independently hosted website was synchronized and read back. Persistent runtime and active host verification remain separate release states.