v0.0.3

Butler v0.0.3

Change Log

• Replaced misleading memory recall labeling with LanceDB-backed vector recall
  evidence. Butler now separates semantic_similarity, lexical_match,
  contextual_match, graph, task, and explicit-memory evidence in recall
  diagnostics.
• Added production hot-cache vector backfill through butler cognition memory maintain --hot-cache-backfill-only --json, so existing hot-cache memory can
  be indexed through the same CLI path used by the real runtime.
• Reworked recall ranking away from anonymous fixed weighted sums and added a
  real-data A/B gate against the legacy keyword-weighted baseline. The latest
  gate showed 4 legacy-baseline lifts, 0 legacy regressions, 0 top decoys, 0
  positive misses, and 0 negative false positives on the Butler data snapshot.
• Hardened work-state persistence with locked atomic file-state writes for
  planned task state, with concurrent TaskStore and PlannedTaskStore E2E
  contention evidence.
• Improved the agent loop so repeated identical failed tool calls can stop
  early without discarding a successful alternate tool result from the same
  safe parallel batch.
• Fixed packaged Butler App startup so the desktop client attaches to an
  already healthy local app gateway, avoids Electron-owned gateway pid-file
  contention, and resolves the managed Bun runtime under Finder-like macOS
  launch environments.
• Added dedicated Butler App release artifacts. Version-tag releases now build
  a signed butler-app-0.0.3-darwin-arm64.zip, a
  butler-app-0.0.3-linux-x64.tar.gz, SHA256 files, app-release-manifest.json,
  and app-update-manifest.json alongside the service artifact.
• Strengthened real-data E2E reporting. The evidence gate now records prompt
  input, internal recall/tool-loop diagnostics, target-vector-backed evidence,
  file-state contention evidence, safe-tool parallel execution, unsafe-tool
  serialization, repeated-failure early stop, and alternate-path recovery.

Validation

• bun run e2e:memory-quality-ab passed on run
  2026-06-02T06-24-01-303Z-05c78913.
• bun run e2e:evidence-gate passed 17/17 on run
  2026-06-02T05-54-51-706Z-fa411ca4.
• bun run release:app:package --out dist/release/app --artifact-base-url https://github.com/Hexpy-Games/butler/releases/download/v0.0.3 generated
  signed app release artifacts and passed macOS codesign --verify --deep --strict plus Finder-like launch smoke.
• bun run release:service:package --out dist/release/service --artifact-base-url https://github.com/Hexpy-Games/butler/releases/download/v0.0.3 regenerated
  the service tarball, SHA256 file, service release manifest, and update
  manifest.
• bun run check passed after release preparation.
• project-ledger check --project $BUTLER_HOME --silent passed with
  no stale views or next actions.