v0.2.2

fix: support 8-char short ID prefix matching for memory tools (#14)

memory_scope_promote, memory_delete, memory_feedback_wrong,
memory_feedback_useful all failed with 'not found' when given an
8-character short ID (first 8 chars of a UUID), even though
memory_search could find the same memory by text/vector.

Root cause: hasMemory(), deleteById(), updateMemoryScope(), and
updateMemoryUsage() all used strict === comparison against the full
36-char UUID stored in LanceDB.

Fix:
- Add matchesId(candidateId, query) private helper on MemoryStore:
  - query >= 36 chars → exact === match (unchanged behavior)
  - query < 36 chars  → startsWith prefix match
- Apply matchesId to all four methods
- Change id schema validation from min(6) to min(8) on all memory
  tools to align with the short-ID convention
- Fix latent data-loss bug: DELETE in updateMemoryScope and
  updateMemoryUsage now uses match.id (resolved full UUID) instead
  of the id argument, which could be a prefix

Bump version to 0.2.2.

v0.2.1

fix: add memories table schema migration for lastRecalled/recallCount…

v0.2.0

fix: resolve regression test flakiness from sidecar config bleed and …

v0.1.6

Merge pull request #9 from tryweb/fix/effectiveness-schema-upgrade

chore: release 0.1.6 schema-upgrade patch

v0.1.5

What's Changed

Added

• RecallSource type to distinguish "system-transform" (auto recall) from "manual-search" (user-initiated).
• memory_search tool now emits a structured recall event with source: "manual-search" and injected: false.
• EffectivenessSummary.recall now includes auto and manual sub-structures with independent hitRate/injectionRate, plus manualRescueRatio — the first instrumented proxy metric from the low-feedback evaluation framework.

Changed

• recall.hitRate and recall.injectionRate top-level fields are retained as blended totals for backward compatibility; consumers that need precise breakdowns should read recall.auto.* and recall.manual.*.
• docs/operations.md proxy metrics table marks manual memory rescue rate as instrumented via recall.manual.requested and recall.manualRescueRatio.

OpenSpec

• Archived add-manual-search-recall-split and add-low-feedback-memory-evaluation changes.
• Main specs synced: memory-effectiveness-evaluation, memory-management-commands, memory-validation-harness, and new low-feedback-memory-evaluation.

Full Changelog: v0.1.4...v0.1.5

v0.1.4

Highlights

• add phase-1 retrieval ranking with reciprocal rank fusion (RRF), recency boost, and importance weighting
• add retrieval configuration controls: rrfK, recencyBoost, recencyHalfLifeHours, and importanceWeight
• add foundation/regression coverage for ranking behavior and config defaults/overrides
• archive OpenSpec change for ranking phase-1 and sync updated specs into main

Verification

• docker compose build --no-cache && docker compose up -d
• docker compose exec app npm run verify

v0.1.3

v0.1.3

v0.1.2

release 0.1.2