Ticket changelog (ticket-004)
[0.1.0] - 2026-07-31
- Created the separately scoped language-independent matching experiment.
- Recorded user continuation as approval.
- Entered
TOOLS with precision, provenance and offline-CI guardrails.
[0.2.0] - 2026-07-31
- Added a multilingual synthetic benchmark with six positive and six nearby
negative pairs across Polish, German, Spanish and French.
- Evaluated pinned MiniLM and E5 models locally.
- Rejected a global cosine threshold because positive and negative score ranges
overlap.
[0.3.0] - 2026-07-31
- Ranked 66 actionable targetless platform declarations against 133 module
aggregates.
- Rejected two new forward-threshold candidates during manual review.
- Confirmed reciprocal top-1 removes the false positives but adds no coverage;
no production matcher was retained.
- Added a separately reported cross-language gold cohort with six known
positives and six gated hard negatives; transitioned to
VERIFY.
[0.4.0] - 2026-07-31
- Passed 244 tests (243 pass, zero fail, one allowed local Java skip), gold
v1/v2, all five SDK examples, dependency audit, CLI/MCP/A2A and Docker smoke.
- Updated
READINESS.md, TEST_REPORT.md, VALIDATION.md and TODO.md.
- Closed the rejected matcher experiment in
DONE without a production
semantic rule.
[0.4.1] - 2026-07-31
- Corrected repository layout after review: moved both executable embedding
experiment reproducers from the ticket evidence directory to
scripts/research/.
- Preserved benchmark inputs, captured outputs and decisions in the ticket.