v0.11.1 Release Recap
v0.11.1 is a non-breaking ranking hardening patch. It adds a representative corpus regression test across the main retrieval scenarios and widens the internal candidate pool before final scoring.
What Changed
| Area | New in v0.11.1 | Why it matters |
|---|---|---|
| Ranking candidate pool | Vector retrieval now brings at least 50 candidates into fusion and scoring | Prevents crowded near-match topics from hiding stronger results before the ranker can apply confidence, recency, and utility |
| Representative corpus test | TestRepresentativeCorpusRankingGolden covers poisoning resistance, temporal memory, procedural memory, and general RAG fixtures | Gives future ranking changes a broader regression tripwire than the small unit fixture |
| Release docs | Proposal and release notes now record the ranking hardening slice | Keeps GitHub Pages aligned with the live tagged release |
Non-Breaking Behavior
The public SDK, REST, gRPC, and GraphQL contracts are unchanged. Ranking may improve in crowded result sets because more candidates are rescored before the final top-k is returned.