v1.1.1

Patch release. The CLI and MCP entrypoints now default to the bm25 retriever, matching pre-registered decision D3 of study c36 (Finding More, Fusing Less, doi:10.5281/zenodo.21969901), which retired the fused default. Previous releases shipped surfaces that still defaulted to fused while the library documentation claimed bm25-primary. A regression test now asserts no surface reintroduces the fused default. No other behavior changes.

v1.1.0

First release of the open-development era.

Changes since the v1.0.x public cut:

• The lineage classifier recognizes any Jira-style ticket key instead of one hardcoded project prefix, with a stoplist so utf-8 / sha-256 don't misclassify, and regression tests in both directions.
• JOSS paper completed: State of the field, Software design, Research impact statement, AI usage disclosure; references verified against Crossref/arXiv (PVLDB DOI added; the preregistration position paper re-attributed to its actual author).
• CI on Node 20/22 (npm run verify, 47 checks) and the JOSS draft-pdf action on every push.
• .zenodo.json stages the software's own archive.

Evidence behind the defaults (unchanged): BM25-primary per the pre-registered bake-off 10.5281/zenodo.21969901; parent study 10.5281/zenodo.21960138. The original fused default remains retired by its frozen prediction.

Development happens here in the open from 2026-08-19 onward: see docs/DECISIONS-2026-08-19-open-development.md.