0.9.1 — Requirements and stories now check their own glossary
Released 2026-09-22.
Term-definedness checking (R-D and D-D) has moved from asking the AI to judge which terms are defined, to deciding that deterministically from your glossary — and a long-standing case-sensitivity bug in the term matcher is fixed along the way.
What's new
Deterministic definedness. The AI's role in R-D and D-D is now narrower: it extracts the significant terms from a requirement or story, and REQQA's own code decides which of those are actually defined in your glossary. Every undefined-term finding can now be checked independently against the glossary itself — it is no longer only the model's say-so that a term was missing.
Mission glossary checking is now scoped, not global. Analysing an application's mission statement for definitions used to hand the AI your entire glossary, however large. It now only receives the terms that actually appear in the mission text.
Self-hosted AI provider support. Organisations can now point REQQA's AI calls at a self-hosted or alternative OpenAI-compatible endpoint instead of a cloud provider.
Bug fixes
Capitalised glossary terms could never be matched. A defect in the term matcher meant any glossary term containing a capital letter — for example "Gherkin" — was never recognised in a requirement or story, however clearly it was used, because the artefact text was checked in lower case while the stored term was not. This is fixed: matching is now case-insensitive, and also recognises singular and plural forms of a term automatically.
A failed AI call left no record. If an AI provider call failed outright, nothing was logged. Failures are now recorded with the reason, so a failed analysis run can be diagnosed rather than silently vanishing.
Behind the scenes
87 scenarios covering the glossary-matching changes are executed by automated tests, plus a live verification run against the model whose output originally exposed the matching defect.
Upgrade notes
Existing glossary term indexes are marked for a one-time refresh, applied automatically the next time each requirement, story, or mission is analysed — no immediate bulk reprocessing is triggered, so there is no delay applying this release.