Skip to main content

0.8.5 — Three more ways to be wrong

Released 2026-08-10.

A well-formed requirement has nine characteristics under ISO/IEC/IEEE 29148:2018. REQQA checked three of them. This release adds three analysis steps and takes that to seven.

What's new​

R-N — Necessity and Traceability. Asks whether a requirement should exist, and whether it comes from where it says. It reads the Rationale section your template already asks for — until now, nothing did — and reports requirements that are orphans, with no parent, no recorded source and no linkage to the mission.

R-P — Appropriateness. Asks whether a requirement states a need or a solution, and whether it sits at the right level. Detection is semantic rather than a list of product names, so "a relational store with row-level locking" is treated exactly like naming a database. Every finding quotes the span it judged and the need that span appears to serve, so you can check the reasoning instead of taking it on trust.

R-W — Writing and Conformance. Asks whether a requirement states one thing and follows its template. Most of it needs no AI at all — escape clauses, open-ended lists, unexpanded acronyms and unfilled placeholders are decided by pattern in milliseconds. Only sentences that genuinely need judgement are sent to the model, so a clean requirement costs nothing. It also checks your template's sections, sentence pattern and placeholders through three independent probes. A probe that cannot run is reported as skipped, never as passed.

These four steps — with R-D Definitions — are now recommended by default on every requirement template, because every requirement needs to be necessary, appropriately pitched, singular and conformant. Existing templates and requirements were updated in place; your own selections were added to, never replaced.

Cleanup and reformat. The cleanup tab now says so plainly when a requirement has nothing to fix, instead of showing an empty table. Section reformatting gained a guard: a proposed rewrite that drops content — a policy number, a threshold, a named routine — is refused, and the original is left untouched.

Bug fixes​

  • Analyses could run with an empty glossary index, which made every defined term look undefined. Indexing now happens once, before any analysis starts.
  • Deleting a glossary term explains that the effect is organisation-wide, tells you how many artefacts use it, and offers to add it to the do-not-suggest list in the same step.
  • Terms defined by a standard your requirement cites are no longer reported as high-severity undefined vocabulary. Citing your sources should not be penalised.
  • Re-indexing a large application no longer times out. All bulk re-index actions now run in the background with progress you can follow.
  • Exporting a release report as PDF failed when any requirement, story or glossary entry contained a backslash.

Behind the scenes​

Analysis prompts were restructured so the parts that do not change between steps are sent once and reused, cutting the cost of a full sweep substantially.

Upgrade notes​

No action required. Existing recommended-analysis selections were extended, not replaced — if you had chosen specific steps for a template, those choices are intact and the universal steps were added alongside them.