test plan generator
Acceptance criteria in, traceable test cases out — with a quality gate that fails the run until every case passes nine explicit criteria.
client program generic: a pricing-compliance feature on a dealer-inventory surface. no vendor, field, or ticket specifics.
The 2025 version turned feature specs into structured test plans. The 2026 version grew teeth after a client QA vendor rejected a four-row test workbook. The pipeline: fetch the acceptance-criteria page from the wiki → catalog it → a gap review that flags missing criteria as proposed, never silently added → derive Given/When/Then cases with action-first numbered steps and {PLACEHOLDER} data instead of fake values → validate every case against nine explicit quality criteria as a hard gate (the run asserts the validator fails on bad input) → export a workbook with how-to-use, test cases, an AC-traceability matrix, and the proposed-AC list.
First run: 68 cases covering 71 of 72 criteria, and the validator caught real defects in the generator's own output — 25 failures, then 8, then 4, then zero across three loops. That the gate rejected the first draft is the point: a generator that grades itself with the same model that wrote the cases learns nothing.
Full case study in progress — artifacts pending clearance or writing time. The insight below is already earned.
Traceability is the client-confidence artifact, not the case list. And a validator that rejects the generator's own output on the first run is the generator working — fixed verb lexicons don't scale; explicit quality criteria do.
the insight