Versions, tests and promotes prompts as controlled artefacts rather than strings embedded in code.
Treats prompts as configuration with a lifecycle: versioned, tested against an evaluation set, promoted between environments, and rolled back when a change regresses.
Any deployment where prompts change with meaningful frequency, or where non-engineers should be able to refine them.
Small stable prompt sets that genuinely never change.
Where a prompt determines output entering a record, changing it is a control change — versioned, approved, dated. A tolerance widened in a prompt is no different from one widened in a rule.
Prompt changes shipped without evaluation, where a wording improvement for one case silently regresses another nobody tested.
A narrative-generation prompt versioned alongside the evaluation corpus, so a change that improves readability and reduces citation rate is caught before release.
Two different absences share this shape. Foundational solutions get built whatever the domain, so no domain links them; the rest are solutions this domain genuinely does not reach for. v_ai_solutions_unlinked separates the two.