Measure which prompting strategy works best for explanations #411

As the AI-layer owner, I want a measured comparison of prompting strategies for ai/explain.py, so that the current prompt is a tested choice rather than an unexamined first draft.

Acceptance criteria

• At least two alternative prompt structures (for example structured/JSON-constrained output versus the current free-text instruction) are run against the same held-out set used in A held-out, repeatable score for the explanation layer (#346) #385/A repeatable score for the question layer, run in CI (#393) #407.
• The comparison is scored the same way A held-out, repeatable score for the explanation layer (#346) #385 already scores explanation quality, so the result is directly comparable to the current baseline.
• A decision (keep the current prompt, or switch) is recorded with the measurement as the stated reason.

Shape

AI layer, tooling and measurement only. No change to the F-1 finding shape or to /api/findings's response contract.

Why this is worth doing

The current prompt in ai/Modelfile was never compared against an alternative. This is the cheapest possible test of that gap, reusing infrastructure that already exists rather than building new measurement tooling.

Epic: Candidate idea (AI layer) - not yet agreed as a specialist pick
Size: not yet estimated
Assigned: @patelankeet2

Proposed 27 October 2026 as a candidate feature idea. Not yet discussed or agreed by the team; open for refinement or rejection at the next planning session, the same status every other idea in this batch carries.