Post-merge QA (glm)

Documentation-only merge: one bullet added to the beta1 declared matrix in docs/verification.md stating that model routes are user-owned and outside the supported matrix, with each result recording the backend, provider and model as context only. The bullet is present in the diff exactly as described in the PR body, is consistent with the surrounding matrix wording (platforms, runtimes, Pi experimental), and introduces no code, no behavior change, and no contradiction with the linked issue #426 acceptance criteria, which concern the platform/runtime/capability matrix. No defects found.

No findings.

Verdict: clean

Post-merge QA (deepseek)

Lens: evidence + test design. Docs-only change: one bullet in docs/verification.md (line 92) that states model routes are user-owned and outside the beta1 matrix, which is exactly what the PR promised and what a maintainer clarification like #426 needs (it adds a reason-bearing statement instead of a new matrix dimension, so the declared 2x2x12 scope is unchanged). Evidence is proportional for documentation: the only factual claim (the bullet sits after 'Runtime versions') is a manual read any reviewer can repeat, and the python3 -m unittest discover -s tests claim is reproducible. The untested part is the bullet's own second sentence, 'Each result records the backend, provider and model it ran with': the section that actually tells a verifier what to record is 'Reporting results' (line 109), and it still lists only 'backend/version'. Test condition that is missing: a check that the reporting guidance and the matrix bullet name the same result fields.

Suggestion (1)

• 'Reporting results' does not ask for provider or model, so line 92 promises a record the guidance never requests (docs/verification.md:109): Line 92 asserts that each result records the backend, provider and model it ran with, but 'Reporting results' (line 109) enumerates only 'platform, archive filename and SHA-256, backend/version, QA role, exit status, and bounded redacted output'. A verifier following the reporting section literally produces results with no provider and no model, which is the situation the new bullet says it fixed. The claim and the guidance should name the same fields, or the bullet should say the draft result template does not exist yet.

Verdict: clean