API: enumerate as a table — change, public, class, per-row verdict
SOLUTION: a holistic design pass is one decision; behaviour and new API still split out
JUDGEMENT: four verdicts in priority order — comment only when problem/solution/impact are unclear
put the VERDICT line in every slot of the output template, not just the convention note
PERFORMANCE: the evidence a perf block must carry — counts not clocks, both builds one run, multiple N, banked repro
every slot ends in a VERDICT line; JUDGEMENT curates them instead of re-deriving
PERFORMANCE: a measured degradation is request-changes, not a tradeoff note
A11Y & I18N is its own slot, wired to rubric section 1 and section 9
VISUAL EVIDENCE: label every difference intentional or not; intentional cites the PR's problem, Design Conventions or rubric §5
ossification: a new concept is marked [Needs human judgement], not argued
ossification: ask who else has this problem and what they called it
scatter the audit rubric across the template: §3 in API, §2 in THEMING, §1+§9 rendered in VISUAL EVIDENCE, §4+§9 source in BREAKING
API absorbs ossification — answered per enumerated line, not per PR
JUDGEMENT: name which half needs a human — design owner stands in for design, never for API
STEP 5b: grep the siblings before writing a convention finding; R31: merge prose is not a trailer; R32: green CI is not evidence CI ran
R33: 'do we want this at all' is a human step — capability PRs never merge unattended
R32: verify the class, not the case — from the #5321/#5214/#5110 retractions
R31: merge reads a trailer not prose; R31b: reconcile with our own open review; R31c: check main did not invalidate the PR
R3l-cite: cite the rule and its sentence, not the page name
R3l: design/API judgement calls have wiki pages; BREAKING gains behaviour, PERFORMANCE gains effects, THEME TARGETS renamed THEMING
Review-5420: request-changes posted; contrast finding, nightly follow-up noted
Review-5418: request-changes posted; verdict moved twice, nightly follow-up noted
Review-5434: visual-gate clock freeze — approve, merged by the loop
Review-5426: approve + merged; set-level findings for the #5417 theme-target rows
Record 2026-08-24 incoming run: 5321, 5394, 5414, 5419
R17: gate 3 clean on #5385 and #5428; correct the render-prop count to its enumeration
R17: record reviews for #5385 (minQueryLength) and #5428 (renderToast), both drafted not posted
Review-5353: round two — request-changes, and the critic overturning the reviewer
Point the derived-gate note back at #5421
Point the derived-gate note at its knot, not the closed GitHub issue