review/code-gpt Firstmate read the diff and evidence (native before/after frames, hero decode proof, episode art chain, gear centered per owner answer); apk-debug green; full serial suite NOT green (4 PersonPickerSessionTest timeouts, cause unisolated, disclosed in the PR body), accepted by firstmate with real-box person-switch verification before anything ships to the owner
review/design-claude Firstmate read the diff and evidence (native before/after frames, hero decode proof, episode art chain, gear centered per owner answer); apk-debug green; full serial suite NOT green (4 PersonPickerSessionTest timeouts, cause unisolated, disclosed in the PR body), accepted by firstmate with real-box person-switch verification before anything ships to the owner
review/code-gpt Firstmate read the PR evidence: TV-only labels and seed grouping, recorded-wire acceptance matrix, fixtures contain no secrets (checked), native before/after frames, full serial suite 697/190/17 green plus apk-debug; no tokens, ring or spacing changes; device check follows
review/design-claude Firstmate read the PR evidence: TV-only labels and seed grouping, recorded-wire acceptance matrix, fixtures contain no secrets (checked), native before/after frames, full serial suite 697/190/17 green plus apk-debug; no tokens, ring or spacing changes; device check follows
review/code-gpt GPT review blockers fixed and read by firstmate: poster progress inset back to Title.cardProgressInset (220 px track) with test tag + regression; library focus-follow effect keyed on columns/focusedRow/header/rowHeight with a 199<->200 regression; token values/ring bounds/toolbar gap passed review on c17447c5; walked on the real box
review/code-claude Hotfix verified on the real Xiaomi box (B3 runs, no crash across Home/Discover/Movies/Shows/title/Settings/Requests); 5-file delta read by firstmate: ShellState.rememberedView no longer writes on read, 3 regression tests incl. snapshotFlow and an audit
review/design-claude B1-B3 re-measured by firstmate from the lane measurement file against the locked numbers (all within 2 px); delta is the three scoped fixes plus an additive main merge; N1/N2 3-4 px offsets remain as listed follow-ups
