[garrytan]

Daily flaky-fix pass (GSTA-57), group 1 plus one group 5 detector.

plan-design-with-ui-scope went red three times on the flaky ledger with ceiling_reached; no acknowledged Design focus and subsequent review question (runs 39315041557, 39344306426, 39363705692, attempt 1 each, slice 4). In all three the agent asked Step 0D and answered an issue-labelled finding. Two of them already pass on main's T1b detector. Run 39363705692, the one after T1b, worded the 0A rating "I rated this plan 4/10 for design completeness" and the option "All 7 areas (recommended)", and the detector only accepted "on design completeness" and "dimensions|passes". The repair defines the input class instead of adding another spelling:

• The rating is any sentence that scores design completeness out of 10. That covers the template's own 0A example, "This plan is a 3/10 on design completeness", which the old regex rejected.
• The all-seven option is any option label that leads with "All 7" or "All seven" followed by a word.
• The focus-title, option-count and option-order requirements are unchanged.

tpa-absent-darwin (run 39315041557, slice 14): "If you want to use Aside: download it (macOS 15+) at aside.com, open it, sign in, and re-run" is the contract pitch with a pronoun. The canonical branch now takes Aside|it before the macOS 15+ floor and keeps its clause-start, negation and quoted-mention guards. Short forms without the floor still count only at a sentence start.

Evidence

• Free replay of every captured slice-4 observation (3 failed and 4 passed attempts, including 39355689639) through the repaired predicates: 7/7 pass. Before the repair, 39363705692 a1 fails.
• New fixture design-ui-focus-capture-39363705692.json (public native calls only), with a positive replay and a no-focus negative. New negative controls: a score in a different sentence from "design completeness", a score out of 7, a non-design completeness score, all-7 without a focus title, a non-leading, issue-labelled or bare "All 7" option. All of them still fail.
• TPA: the 39315041557 transcript is now a pitch capture and counts exactly once. New negatives: negated, line-negated, reported and quoted "download it (macOS 15+)" still don't count. All earlier captures and controls are unchanged.
• test/plan-design-with-ui-fixture.test.ts 28/28, test/third-party-actions.test.ts 58/58, free-fixtures and touchfiles tests, typecheck, and the typecheck:test ratchet (none new) all pass. Both new regressions fail against main's detectors.
• Focused paid runs, Opus 5.5: plan-design-with-ui-scope 5/5 (129, 106, 147, 153 and 127 s); tpa-absent-darwin 5/5 (15–19 s, $0.41 total).

Gate integrity: test-only detector edits. No threshold, trial count, case kind, quarantine or budget changed.

[trunk-io]

😎 This pull request was merged.