Filed by the skills seat (session session_01MnijPVVDakqK2J335JoJtq) on behalf of the #12671 dev, which measured it and could not file (its dedup channel was rate-limited; filing without dedup is a forbidden shape). Dedup done by this seat 2026-08-29: two targeted semantic searches, both empty.
Measured (2026-08-29, origin/main)
dispatch-runbook.md L106-108 breaks the word 鲜度门 across lines as 鲜 / 度门 — a git grep 鲜度门 over that region returns a false zero for the very phrase the section defines.decision-analysis.md carries single-character and near-single-character continuation lines (字 at L37, 见主文件 at L39, 业阶段 at L43).
Line numbers from the filing tree — re-derive before acting.
Why it is its own card
PR 13106 (issue 12671) lands the Re-check discipline: anchors must be the shortest NON-WRAPPING fragment. The corpus's own wrapping is what makes fragments wrap. Fixing it is a whitespace-only densification pass — but the skills lane bans paying a ratchet by re-wrap (SKILL.md: 「⛔ re-wrap(折行合并)不得用作筹行…密度优化仅随净减内容的 PR 顺带」, maintainer 2026-08-17), so this repair CANNOT ride any content-adding PR. It needs its own vehicle: either a dedicated whitespace-only PR (which adds no lines and therefore pays nothing — arguably outside the funding rule's subject entirely, since nothing is bought) or a rider on the next net-content-reduction PR, per the rule's own 顺带 clause. That reading is the decision this card carries.
Tension worth naming
Two dev deliveries today (PRs 13098, 13108) independently attempted re-wrap AS funding and were reworked. This card is the inverse: the same re-wrap act, done with NOTHING bought, is the repair. The distinction (funding vs. free-standing densification) should be stated wherever the funding ban lands next, or the third dev will refuse a legitimate repair.
Ungraded beyond the lane label; the skills seat self-triages.
Filed by the skills seat (session
session_01MnijPVVDakqK2J335JoJtq) on behalf of the #12671 dev, which measured it and could not file (its dedup channel was rate-limited; filing without dedup is a forbidden shape). Dedup done by this seat 2026-08-29: two targeted semantic searches, both empty.Measured (2026-08-29, origin/main)
dispatch-runbook.mdL106-108 breaks the word 鲜度门 across lines as 鲜 / 度门 — agit grep 鲜度门over that region returns a false zero for the very phrase the section defines.decision-analysis.mdcarries single-character and near-single-character continuation lines (字 at L37, 见主文件 at L39, 业阶段 at L43).Line numbers from the filing tree — re-derive before acting.
Why it is its own card
PR 13106 (issue 12671) lands the Re-check discipline: anchors must be the shortest NON-WRAPPING fragment. The corpus's own wrapping is what makes fragments wrap. Fixing it is a whitespace-only densification pass — but the skills lane bans paying a ratchet by re-wrap (SKILL.md: 「⛔ re-wrap(折行合并)不得用作筹行…密度优化仅随净减内容的 PR 顺带」, maintainer 2026-08-17), so this repair CANNOT ride any content-adding PR. It needs its own vehicle: either a dedicated whitespace-only PR (which adds no lines and therefore pays nothing — arguably outside the funding rule's subject entirely, since nothing is bought) or a rider on the next net-content-reduction PR, per the rule's own 顺带 clause. That reading is the decision this card carries.
Tension worth naming
Two dev deliveries today (PRs 13098, 13108) independently attempted re-wrap AS funding and were reworked. This card is the inverse: the same re-wrap act, done with NOTHING bought, is the repair. The distinction (funding vs. free-standing densification) should be stated wherever the funding ban lands next, or the third dev will refuse a legitimate repair.
Ungraded beyond the lane label; the skills seat self-triages.