Hold reasons: a typed hold kind from one shared list, so a new merge.decided reason can't be emitted without its plain words (by design; #816 follow-up) #824

Outcome: a merge.decided reason can't be emitted without its plain words, by design. Each reason carries a typed hold kind from one shared list. mergeDecision, mergeDecisionFor, mergeOrWait's pushes, combinedGate, rulesetHold, conflictWaitReasons and the weakening reasons draw from it, and src/hold-words.ts is keyed by it. A new kind without words then fails to compile.

Why (measured on #816 at 640b6cd, round-6 review #816 (comment)):

• First run: a card follows the first task from File one to its PR; the Start card counts only steps before Start (#741) [PROTECTED: owner merges] #816 (First run: a card follows the first task from "File one" to its PR, and the Start card stops counting steps that come after Start #741's First task card) guarantees two things by design. Both are runtime checks:
  • the card shows only fixed words: the fold checks each field, holdWords uses its fixed table, and any unknown prefix reads "Another reason (see the Board)";
  • the page never goes down: firstTaskCard's try.
• One guarantee is lint-only. "Every kind merge.decided carries gets its own plain words" rests on a source-scan test, test/first-task-progress.test.ts. The owner's 2026-10-19 rule accepts a lint only as a backstop.
• The scan has named gaps. Each only loses the plain words; nothing leaks.
  • Emitter forms it misses:
    • emit ( with a space;
    • emit\n(;
    • emit<'merge.decided'>(;
    • emit(/* c */ '…';
    • an emitter by a constant event type, by another method name, or by emit?.(.
  • Write forms it misses:
    • push?.(, push (, push.apply, Array.prototype.push.call;
    • an index write, ??=;
    • a shorthand { hold } or a hold variable.
  • Regex literals: paired regex literals holding quotes can hide a spread or computed key from the recovery exemption. This needs deliberately crafted code.
• Why it isn't done in First run: a card follows the first task from File one to its PR; the Start card counts only steps before Start (#741) [PROTECTED: owner merges] #816: the fix touches src/merge-policy.ts, src/coordinator/merge-flow.ts, src/coordinator.ts, src/pr-rules.ts and src/conflicts.ts, which open PRs hold (Merge: a PR that removes or weakens a test, lint or guard waits for the owner (#783) [PROTECTED: owner merges] #820, Required checks count only from the App or workflow pinned for their name [PROTECTED: owner merges] #788 and others).

Constraints:

• mergeDecision's decisions are byte-identical before and after. Only the reason's carrier changes, never what holds.
• Stored merge.decided events keep their reasons: string[], so older rows still read. The kind is added beside each reason, or derived from it in one place.
• Once the type makes words compulsory, the source scan in test/first-task-progress.test.ts can shrink to a backstop for untyped writes, or go. Removing it is the owner's call (rule 10).

Deliverable:

- test: test/hold-words.test.ts
- command: |
    npm run typecheck

Acceptance, each with a fail-without test:

• a new hold kind added to the shared list without words fails npm run typecheck, shown on fixture code;
• every reason that mergeDecision, mergeOrWait, combinedGate, rulesetHold and conflictWaitReasons emit carries a kind from the list;
• mergeDecision's output is identical before and after (a snapshot over test/merge-policy.test.ts inputs).

Review point: L3, because it touches the merge policy. Protected, so the owner merges.

Related: #741 / #816 (the card and its words), #706 (plain words for every hold, slice 2's table), #820 (the weakening reason).