Anchor recorded replays to surviving commits, and content over positions

[claude]

Fixes #13739

One bullet added to the record-discipline block of
.claude/skills/pm-dispatch/references/dispatch-runbook.md — the section that
governs how cards and rulings are written down. That section asks for
reproductions, but says nothing about what a record is pinned TO, and both of
the usual pins expire silently.

The two failure shapes, both measured

A replay stated at its own PR head. The branch is squash-merged and
deleted, the objects go unreachable, and the command block still reads
replayable. On the card behind this PR, two of the three shas in its subject's
repro block were unresolvable — both PR-branch commits — while the one that
survived was the main-side diff base.

A citation stated as a line number. Files grow and the number drifts. One
triage round hit three dead line-number citations, and every one of them was
relocated only by searching for a symptom string.

Same defect in both: a record that reads checkable and is not. A number nobody
can re-derive is a number nobody can check.

The rule

State a replay against a commit that survives — the main merge commit, or
base plus the PR number — never a PR head. Where the PR head genuinely is the
subject, say so and record the inputs beside it (the diff, or the changed
hunks) so the run can be rebuilt without the object. And anchor content, not
positions: symptom strings or distinctive identifiers rather than line numbers.

No gate, deliberately

Scanning card bodies for sha liveness would false-positive on every unmerged
PR, where a PR-head sha is alive and correct. That restraint is the card's own
and triage concurred with it, so this PR adds no script and no check.

Budget and verification

Line budget on the runbook: 271 → 276 lines, ceiling 278 (headroom 7 → 2).
The five added lines were wrapped by the ratchet's own wrapLine, so each is a
canonical break: widths 118 / 116 / 119 / 119 / 97 bytes against the 120-byte
cap, and no table row is introduced (that file's widest-table-row pin is 0).
No issue numbers appear in the operative text, per the id lint.

Gate union re-derived in the worktree with
node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands
and run after the final commit, at HEAD f952cf8fd. Every exit code was
captured by redirect-then-capture, never after a pipe:

GATE_EXIT=0 :: pnpm --filter @objectstack/lint run check:doc-formula-expressions
GATE_EXIT=0 :: pnpm check:agent-test-spelling
GATE_EXIT=0 :: pnpm check:doc-authoring
GATE_EXIT=0 :: pnpm check:pm-governed-merges
GATE_EXIT=0 :: pnpm check:pm-skill-id-lint
GATE_EXIT=0 :: pnpm check:pm-skill-ratchet
GATE_EXIT=0 :: pnpm check:skill-frame-sync

Ratchet verdict for the edited file, verbatim:
check-skill-line-ratchet: .claude/skills/pm-dispatch/references/dispatch-runbook.md is 276 lines (ceiling 278; headroom 2).

check:doc-formula-expressions first refused with exit 3 — PREREQUISITE NOT
MET, @objectstack/formula then @objectstack/lint unbuilt in a fresh
worktree. Both were built and the gate re-run to the exit 0 above; the refusals
measured nothing and are recorded here only so the sequence is reconstructible.

Changeset: none. The diff is .claude/**-only and publishes from no
package, so it takes skip-changeset per the workflow's own prescription.

Draft only — governed surface, human merge.

Generated by Claude Code

---

Generated by Claude Code