GRANT: tools/progress.py → pxx-a5, scoped to the UNCITED-RESOLVE check

Granted 2026-08-30. tools/progress.py is every lane's tool on every ticket move, which is why pxx-a5 correctly declined to edit it unasked and filed bug-t-a-resolve-that-never-wrote-a-placeholder-is-uncited-and-nothing-says-so [T p45] instead. The coordinator holds it and is the bottleneck; the design is pxx-a5's, complete, with three cautions it derived itself. Granting rather than re-implementing from a description.

Scope: the UNCITED-RESOLVE check inside check(), and its devtest. Nothing else in the file. The coordinator is not editing progress.py while this is open.

The defect

A ticket resolved by a hand-written Log line never gets a PENDING-COMMIT placeholder — so sync.sh has nothing to fill and check has nothing to count. It is uncited and silent, which is strictly worse than PENDING-COMMIT, a state that at least greps, counts, and has a tool that knows how to repair it. Measured: 3 of 681 in the 2026-08-16..31 window.

The three cautions, all pxx-a5's, all load-bearing

1. Warn, never repair. check's own bookkeeping note records that ~82% of bad citations look fixable by git-log matching and are not.
2. A date floor. 881 of 2806 done/ tickets predate the convention and would flood it with 881 findings on day one — which is how a guard gets muted (129, and STALE-EDGE-HIDDEN's calibration comment).
3. Some resolutions are not commits. "Profiled; findings filed; two levers measured and declined" is a real outcome, and a check that flags it teaches people that uncited means nothing.

Three distinct routes to the same worthless-check failure, which is why this is a grant and not a ticket description.

Convention it must encode

Ruled 2026-08-30: for a HAND-FILLED citation, cite the FIX, not the close. The "cite the resolve commit" convention exists because that is what sync.sh can automate at push time, not because the resolve commit is more useful. When a human fills it and the two are distinguishable, cite what a future reader needs — what changed — and name the close in the Log line. The automated resolve + sync.sh path is unchanged; a placeholder can only ever know the resolve sha.

RETURNED 2026-08-30 — unused, and the check landed somewhere else

tools/progress.py was not edited. The grant is returned unspent; the coordinator may reopen the file immediately.

The grant rested on a number I had already retracted

It cites "3 of 681" and the ticket as filed. Both were superseded on the same day, in 72538cc79, which closed bug-t-a-resolve-that-never-wrote-a-placeholder-is-uncited-and-nothing-says-so as rejected/. The 3-of-681 used an ad-hoc test that counts a ticket discussing a commit range 8fb3f776..b3fd1c76 as cited. Under the house definition (CITATION_RE plus a line-start key — what _audit_citations already uses):

The freshest window is the date floor, and it still yields 31. Caution 2 does not survive contact with the data — as a standing check report this is the muted guard its own cautions were written to prevent.

What changed: the scope, not the check

The same 31 findings are worthless as a standing report over the tree and valuable one at a time, addressed to the person who just resolved the ticket, at the moment fixing it costs one line. So it belongs beside verify_citations_landed in tools/sync.sh — scoped to what this run resolved — not in check() scoped to the tree.

That also dissolves caution 2 entirely: there is no date floor to choose, because "this push" is the floor.

manifest_resolved is captured beside the subject manifest — --diff-filter=A --no-renames over the resolved buckets. --no-renames is load-bearing: a git mv backlog/ → done/ is reported as R, and --diff-filter=A would never see the very move this is about.

Calibrated against the live board, as asked

Over the last 400 commits touching a resolved bucket: 469 resolutions, 43 would nudge (9%). But 23 of the 43 are regression-, decide- and grant- slugs whose resolution is a verdict rather than a commit — the watcher files and closes regression tickets from tstate; a decision closes when the user rules; a grant closes when it is returned, as this one is. Nudging those teaches that uncited means nothing, which is caution 3 in live data.

Excluded by prefix: 19 of 469 — 4%, about one nudge per twenty-five resolutions, and every survivor is a bug-/feature- ticket that changed code and cited nothing.

All three cautions are honoured, and two of them by construction rather than by care: warn, never repair (it prints, exit 0 — the push already succeeded); the date floor (there is none to get wrong); some resolutions are not commits (the prefix exclusion, plus the message's closing line saying so in words the reader sees).

Convention encoded

The message says "cite the FIX, not the close, and name the close beside it" — the ruling, in the place a person is standing when it applies.

Guard

tools/sync_citation_guard_devtest.py — 36 guards, 0 FAIL, up from 24. Section 8 covers the nudge and the three excluded prefixes; section 9 is the scope guard that stops it firing when a resolved ticket is merely edited, which is what every write-up appended to old work would otherwise do.

One more caught by running rather than reading: the function's early-out was still keyed on manifest_tickets alone, making the nudge unreachable whenever that list was empty. Harmless live — manifest_resolved is a subset — but accidentally correct is not correct, and the devtest passes the two separately, so it failed on the first run.

Log

• 2026-08-30 — resolved, commit c1a8e092c.

Folded from a duplicate copy that was sitting in backlog/

check's DUPLICATE-SLUG scan caught this ticket existing in backlog/ and done/ — the backlog copy was a headless fragment (no frontmatter, so NO-FRONTMATTER flagged it too) holding my closing note. Diffed before removing, per the scan's own instruction; the done/ write-up above is strictly the fuller one. The single framing worth keeping from the fragment:

23 of the 43 are regression-/decide-/grant- slugs whose resolution is a verdict — that is caution 3, in live data, at 53% of findings.

Naming the fraction is what makes caution 3 an argument rather than a worry. Noted also that my own coordinator check flagged my own duplicate within an hour of my filing it, which is the check working on its author.