Planning bug: a falsified-premise card named TWO of the ten cards its finding invalidated — the referrer sweep was done by memory, not by grep
Telemetry about the planner. The correction is already applied — this card holds the record, and blocks nothing.
What happened
MOTIR-3492 recorded a finding that falsified a premise: a loading.tsx at a route-group root destroys the 404 on every route beneath it that decides existence, so the "group frame" every card in MOTIR-3440's subtree was written against cannot exist.
Its own step 3 named the cards that needed restating: "Restate MOTIR-3440 and MOTIR-3442." Two.
Running it found ten:
| card | the clause the finding falsified |
|---|---|
| MOTIR-3440 | its TITLE — "settings gets a frame of its own" — named in the card |
| MOTIR-3442 | its TITLE — "instead of the group's" — named in the card |
| MOTIR-3443 | its TITLE and deliverable 1 name app/(authed)/settings/loading.tsx — the exact file the finding forbids. Not named |
| MOTIR-3441 | the shared settings boundary its shape was drawn for. Not named |
| MOTIR-3444 | /items/[key]/edit is a decider. Not named |
| MOTIR-3445 | /plans/[id] is a decider. Not named |
| MOTIR-3447 | /sprints/[id]/report is a decider. Not named |
| MOTIR-3446 | the group frame it falls back to. Not named |
| MOTIR-3449 | the boundary its seams and ratchet are written over. Not named |
| MOTIR-3450 | "watch the frame land before the content" — a different moment now. Not named |
The most expensive omission is MOTIR-3443, whose title and first deliverable name the forbidden file outright. Dispatched as written it would have built app/(authed)/settings/loading.tsx above the five existence-deciding routes under settings/ — re-creating the defect MOTIR-3492 exists to prevent — and been caught only by tests/navigation/loading-boundary-guard.test.ts after the work was done.
Why it happened
The sweep was performed from the two cards the reporting run had read while doing its own work, rather than from a search of the tree for the premise's own words. The finding was precise; the enumeration of what it reached was not. Both survivors it did name were cards the run had opened; every card it missed was one it had not.
The graph could not have supplied the answer either. MOTIR-3440 is blocked_by MOTIR-3492 at the STORY level, so its children read as held — correctly — while their TEXT went on asserting a premise nobody had re-read. A readiness edge holds a card; it does not correct it.
The correction, already applied
- MOTIR-3440, MOTIR-3442, MOTIR-3443 — title and body amended, each with a dated ⚠️ AMENDMENT block; MOTIR-3440's carries the per-child table.
- MOTIR-3441, MOTIR-3444, MOTIR-3445, MOTIR-3446, MOTIR-3447, MOTIR-3449, MOTIR-3450 — commented with the clause that moves for each.
The takeaway
A finding that falsifies a PREMISE owes a grep for that premise's own words across the tree, not a list of the cards the finding's author happened to have open. The two are indistinguishable in the written card — both read as "and these cards need restating" — and only one of them is complete. Where the premise has a literal carrier (a filename, a phrase, a mechanism name), that literal IS the search; where it does not, the search is over the subtree the falsified card gates.
Recorded as a lesson as well as here.