Planner telemetry, filed into the planner-bug home. It blocks nothing and is in no sprint. The correction to the tree was submitted as plan cmt5j2uz900dei3phybrnq52g at the time of filing.
motir run MOTIR-3264 (2026-08-23), which halted at run.md guard #4 before creating a worktree. Full evidence table in the run comment on MOTIR-3264.
MOTIR-3274 exists to make a Neon control-plane credential reachable from a run session. Its Context refs name two consumers: MOTIR-2853 (motir-core) and, in plain language, "MOTIR-3264 — the motir-gateway half of the same measurement, still open, which will need the same credential."
Its four acceptance criteria were then written entirely over motir-core's project:
GET /api/v2/projects/{id} returns 200 — satisfied against snowy-truth-13928044.ep-billowing-wildflower-au5s16ri is recorded — motir-core's endpoint.Every one of them is true. The card closed done on honest evidence. And the second consumer it names is exactly as blocked as before: the key is scoped to snowy-truth-13928044, the boundary is enforced server-side (GET /api/v2/projects → 404 subject_project_id:"snowy-truth-13928044"), and motir-gateway's database is ep-crimson-dew-ai3vmlx0 in a different Neon project.
The shape, stated generically: a prerequisite card's acceptance criteria named ONE consumer's need while its prose promised TWO. Criterion 2 even asked for the scope to be stated — and stating it was enough to pass, because no criterion asked whether that scope covered the consumers the card had named. Nothing downstream could catch it either: readiness.ready was true, openBlockers empty, validate_work_item valid: true, and the one advisory on MOTIR-3264 pointed at an unrelated merged PR. The blocker was not a work item failing to reach done — it was a done work item whose deliverable did not cover the consumer, and no edge can hold that.
notes.html #30) and a soak window no edge can express. This is the next state along: the credential card was planned, was built, and was closed correctly. Its criteria simply did not reach one of its own named consumers.run.md's artifact-obtainable check — pull it from where its CONSUMER will obtain it — is the closest existing rule and is the right instinct, but it is written for one consumer and one artifact. MOTIR-3274's filer did pull, as the consumer, from where that consumer reads. There were two consumers and only one was standing there.plan-rules/ gains, or sharpens, a rule for a card whose DELIVERABLE is consumed by more than one other card: every consumer the card names — in its criteria or in its prose — appears in an acceptance criterion as a check performed against that consumer's own target, not against the first one. A manual credential/provisioning card is the canonical carrier, and the rule states that naming a consumer in Context refs creates a criterion, not a note.notes.html mistake #30 family is extended, not duplicated. #30 says every external prerequisite is its own planned subtask. The addition is that such a subtask's criteria must be written over the SET of consumers it serves; whether that is a new lesson or a sharpening of the existing one is decided here.notes.html (frozen since MOTIR-3296) — a row in motir-ai src/seed/lessons.base.ts plus its generated migration, written generically, with sourceRef naming this card. (Written by the filing run; this criterion records that it is owed and closes on the row being present.)motir run MOTIR-2853.cmt5j2uz900dei3phybrnq52g — the submitted correction (the widened-credential card + the blockedByAdd edge on MOTIR-3264).cmt5j15z500d4i3phegz52470 — abandoned unclosed by the same run (a modify appended with a placeholder temp-ref, and a proposal cannot be withdrawn). Cancel it. That a mis-ordered append is unrecoverable is itself worth a corpus line: append the add, read its id back, and only then append the modify that references it.