Get Pipeline Health

Where ingestion is stuck, across every tenant (admin).

Three per-tenant stall signals — failed analysis, failed archiving, and intakes sitting in failed / needs_input / awaiting_analysis — which the customer experiences identically ("my documents never arrived") and which need different fixes.

poison is the one that justifies this being a platform surface. file is a cross-tenant cache keyed on sha256: an existing hash reuses the row as-is, and a terminally failed row is never re-queued. So one file that failed once makes every FUTURE tenant ingesting those same bytes born stuck, claimed by no queue, with nothing on any screen to explain it — and a shared Swiss supplier's invoice template is exactly the kind of bytes that recur across tenants. The row belongs to no company, so no per-tenant page can show it.

The companies count on each poison row is the finding, not decoration: 1 would be one tenant's bad document; more than 1 is a cache entry poisoning the estate. Only rows above 1 are listed.

⚠️ There is no bulk requeue yet. This surface diagnoses; clearing a poisoned entry is still a manual operation. Said plainly because a health page with no remedy is easily mistaken for one that already fixed things.