Corpus & Ingestion
Corpus growth is the forcing function for the discovery claim, and ingestion is scheduled with a budget-derived cap that refuses rather than overspends. One screen for how big the catalog is, how much of it is affinity-bearing, what the last runs did and declined to do, whether the jobs are on, and how much budget is left.
Catalog coverage
How much corpus there is, and how much of it carries anything to match on.
Budget posture
Ingestion has its own daily sub-ceiling, well under the platform-wide kill switch. It exists so a big ingest cannot starve interactive discovery — and so a run with $2 of headroom does not attempt 5000 embeddings and abort at record 40.
Ingestion runs
Every run appends a row to corpus_ingestion_runs recording what it wrote, what it skipped and why it stopped. A run that was refused by a guard and a run that failed are different rows here and are different rows on screen.
Enrichment sweep runs
The weekly sweep over stale public-catalog organizations (catalog_enrichment_sweep_runs, migration 160). It queues enrichment; the worker enriches later, so nothing on this panel is a count of organizations enriched.
Partnership verification ledger
Every live search the platform ran to check a rumoured partnership, confirmed or refused. A refusal writes nothing to the graph, so this table is the only record that the check happened at all — which is what makes "the sweep refused everything" distinguishable from "the sweep never ran".
Schedule
Both jobs ship disabled — a run spends real budget against the live catalog, so an operator turns them on deliberately, per environment.