Pending Industry Orgs

Organizations waiting on Rooster's industry-map desk, worst gap first.

Same {"conversations": [{"id": ...}, ...]} shape every other pending surface uses (/cs/tickets/pending, /issues/intake/pending, /playbooks/author/pending) so tools/rooster_worker/worker.py's existing fetch_pending drains it with no new client code -- naming kept consistent across surfaces rather than inventing an industry-specific key.

THE QUEUE IS ALREADY BUILT AND ALREADY RANKED. This is a thin read over enrichment_service.select_tier_a_adp_queue, which orders by domain.rank_adp_queue: the segment furthest below the 90 percent ADP target first, then the oldest attempt, then the organization missing MORE surrounding data. Nothing new decides priority here; a second ranking would be a second truth.

WHY THE TIER A SLICE IS THE RIGHT QUEUE FOR A DESK, and not a limitation: AUTONOMOUS_FACILITY_ENRICHMENT.md S5's cost ladder puts these at the floor -- an org whose website resolves to .gov/.us "need[s] NO search and NO LLM" for its cheapest facts. Handing the desk the cheapest work first is what keeps the model seat free for the judgment cases and for everything else Rooster does; see tools/rooster_worker/desks.py's industry block for that argument in full.

Read-only, and deliberately so. Claiming, writing and promoting a fact all happen through the enrichment service's own promotion rules (enrichment_service.promote_org_field), which enforce source tier and never overwrite a human-set field -- a desk that could write around those rules would be a second, weaker copy of them.