Targeted burst runbook — newly-canonical companies only

Follow-up from the 2026-04-19 /companies canonicalization pass. This is a small burst, explicitly scoped to the rows that just got promoted, and is gated on CompanySpec modules existing first. The full 170K-entity backlog identified in the 10× cluster debrief is out of scope here.

Prerequisites

1. Create a dedicated Supabase environments row for each of the following slugs (one row per slug, id is generated always):
   • unitedhealth-group-incorporated
   • optum-care
   • optumrx
   • optum-bank
   • birdsong-capital Use scripts/ensure_company_environments.py (or equivalent) so the IDs can be pinned in both seed_company_deep/environments.py and frontend/src/lib/company-environments.ts.
2. Land seed_company_deep/data/<slug>.py for each of the five companies, modeled on optum_360.py / unitedhealthcare.py.
3. Run the generic seeder (python -m seed_company_deep <slug> --apply --skip-embed) once per slug so entity_details rows carry seeded_by=seed_company_deep_<slug> — the /companies/[slug] deep API filters on that tag.

Provisioning (single 4×ADA cluster)

• Follow the SSH rule for the Linode GPU burst server: -o StrictHostKeyChecking=no -o PubkeyAuthentication=no -o ConnectTimeout=10 with sshpass + LINODE_GPU_BURST2_PW from .env.
• One cluster is enough for the ~5-company smoke pass; do not duplicate until the first 100 entities complete cleanly.

Models + context

• NER: gemma-4-E4B-it-Q4_K_M.gguf (alias gemma-4-e4b), 16k context.
• Embedding: jina-embeddings-v5-text-small-retrieval/v5-small-retrieval-Q8_0.gguf (alias jina-v5-embed), 16k context.
• Both served via llama-server (no LM Studio, per project rules).

Guardrails (to avoid the 10× debrief's re-enrichment loop)

• Gate the enrichment priority query on enrichment_count < 3.
• Cluster-partition the queue (e.g. entity_id % 1 = 0 for a single-cluster run, % N per cluster otherwise).
• Use FOR UPDATE SKIP LOCKED or a claim_batch RPC so parallel workers do not fight over the same entity.
• Populate enrichment_model + enrichment_host on every write so future debriefs can attribute throughput correctly.
• Cap the overall batch to ~100 entities for the smoke pass, then pause and sanity-check before scaling.

Sequencing

1. Smoke-run 100 entities on a single 4×ADA cluster.
2. Spot-check /companies/optum-care, /companies/optumrx, /companies/optum-bank, /companies/unitedhealth-group-incorporated, and /companies/birdsong-capital (after flipping its override_href). Confirm the OrgStatisticsBlock renders non-empty headcount, function mix, and maturity bars.
3. If results look good, duplicate to a second cluster and expand the batch. Do not expand before step 2 passes.