● Atlas Commons · live build status

Everything in flow

Every atlas, every bronze / silver / gold layer, every serving surface, and what's blocking what — read from the live atlas-commons-lakehouse catalog, 2026-06-20.

8
atlases · one catalog
318
tables live in unified R2
9
gold serving surfaces
2
cross-atlas joins proven
live partial / sampled migrating blocked / not started L5 regulated routing

The eight atlases

b / s / g = bronze / silver / gold tables in the unified catalog · route = the gold serving surface

Coverage

insurance · health plans
live
12
bronze
3
silver
1
gold
plan_serving_surface · 22,059 ACA marketplace plans, structured for comparison
ACA plans only cover federal-marketplace (FFM) states; 26 state-based-exchange states read 0 — add SBE data + a footnote. 13 gated sources named residue.

Care

healthcare · hospitals
L5
18
bronze
2
silver
1
gold
hospital_quality_serving_surface · 5,432 hospitals, CMS star ratings. Compares, never diagnoses.
NPPES un-capped 200 → 9,606,683 providers (FULL registry) via the monthly bulk file + typed schema + chunked write.

Property

home risk
live
9
bronze
4
silver
1
gold
state_risk_profile · storm + wildfire + disaster-mitigation + FEMA NRI composite: risk, EAL, dominant hazard per state
NFIP Claims un-capped → 2,721,780 (FULL flood-loss record) via projected CSV API + resilient chunked write. NFIP Policies 73.6M stays a named 10k sample — the bulk file 403s datacenter IPs (gated, confirmed from a runner).

Finance

consumer finance · L5
L5
2
bronze
2
silver
2
gold
lender_directory (FDIC) + state_lending_access (NEW HMDA): denial rates by race + Black/White disparity. Facts + routes, no discrimination finding.
CFPB FULL 15,979,491 complaints + FDIC FULL 27,836 institutions (credit bureaus dominate: TransUnion 4.3M, Equifax 4.2M, Experian 3.8M). Both un-capped, no sample remaining.

Justice

civil legal · L5
L5
6
bronze
3
silver
1
gold
legal-aid directory + statute-of-limitations · 193 LSC providers; the free-path referral surface
FJC IDB un-capped → 10,857,396 federal civil cases (FULL SY1988-2026) via chunked streaming write. CourtListener (54.2 GB bz2) stays residue — exceeds runner disk.

Compliance

business registration · L5
L5
1
bronze
1
silver
1
gold
entity_registry_serving_surface · 3,347,057 GLEIF legal entities (FULL golden copy) for counterparty due-diligence
Un-capped 1k→3.35M via the GLEIF golden-copy bulk. OFAC SDN + IRS 990 (entity-legitimacy pair) still to build.

Singulariki

work · labor · the original
complete
67
bronze
51
silver
54
gold
172/172 COMPLETE ✓ — fully migrated to the unified catalog (incl. licensed Lightcast via private R2 bucket): O*NET (46) + 11 reference (git) + OEWS 414k + 9 academic AI-exposure via R2-staging (Anthropic AEI 482k · OpenAI 81k · Microsoft 36k · Felten 6k · Frey/Brookings/Dingel/ILO).
R2-staging mechanism PROVEN (public bucket + staging_bytes helper). Lightcast (33,040 skills) landed via private R2 bucket (boto3 S3). The original 172-table atlas is now 100% in the unified catalog. NEXT: money-pipelines (74 tables, code migrated) via catalog-copy.

Money-Pipelines

public money
complete
19
bronze
9
silver
46
gold
74/74 COMPLETE ✓ — migrated into the unified catalog via catalog-to-catalog copy (vendor_genome 74k, sam_entity 884k, money_scent_signals…)
Done. 46 gold serving surfaces (proof_bomb_queue, no_bid_economy, incumbent_recompete_radar…). USAspending + SAM.gov federal-contract intelligence.

Cross-atlas joins — the unified-catalog payoff

questions no single atlas can answer · one R2 SQL warehouse · live
JOIN coverage ⋈ care ⋈ property ⋈ finance ON state  → live on /commons (56 states: plans · hospitals · lenders · storm · mitigation)
JOIN finance.lending_disparity ⋈ property.dominant_hazard ⋈ property.nri_risk ON state  → live: "where lending bias meets disaster risk"
Queued: + HMDA × ACS × NRI by county (the "can a family build wealth here" surface, once ACS lands).

What's blocking what

the constraint list, by severity

State / staging — resolved

Singulariki 172/172 + money-pipelines 74/74 fully in the unified catalog. R2-staging (public + private buckets) proven; the keystone corpus-to-R2 fix is done.

No-sampling residue — cleared

GLEIF 3.35M · CFPB 16M · NPPES 9.6M · FDIC 27,836 · NFIP Claims 2.72M · FJC IDB 10.86M — all un-capped to the full federal record. Only NFIP Policies 73.6M remains a named 10k sample (gated, below).

Heavy / gated sources — named, never faked

NFIP Policies 73.6M (bulk file 403s datacenter IPs; API past the $skip cliff) · CourtListener 54.2 GB bz2 (exceeds runner disk) · sam_gov key-gated · OFAC + IRS 990 not built. Each PROVEN-gated and named as residue, never sampled-as-complete.

Visible payoff

Property page lacks the NRI section; finance page lacks the HMDA surface; commons ACA-plans-0 needs a footnote; this status page is the new at-a-glance truth.
SOURCE · live GET on atlas-commons-lakehouse namespaces/tables · row-truth via R2 SQL · 2026-06-20
Companion: runtime architecture (how to run this forever) → · cross-atlas surface → · home →
Counts are live table counts in the unified catalog; "target" reflects each atlas's full medallion. Capped/sampled sources are named, never presented as complete.