Atlas Commons · editorial edition Atlas
Atlas Commons — Corrections & Known Issues Ledger
What Atlas Commons knows is wrong or limited about its own data: data withheld rather than shipped wrong, errors fixed, named limits, and pending fixes. Check before treating any figure as authoritative. All 25 published gold datasets have been run through a reconciliation audit (part-vs-whole, bounds, source-id, anomaly + targeted-invariant checks): it found 2 silent corruptions (both fixed below) and 0 outstanding. CC0.
Benefits
known-limit
Initial 50,000-filing SAMPLE, not the full ~800K-filing DOL Form 5500 series. Labeled as a sample wherever shown, never treated as a census; full ingest queued. An employer's absence means 'not in the sample'.
Justice
fixed
Case-mix was corrupt: the AOUSC C-3 workbook has 36 columns (merged headers, spacer, prisoner sub-columns) but the parser assumed 27, so a positional misread put the private all-civil total ($221,592) into the intellectual-property row (real federal IP ~12,500/yr). Fixed: decoded the real 36-col layout, reconciliation-verified (party total = sum of categories per district), + a reconciliation guard. Now PUBLISHED at /justice/jurisdictions; IP reads 12,500.
Care
fixed
Headline total came from the CMS by-nature report (general payments only) while companies summed the by-company report across ALL payment types -> research/ownership inflated companies above the total. Fixed: general / research / ownership kept DISTINCT (general = influence signal + reconciling companies; research + ownership separate labeled figures; a $91M ownership stake no longer reads as a payment). 979,136/979,136 rows reconcile.
Property
fixed
The structured data advertised a FIO/NAIC insurance non-renewal rate that does not exist in our data (source gold empty). Struck rather than fabricated.
Property
fixed
Cross-state storm-death comparison was a raw total (size-confounded). Added population-normalized deaths-per-million + rank; labeled the raw count. TX is #1 raw but #5 per-capita.
Coverage
known-limit
Sticker prices before income-based tax credits; most enrollees pay less. Net cost not computed (2026 subsidy rules unsettled). Scope = 30 FFM states; state-based-exchange states labeled, never $0.
Property
known-limit
Single recorded year (2025), not a multi-year trend; one severe event can swing a per-capita rank. Read as direction, not a precise standing.
Compliance
known-limit
The full GLEIF register (~3.35M legal entities), one row per LEI. Registration status + structure only (ISSUED/LAPSED, jurisdiction, name) — a point-in-time public record, never a conduct/solvency verdict. Absence means no LEI on file.
Singulariki
known-limit
OEWS lags ~12 months and suppresses small-population cells; a missing wage means suppressed, not zero. Anthropic Economic Index is non-government, labeled where used; BLS/O*NET are the authoritative spine.
Finance
known-limit
Cross-registry corporation match is a name heuristic (DERIVED), superseded by hard-code joins; a lead, not proof.
Finance
pending
Closed/merged FDIC institution pages lack closure date and successor (gold lacks FDIC CHANGEC/successor fields). Page routes to FDIC BankFind; successor re-ingest queued.
Coverage
pending
Per-rating-area/county localization (needs CMS rating-area->county file) and subsidy-adjusted net cost (SLCSP, gated on 2026 law) are queued.