Dataset Changelog

Full-data review found relationship defects; new purchases paused

Full archive integrity and table counts passed. New sales paused: millions of driver/occupant rows have duplicate candidate keys or no matching crash. Linked tables require source-level repair. Source records were not refreshed.

Full archive and table validation completed; limitations retained

Full archive integrity, table counts and crash relationships passed. The delivered snapshot has unique table keys; coordinate sentinels, parked/working-vehicle scope and differences from revised source files remain disclosed. Source records were not refreshed.

Full archive and table validation completed; limitations retained

Full archive integrity, table counts and record keys passed. The listing now distinguishes 2,189,824 complaint/component rows from 1,582,110 ODI complaint numbers. Source-date anomalies and later source revisions remain limitations. Source records were not refreshed.

NTSB six-table snapshot rebuilt with verified joins and explicit coordinate quality

All six scoped source tables reconcile field by field to 224,906 delivered rows. Unique keys, every event/aircraft relationship, derived fields, CSV/Parquet values, sample membership and ZIP integrity passed. Missing/invalid coordinates and absent narratives/findings remain explicit. Rebuilt from the official September 1, 2026 avall.zip revision retrieved September 19, 2026. Actual event coverage is 2008-01-01 through 2026-08-27.

Full archive and table validation completed; limitations retained

Full archive integrity, table counts and record keys passed. The product remains a dated snapshot with unavailable advisory flags and unresolved differences from current official affected-product listings. Source records were not refreshed.

EPA TRI rebuilt with typed quantities, versioned facility joins and verified source provenance

Both pinned EPA source years and all 122 source fields reconcile to 158,687 reports. Facility relationships, exact decimal quantities, CSV/Parquet values and every ZIP member passed full checks. Original totals and explicitly documented source formula discrepancies are preserved. Rebuilt from EPA annual CSVs pinned September 19, 2026; coverage remains reporting years 2022–2023.

FAERS 2023 case versions and deletions reconciled; one-time offer restored

Rebuilt all seven CSV/Parquet tables using one latest-case cohort and the supplied 2023 deletion lists. Source key multiplicities and every CSV/Parquet value match; no unmatched parent reports or indication/therapy drug groups remain. Full ZIP CRC and published archive SHA-256 checks passed. Drug dosing multiplicity is preserved as documented by FDA FAQ 53. Source records were not refreshed.

Full archive and table validation completed; limitations retained

Full archive integrity, table counts and rate-formula checks passed. Candidate establishment keys have duplicates or missing fields; extreme rates and absent ZIP prefixes remain disclosed. Source records were not refreshed.

Full archive and table validation completed; limitations retained

The complete production ZIP passed SHA-256, every member CRC, complete CSV parsing and all-column Parquet decoding. Complaint IDs, counts, coverage and narratives agree with the earlier repair. Current source API reconciliation remains unverified. Source records were not refreshed.

DOT airline snapshot rebuilt across all 84 months with all BTS source fields and preserved flight records

All 84 monthly source files, all 109 BTS fields and every source row reconcile to persisted Parquet with an independent CSV parser. Monthly counts, nulls, exact delay sums, dates, provenance keys, sample membership and complete ZIP integrity pass. Duplicate flight candidate keys and source anomalies are retained. Rebuilt from all 84 official BTS Reporting Carrier monthly ZIPs for 2018–2024, retrieved and hashed September 19, 2026. Restores October–December 2021 and all 2022–2023, and preserves duplicate candidate keys instead of applying the former cleaner’s deduplication. This is the pinned source revision, not a claim that historical files never change.

NOAA archive rebuilt from pinned sources; event relationships reconciled and decimal measurements preserved

All 1,940,615 events, 1,748,259 source location rows and 23,766 fatality rows reconciled to pinned NOAA inputs. The primary location table contains 1,465,798 rows; 282,461 excluded rows remain in an audit table. All five CSV/Parquet pairs, source-field comparisons, event joins, numeric derivations and ZIP member checks passed. Rebuilt from NOAA source revisions pinned September 19, 2026; primary event coverage remains 1950–2024.

FEMA claims rebuilt from NFIP v3 with preserved geography, exact payments and full source reconciliation

All 2,725,989 IDs and 84 source fields reconcile between official CSV/Parquet and persisted output. Independent exact annual financial totals, complete delivered-value equality, diagnostic fields, sample membership and ZIP integrity pass. Missing and irregular source values are retained and disclosed. Rebuilt from the official NFIP v3 September 9 bulk files, with September 8 asOfDate, retrieved and hashed September 19, 2026. The v2 endpoint is frozen and scheduled for removal October 15. Acceptance is against this pinned v3 revision, not an exact recreation of the former snapshot.

Full archive and table validation completed; limitations retained

Full archive integrity, table counts and record keys passed. Coordinates remain absent; source completeness and predictive usefulness of descriptive species scores are not established. Source records were not refreshed.

USDA indemnities rebuilt with corrected financial fields and exact source reconciliation

Every source field and all 4,173,884 records reconcile to persisted Parquet across 35 years. Annual financial totals, typed conversions, source row keys, CSV/Parquet equality, sample membership and ZIP integrity passed. Source duplicates, negative amounts and invalid/missing dates are retained and disclosed. Rebuilt from all 35 official RMA annual files for commodity years 1989–2023, retrieved and hashed September 19, 2026. Current files contain 1,872 fewer rows than the former snapshot; acceptance is against the pinned source revision.

New purchases paused; package limitations published

Model years 2010–2023. New purchases are paused. The promised timeline table is not present. Every complaint_trend is stable and every do_not_drive_flag is zero; these fields must not be treated as verified trends or absence of safety warnings. The delivered model years are 2010–2023, not 1985–2025. Fatality counts are not exposure-adjusted rates. This is a listing correction, not a data refresh.

Complete public inventory offered free; performance-history claims removed

26 states and 60 waterways; age reference year 2024. This is a physical inventory, not traffic, throughput, delay or performance history. age_years was calculated relative to 2024; it is not a current age field. The public CSV already includes every row and all 56 fields. No paid dataset purchase is needed. This is a listing correction, not a data refresh.

New purchases paused; package limitations published

1999–2016. New purchases are paused. There is one mortality table, with no SDOH or joined table. The file does not contain the advertised age, race, sex, cause-of-death or deprivation breakdowns. years_potential_life_lost is missing throughout. This snapshot supports county-year totals only. It cannot support the previously advertised demographic or cause-specific analyses. This is a listing correction, not a data refresh.