Federal atlas source and candidate audit

All 1,106 catalog datasets retain eight demographic groups and Dem/Rep/Other/Nonvote counts. County values sum unrounded fitted block-group counts before display rounding. All 2,671 referenced numerical gzip streams pass SHA-256 and CRC verification. This report is rebuilt from the selected bundle metadata and unrounded export totals; it does not inherit an earlier source audit.

Every newly selected model matches its certified reference-stage Democratic, Republican and Other category targets separately within 0.0001 vote. This verifies the recorded reference scope and does not establish completeness of ballots or write-ins omitted by that source. Explicit source coverage qualifications remain attached to each race. Only the six unchanged prior joint-national 2020/2024 variants may retain an existing calibration difference, explicitly recorded in `retainedNationalCalibrationDifferences` (0 datasets). Candidate labels never change numerical totals.

Independent raw Census checks passed for all 596 prepared state/vintage caches, and independent demographic, vote-margin and source-cast checks passed for all 1,249 corrected prepared inputs (1,213 original prepared jobs and 36 separately certified supplements). Black CVAP uses Census line 5 and Asian CVAP line 4. All 1,100 selected corrected models use those certified inputs. The six earlier joint-national 2020/2024 variants use independently verified, correctly labeled original inputs and remain unchanged.

Margin is Democratic minus Republican share of all valid votes, including Other. Shift is the change in that margin in percentage points. Turnout is all valid votes as a share of eligible CVAP, and turnout shift is its change in percentage points. Aggregation uses counts before calculating shares.

California refinements use numeric SWDB precinct returns and the publisher registration-to-block crosswalk. Suppressed or unlocated counts remain certified county/district residuals. Any county/district exceeding a certified party total is rejected and replaced by the certified county allocation; margins are never globally rescaled to conceal an excess. California 2022 CD02 uses verified SoS party affiliations: 229,720 Democratic and 79,029 Republican votes. SWDB source documentation.

Iowa retains mixed source precision. Available 2022 precinct boundaries are conservatively qualified for documented municipal boundary corrections; 2024 fine coverage is about 95%. County/district residuals do not imply observed precinct detail. Per-race bundle receipts and the machine-readable audit record actual source fractions and qualifications. Iowa precinct GIS.

477 datasets have a quantified coarse vote fraction; 438 have a positive coarse residual. Unknown fractions remain null and explicitly unquantified. 55 datasets warn that some local Other votes are spatially imputed; certified aggregate Other totals remain fixed. Preserved qualifications identify historical demographic proxies, district portions, source/reference corrections and coherent ranked-choice election stages.

Same-party nominees and independent/minor-party candidates remain available as party-category estimates. Same-party nominees are aggregated, and Other can contain multiple candidates. The displayed margin includes Other in valid votes but cannot identify individual nominees' demographic voting. Single-candidate uncontested results remain available. Candidate labels come from exact-race certified references. Primary federal returns.

Source-line discrepancies and any retained prior-national calibration differences are explicitly recorded in the JSON report. Known unlocated/water block groups, including New York, retain their counts in county totals; polygon coverage is separately audited. These are ecological estimates of demographic voting behavior, not observed individual ballots.

Independent map checks

All 1,106 model tables passed canonical Census identity, finite/nonnegative count, vote marginal, and county-to-block-group sum checks. Display counts round to 0.001, and county sums are checked against the corresponding rounding bound.

All 436 current congressional filters conserve each demographic’s geographic weight. County labels match Census NAMELSAD. President and Senate district filters use 2024 boundaries consistently in both years of a shift.

The geometry audit found 68 water-only block groups and 1 zero-area district contacts. Positive votes assigned to water by source allocations remain in county totals and are explicitly qualified as unlocated allocations. The largest affected share is 0.583% of valid votes in NY-2024-CONG-08.

7 omitted land block groups were restored across 8 district scopes using exact Census block-group intersections with the actual district boundary. Their numerical estimates are unchanged.

Demographic source correction

Black and Asian Census CVAP categories are checked against their named source lines. Corrected local electorates, House district allocations, demographic fits, county sums, and browser table columns are independently verified. Existing precinct, municipality, and county source-resolution qualifications remain explicit; fine map polygons do not imply equally fine observed returns.

Within-county precision screen

All 1,106 model tables were screened for nearly uniform All-voter block-group margins, weighted by valid votes and allowing for display rounding. Uniformity flags are diagnostic, not proof of coarse data or an error; a single precinct or homogeneous area can legitimately be uniform. Declared coarse residuals are checked before unexplained flags are listed. The compact evidence is downloadable below; the full county ledger is retained locally.

Source resolution across the published catalog

Each race is counted once. Quantified fine sources have no declared coarse votes; mixed residuals use larger units for some votes. Publisher block estimates are disaggregated estimates whose original reporting-unit precision is unquantified. All remain estimates of block-group totals. County or district totals cannot supply observed within-county detail.

Download the evidence

• demographic_correction_summary.json.gz
• demographic_order_hosted_audit.json.gz
• missing_geometry_land_audit.json.gz
• missing_geometry_repair_table_audit.json.gz
• repaired_geometry_hosted_audit.json.gz
• source_resolution_audit.json.gz
• systemic_table_audit_corrected.json.gz
• systemic_table_audit_iowa_final.json.gz
• water_allocation_qualifications.json.gz
• within_county_precision_review.json.gz
• within_county_precision_summary.json.gz