Skip to main content
Source inventory

Where the data comes from — and which source wins when they disagree.

We favor official and academically maintained datasets, and apply a clear priority order when multiple sources overlap. Every row can be traced back to its origin.
How to citeMargin convention
Source families
3
Historical, modern, census
Row count
19.4M+
Election results across 51 state partitions
Validation cadence
Daily
Automated reconciliation checks
Coverage

What we have, by office and year

Mixed (county + precinct + block)Historical (Algara-Amlani / ICPSR)Modern (MEDSL / VEST / NYT / Klarner)
186018801900192019401960198020002020President1868–2024U.S. Senate1908–2024Governor1865–2025U.S. House1976–2024CD Presidential2008–2024State Senate1968–2023State House1968–2023
Election Results
19.4M+
rows across 51 state partitions
Elections Covered
224
1868–2024
Contests
151,488
President, Senate, Governor, House, State Leg
Counties
3,143
with geometry, demographics, and embeddings
Congressional Districts
435
with redistricting vintage tracking
Precincts (2024)
163,926
with full boundary geometry
Census Blocks
8M+
50 states, disaggregated election data
Geographic Entities
8.3M+
states, counties, CDs, SLDs, precincts, blocks
OfficeCoverageGeographyNotes
President1868–2024County + Precinct + BlockAll 51 states. Block disaggregation for 50 states.
U.S. Senate1908–2024CountyAll regular and special elections.
Governor1865–2025CountyIncludes off-year elections (Virginia, New Jersey).
U.S. House1976–2024Congressional districtMEDSL source. 10,869 contests.
CD Presidential2008–2024Congressional district (120th canonical; 118th archive)435 CDs. Current profiles use 2026-adopted / 120th Congress boundaries.
State Senate1968–2023State legislative districtKlarner dataset, all 50 states.
State House1968–2023State legislative districtKlarner dataset, all 50 states (excluding Nebraska unicameral).
Source families

Three families, one lineage

Click any card to expand. Each source shows coverage, geography, license, and a link to the upstream provider when available.
Historical data

Pre-2000 federal and state historical returns.

Modern elections

County, district, and precinct-era results.

Census and geometry

Population weights, demographics, and boundaries.

Priority order

Which source wins when they disagree

When sources overlap, this hierarchy decides which row lands in the warehouse. Differences are logged and reviewed.
PeriodCounty sourcePrecinct source
1868–1949Algara-Amlani → ICPSR 0001N/A
1950–1968Algara-Amlani → ICPSR 0013N/A
1969–1999Algara-Amlani → State SOSN/A
2000–2015MEDSL → Algara-Amlani → State SOSState SOS (partial)
2016–presentMEDSL (county)VEST (2020) → NYT (2024) → State SOS
Worked example
Cook County, Illinois · 2020 Presidential
MEDSL is the primary county source for 2000–present. Algara-Amlani overlaps but is held in reserve for disagreement review. If a row diverges by more than the validation threshold, the anomaly is logged for manual inspection and the higher-priority source keeps the warehouse row.
Validation

What we check — and what happens when something fails

Automation catches most problems. Humans resolve the rest.
Vote reconciliation
Do precinct totals sum to county totals? Do county totals sum to state totals? Any discrepancy over a threshold is flagged.
Geographic completeness
Is every expected county, district, or precinct present for a given contest? Missing geographies get logged by state and year.
Cross-source consistency
When multiple sources cover the same race, we compare margins and vote totals. Large disagreements are routed for manual review.
Margin sanity checks
Does the margin math match? Our convention is (D-R)/total. A mismatch usually means a bad denominator, not a bad source.
Boundary integrity
Crosswalks between precincts, blocks, and districts are checked for coverage. Orphan blocks or unmapped precincts surface immediately.
Failure handling
Flags are written to a review queue. We hold the warehouse row pending resolution rather than ship bad data silently.