Statistical screening workspace
Review Center
Start with the priority list, inspect why an area was flagged, then use the graphs and ward rows to decide which official records to request.
- 1Pick an areaStart with the flagged-area list.
- 2Read the flagSee which threshold was crossed.
- 3Inspect local rowsReview the underlying official totals.
- 4Export evidenceDownload a focused CSV for follow-up.
- 5Request recordsUse official records to verify the pattern.
Statistical screening checks
Test Status
Several screening methods rely heavily on precinct-level data. Any turnout source using registration counts that exclude Election Day updates must be treated as provisional in states where same-day registration or late denominator updates apply.
Presidential elections only
Historical Baseline
Compare earlier presidential elections with 2024 before treating a visible pattern as new. These charts help answer whether a shape was already present in earlier data. They do not determine why a pattern exists.
Historical baselines are loaded state by state from the best available official or clearly documented public source rows. Some states use native county, precinct, or municipality rows. Others use harmonized comparison rows placed onto common geography so years can be compared. Any allocation or harmonization is labeled in the selected state's source notes.
Across presidential elections
Statewide comparison
| Election | Source row type | Democratic | Republican | Other | Total | DEM share | REP share |
|---|
Start here
Compare all four election years at once
- Begin with the four distribution panels below. They show 2012, 2016, 2020, and 2024 side by side.
- Compare the shape of the bars, the dashed blue average line, the gold busiest-bucket line, and the skew value.
- Use the single-year scatter chart afterward when you want to inspect one election in more detail.
Vote-share distribution histogram
Republican Vote-Share Distribution Across Years
How to read this
Each panel is one election year. Each bar counts how many local source rows landed in a Republican presidential vote-share range. Think of sorting result rows into labeled buckets: 0-5%, 5-10%, and so on. The dashed blue line is the average row. The gold line marks the busiest bucket. Compare the overall shapes across election years.
Skew is a simple measure of whether the bars lean more heavily toward one side of the chart. A value near zero is more balanced. A positive or negative value describes the direction of the lean; it does not explain the cause.
This is a screening comparison, not proof of a cause. The displayed rows use the selected state's loaded historical source method. A statewide turnout-versus-vote-share heat map requires reliable row-level turnout denominators for each year, which are not available for every state.
All-years summary
Presidential Vote Share Across Years
How to read this
Each point is one presidential election for the selected area. Think of it like checking several snapshots instead of one: if a share level was already visible in earlier elections, the line makes that easier to see.
Optional single-year drilldown
Selected-Year Vote-Share by Vote-Count Scatterplot
How to read this
Each dot is one source row for the selected year. Left-to-right is candidate vote count and up-and-down is candidate share. Compare the shape with the 2024 chart, while keeping the displayed source label in mind.
Coverage and provenance
What is ready, and what is still missing
- Statewide presidential totals reconcile for every displayed year.
- The native official WEC 2016 recount workbook matches the LTSB 2016 statewide total.
- The harmonized LTSB 2016 layer shifts seven votes from Buffalo County to Trempealeau County relative to the native WEC recount workbook. The app keeps those row types separate.
- The official LTSB 2024 harmonized layer matches the native WEC 2024 presidential totals statewide and in every county. Its ward values remain labeled as harmonized comparison rows because LTSB documents population-based disaggregation.
- Seventy LTSB geography rows contain masked presidential values. They are preserved as missing and excluded from the graph-ready rows rather than treated as zero votes.
- One hundred forty January 2025 LTSB ward features contain no 2024 presidential values. They are also preserved as missing and excluded from graph-ready rows rather than treated as zero votes.
- Ward boundaries changed between elections. The older LTSB series use a common harmonized geography so shapes can be compared cautiously; they must not be described as exact election-year native ward rows.
- Native official 2012 and final 2020 ward-level WEC workbooks are still being requested.
Coverage, limitations, and provenance
Data & Sources
See what is complete, what is partial, and which official or local records support each part of the app.
Turnout coverage
County Tracker
Checking turnout coverage...
Sources & Provenance
Used for map shading, county table, statewide totals, candidate breakdown, CSV export, and selected-county details.
Each state records the official source file used for map rows in the Source Planner and generated app data.Used for vote-share by vote-count scatterplots, presidential-versus-comparison drop-off histograms, selected-area filtering, and review drilldowns when a state has same-grain rows loaded.
Availability varies by state. Some states have precinct rows; others remain county-only until a same-grain comparison source is mapped.Used for drop-off charts when President and a statewide or district-level comparison contest are available at the same row grain.
The comparison contest is documented in each state's app data and review graph source notes.Used to draw state, county, or district polygons depending on the selected state.
Most county maps use U.S. Census TIGERweb geometry; exceptions are labeled in the selected state's source inventory. TIGERweb serviceUsed for turnout histograms and denominator-warning labels. Denominator timing differs by state, so warnings are retained when rows are statewide-only, active-voter-only, post-election, or otherwise limited.
Open Source Planner for the selected state's loaded turnout route and caveats.Used for the Historical Baseline tab where multi-year presidential rows have been loaded. Source row type and limitations are shown inside the tab.
Local generated files include historical-data.js and reconciliation summaries for states with historical coverage.Used for graph-category references, reporting context, registration timing, and source limitations.
Election Truth Alliance methodology referenceState expansion workspace
Source Planner
Track the official result sources behind the app by state and county, with each state's loaded capabilities shown separately.
Selected state
Selected state
Building source summary...
County breakdown
Election Number Sources
Building county source rows...
| County | Certified Result Source | Ward Detail Source | Turnout Denominator Source | Follow-up |
|---|
Plain-language reference
Methodology
These screens are statistical screening tools. A flag means that an area is worth reviewing with official records. It is not proof that tampering occurred.
Plain-language glossary
What the technical terms mean
- LTSB
- Wisconsin Legislative Technology Services Bureau. It appears only in Wisconsin historical-source notes as a nonpartisan legislative service agency whose GIS team publishes mapping layers.
- Harmonized
- Adjusted onto one common map so different election years can be compared. Think of redrawing several old neighborhood maps onto the same modern street map. Some row values are estimates allocated by population, not exact original ward records.
- Native official row
- A result row taken from an election-specific official report without converting it onto a shared historical map. This is closer to the original reporting format for that election.
- Reporting unit
- The local result bucket used in an election report. It may be one ward or a group of wards reported together.
- Historical baseline
- Earlier presidential elections used as a reference point. It helps answer: was a visible pattern already present before 2024?
- Turnout denominator
- The bottom number in a turnout calculation. For example, 900 ballots divided by 1,000 registered voters equals 90% turnout.
- Provenance
- The paper trail for the data: where it came from, which file was used, and what transformations or limitations apply.
Vote-Share by Vote-Count Scatterplot
- Calculates
- Whether candidate vote share rises or falls as ward vote totals get larger.
- Can flag when
- The absolute correlation reaches the selected review threshold.
- Normal explanations
- Neighborhood geography, population density, ward boundaries, and partisan clustering.
- Verify with
- Ward canvass totals, tabulator reports, ballot records, and audit materials.
- Cannot prove
- A correlation alone cannot prove manipulation or identify its cause.
Presidential-versus-Comparison Drop-Off Histogram
- Calculates
- The difference between presidential and same-party comparison-contest votes within each loaded local row.
- Can flag when
- The average difference or the count of larger row-level differences crosses a selected threshold.
- Normal explanations
- Split-ticket voting, undervotes, write-ins, ballot design, and local candidate preferences.
- Verify with
- Ward canvass detail, contest totals, tabulator reports, and hand-count records where available.
- Cannot prove
- A difference alone cannot show whether a voter intentionally skipped or changed a contest.
Turnout Histogram
- Calculates
- Ballots cast divided by registered-voter denominators for the local rows collected so far.
- Can flag when
- A source row has an unusual turnout value or uses a denominator that needs additional context.
- Normal explanations
- Election Day registration and mismatched reporting levels can make preliminary denominators too low.
- Verify with
- Final local registration totals, pollbook reconciliation, ballots-cast records, and municipal canvass reports.
- Cannot prove
- A turnout value over 100% does not by itself imply excess ballots or fraud.
Statistical Review Flags
- Calculates
- A triage signal from the vote-share, average drop-off, and row-outlier checks.
- Can flag when
- At least one current threshold is crossed for a county, major city, or rest-of-county area.
- Normal explanations
- Many flagged patterns can have ordinary demographic, geographic, or reporting explanations.
- Verify with
- The official records listed in each Review Center drilldown.
- Cannot prove
- A screening flag is not an accusation, audit finding, or conclusion.
Reproducible defaults
Starting Thresholds
The Review Center sensitivity controls can change these values. Exports include the calculated metrics so another reviewer can reproduce the screening step.
| Setting | Default |
|---|---|
| Minimum ward rows | 10 |
| Vote-share correlation | |r| >= 0.35 |
| Average President-vs-Senate difference | >= 2.0% |
| Row outlier difference | >= 15.0% |
| Minimum candidate votes for outlier rows | 100 |
| Outlier-count trigger | max(3, 5% of scope rows) |
Official configuration with hypothetical scenarios
2024 Wisconsin Equipment-Audit Coverage Simulator
This tab starts with Wisconsin's published statewide 2024 post-election voting-equipment audit configuration and models one narrow question: how often would a sample touch a hypothetical discrepancy concentrated in a limited number of reporting units?
This is not evidence that tampering occurred, not a reconstruction of the actual selected-unit list, and not a conclusion that another safeguard failed. The statewide default uses the WEC-reported sample size. Its probability calculation remains a simplified random-sample illustration because the real selection tool also applied county, equipment-type, central-count, and municipality-size constraints. A simulated miss means only that the illustrative sample did not intersect the hypothetical affected units.
What the report says
Selected-unit equipment audit
WEC reported that a public dice-derived seed was used on November 6, 2024. The selection tool returned 373 reporting units. Twelve selected units had no registered voters and were excused because no ballots were cast there.
For selected units, auditors hand tallied four contests, including President and Vice President by default, and compared those hand tallies with voting-equipment totals. WEC reported no errors solely attributable to the electronic voting system in the audited submissions.
What this model asks
Sample-intersection question
If a hypothetical discrepancy were concentrated in certain reporting units, how often would a random sample include at least one affected unit? If the sample touches an affected unit, the model marks the scenario as detected. If it touches none, the model marks the scenario as missed.
The statewide default uses the report's 373-unit selection and a derived 3,730-unit denominator based on the adopted 10% rule. The real selection tool included county, equipment-type, central-count, and municipality-size rules. The probability shown here is therefore a simplified intersection calculation, not WEC's own detection-probability estimate.
Interactive model
Change the hypothetical scenario
Repeated-trial simulation
How often did the sample miss every hypothetical affected unit?
Ready
Press "Run 1,000 simplified trials" to repeatedly draw an illustrative audit sample against the current hypothetical affected-unit pattern. This is a simplified model, not a reproduction of WEC's constrained selection software.
Not sampled
Sampled
Hypothetical affected
Sample intersects affected
Illustrative area vote totals
Paper-baseline scenario vs hypothetical altered report
Official references
What this tab is based on
About this project
Multi-State Presidential Results Explorer
This is an independent civic-data app for reviewing 2024 presidential election results by state, county, and local reporting unit where official source rows are loaded, then comparing historical baselines where public records support them.
Maps certified presidential results, shows review graphs where same-grain comparison rows are available, and tracks which turnout denominator sources have been imported. A historical tab helps distinguish patterns that were already visible in earlier presidential elections. The Wisconsin audit-coverage simulator remains as a state-specific educational model based on Wisconsin public materials.
Statistical review flags are not proof of tampering. They mark places where the numbers deserve follow-up with official records, ballot audits, or written explanations from election officials.
Coverage varies by state and capability. Each state's map, review graphs, turnout rows, historical baselines, source inventory, and caveats are shown separately in the Source Planner and generated app data.
Built and maintained by Camreyn ( https://github.com/Camreyn/ ) as an independent public-interest data project. The repository is open so data sources, formulas, and assumptions can be checked.
Contact
Report a data issue
The best way to send corrections is through GitHub Issues so the source, county, and proposed fix are visible and trackable.
Open GitHub IssuesProject links
Code and live page
Source summary
What the app relies on
- Official state election authority result files and public results portals.
- Official precinct, reporting-unit, county, or district rows where they are available.
- U.S. Census TIGERweb county or district boundary geometry where applicable.
- Official turnout reports or registration denominator files where imported.
- Wisconsin public voting-equipment audit materials for the Wisconsin-specific audit simulator.
- Election Truth Alliance methodology pages for graph type names and screening concepts, not as a source of Wisconsin vote totals.
Selected county
None selected
Winner
--
Margin
--
Total votes
--
Data confidence
What This App Can Say
Checking data version...
The vote-share and down-ballot graphs calculate patterns from the selected state's loaded official local result rows.
A flagged pattern means review further. It is not proof by itself and does not replace a ballot audit or records review.
Turnout graphs only run where denominator rows have been imported, and warning-gated denominators need extra care.
Review priority
Flagged Areas Summary
Building flagged-area list...
| Area | Type | Flags | Vote-share r | Avg drop-off | Outliers | Rows | Action |
|---|
These are statistical review flags, not proof of tampering. They are a triage list for deciding where to inspect official records first.
Evidence drilldown
Why This Was Flagged
Records that would help verify this flag
Select a county or city scope to see suggested records.
| Ward / reporting unit | Trump | Harris | Total | Trump share | Harris share | DEM drop-off | REP drop-off | Why row matters |
|---|
Other candidate detail
Statewide
57,063 other votes
Statistical screening graph
Vote-Share by Vote-Count Scatterplot
How to read this
Each dot is one local result row. Left-to-right shows how many votes a candidate received there; up-and-down shows that candidate's share of that row's vote. Think of it like a weather map: one dot is just one reading, but the trend line shows whether the readings lean in a pattern.
Simple example: if larger Harris rows also tend to have higher Harris percentages, the blue trend line rises and the check may flag for review. A flag means "look closer," not "proof something happened."
Statistical screening graph
Presidential-versus-Comparison Drop-Off Histogram
How to read this
This compares presidential votes with a same-party comparison contest in the same local row. It is like checking two receipts from the same shopping trip: if one line item is much bigger or smaller than expected, it gets marked for a closer look.
Simple example: if Trump received 1,000 votes in a precinct and the same-party comparison candidate received 950, the Republican presidential-to-comparison drop-off is 50 votes, or 5%. Normal split-ticket voting can cause differences; the chart helps show whether those differences cluster oddly.
Statistical screening graph
Turnout Histogram
How to read this
This graph runs only where turnout denominator rows have been imported for the selected state. Think of turnout as a fraction: votes cast divided by possible voters. We have the vote totals, but the reliable bottom number may still have to be collected from separate official sources.
Simple example: 900 ballots out of 1,000 registered voters is 90% turnout. But if the 1,000 count was from before a state finalized registration updates, and 80 people were added later, the real denominator is 1,080. That is why this section warns about denominator timing.
Major city split
City vs Rest of County
Building city split graphs...
Coverage & confidence
County Data Status
Building county coverage table...
| County | Vote Results | Turnout Data | Denominator Warning | Sources |
|---|
Checked But Not Imported
These sources were reviewed but did not have the denominator fields needed for turnout analysis.
County table
! marks a statistical review flag, not proof of tampering. Hover over each icon for county-specific details.
| County | Review | Trump | Harris | Other | Margin | Total |
|---|