About — Scrutineer

What this is

Scrutineer analyzes Alaska's Cast Vote Records. A Cast Vote Record (CVR) is the ballot-level electronic record of every contest selection on every scanned ballot. It shows exactly what was marked. It contains no voter identity. The Alaska Division of Elections publishes CVRs for elections with ranked-choice contests so the public can verify the tabulation. Scrutineer loads those files and shows what precinct totals cannot.

The data

Election	Ballots	Contests
2022 Primary + Special General (Aug 16, 2022)	192,289	63
2022 General (Nov 8, 2022)	266,797	92
2024 General (Nov 5, 2024)	340,778	73

Source: official Dominion Democracy Suite CVR exports from the Division of Elections results pages. The August 2022 file contains two elections because they shared one ballot: the regular top-four primary and the ranked-choice special general for U.S. House. No CVR exists for the 2024 primary. That ballot had no ranked contest, so the Division did not publish one.

The analyses

Insights — automatic detectors run over every contest: vote-mode divergence (a candidate winning absentee while losing Election Day), Condorcet failures, exhausted ballots exceeding the winning margin, heavy bullet voting, and coalitions that never reached the deciding round.

RCV Rounds — round-by-round replay of the ranked-choice count, with a redistribution table after each elimination showing where those ballots went. Includes counterfactual mode: remove any candidate and re-run the count.

Head-to-Head — every pair of candidates as a two-way race, shown as a color-coded margin grid with the closest matchup flagged. Flags any contest where the head-to-head winner lost the actual tabulation.

Coalition Flows — a Sankey diagram of 1st → 2nd → 3rd choice paths, plus each candidate's second-choice breakdown.

Exhaustion — ballots that ranked only one candidate, and ballots with no candidate left by the final round, grouped by first choice and by house district.

Ranking Depth — how many ranks each candidate's voters used: bullet-voting rate and average depth, split by vote mode and by house district.

Vote Mode — first choices split by how the ballot was cast: Election Day, early, absentee, question. Includes elimination transfers by vote mode.

Cross-Contest — pick any two contests and see how the same ballots voted in both, including second-choice crosstabs. This is the analysis only a CVR can support.

Anomalies — house districts z-scored by overvote rate, highest-exhaustion precincts, and precincts ranked by overvote and ambiguous-mark rates.

Simulator — a forward model for a Democrat-vs-multiple-Republicans ranked race. You set first-choice shares. Each candidate carries an adjustable transfer profile: bullet/exhaust rate and the R-vs-D split of their transferring ballots (defaults from the 2022 calibration: 21.8% exhaust, 75.8% of transfers stay R). Optional turnout programs model GOTV pushes — added voters with a chosen lean — and results show baseline vs. with-program, in votes when a ballot count is set. Output is a hypothetical model, not a prediction.

Every table on every tab has a CSV export.

Method

The ranked-choice replay implements the rules in AS 15.15.350 and 15 AAC 25.270:

Ranks count in order. Ranking the same candidate twice collapses to the highest rank.
An overvoted rank (two or more candidates at the same rank) exhausts the ballot when the count reaches it.
Two consecutive skipped ranks exhaust the ballot. A single skip does not.
The last-place candidate is eliminated each round until two remain.

Rankings are built from all non-ambiguous marks in the export. The Dominion IsVote flag marks only the choice counted in the first round, so it is not used to read rankings. Adjudicated ballots use their modified record where one exists.

Verification

The replay is checked against certified results.

Check	Certified	Scrutineer
2022 Special General, post-write-in tallies	75,799 / 58,973 / 53,810	75,799 / 58,974 / 53,810
2022 Special General, final round	51.48% / 48.52%	51.48% / 48.52%
2024 U.S. House, final round	51.32% / 48.68%	51.34% / 48.66%
2024 Ballot Measure 2, yes share	49.88%	49.85%

Small differences come from write-in resolution and adjudication that happened after the export snapshot.

Limits

CVRs cannot be joined to the voter file. The ballot records are anonymous by design. Geography and vote mode are the finest grain available.
Only elections with a published CVR can be analyzed. That means the three above.
The export is a snapshot. Adjudication after the export date is not in it.

API

Every view is backed by a public JSON endpoint.

/api/elections
/api/{election}/contests
/api/{election}/insights
/api/{election}/errors
/api/{election}/crosstab?c1={contest}&c2={contest}
/api/{election}/contest/{contest}/rcv · /condorcet · /flows · /exhaustion · /depth · /groups

Built by Alaska Targeting. Data is the state's own record. Check the math yourself with the API.