Alaska, 16 August 2022 · one public record, three answers

The Six Thousand Votes That Hurt

Recount the official totals from Alaska's 2022 special House election through a published reduction of its scanned-ballot record. Mary Peltola won the instant runoff, while Nick Begich beat both rivals head to head. Then operate two 6,000-ballot counterfactuals: extra first-choice support makes Peltola lose, and voters who rank Begich above Peltola obtain that outcome by abstaining. Every total is recomputed in the browser, with the reduction's limits visible.

Here are 188,582 continuing first-round ballots, compressed into the nine ranking patterns published from Alaska's scanned cast-vote record. Press the button. Nothing is sampled, estimated, or projected. The counting rule touches every represented ballot.

Layer one · the recount

Let the low pile fall away

Alaska counts each ballot for its highest-ranked continuing candidate, eliminates one last-place candidate, and transfers that candidate's ballots.

Ready. The verified result is already shown; press Recount to watch it arrive.

Round 1 · highest continuing rank188,582 continuing

Begich53,810
Palin58,973
Peltola75,799
Begich to Palin27,053
Begich to Peltola15,467
Inactive group11,290

Round 2 · Begich eliminated177,292 continuing

Palin86,026
Peltola91,266

Peltola wins the official instant runoff. Her final margin is 5,240.

The compressed profile lands exactly on the official candidate totals and transfers. One seam must remain visible: the paper groups 11,290 ballots in a Begich-only column. Alaska's report separates that transfer into 11,243 exhausted and 47 overvotes. The sum agrees; the labels are not interchangeable.

Open all nine published ballot patterns

The symbol at left is the first choice. A missing lower name means the ballot was grouped with a truncated ranking. The paper disregards write-in candidates in this reduction.

published patternballotsfirst choice
Begich › Palin › Peltola27,053Begich
Begich › Peltola › Palin15,467Begich
Begich only / inactive group11,290Begich
Palin › Begich › Peltola34,049Palin
Palin › Peltola › Begich3,652Palin
Palin only21,272Palin
Peltola › Begich › Palin47,407Peltola
Peltola › Palin › Begich4,645Peltola
Peltola only23,747Peltola

The second count hiding in the same rankings

Put them in pairs

Instant runoff asks who remains after sequential elimination. A head-to-head count asks a different, narrower question: on ballots that express the comparison, which name is ranked higher? The answer is not an instruction about who deserves office.

The exact figures below come from the nine-pattern reduction published by Adam Graham-Squire and David McCune. They are not official Alaska pairwise tabulation totals and not totals from a direct replay of the raw CVR export.

head to headfirst candidatesecond candidatemargin
Begich vs Peltola87,859 Begich79,451 Peltola8,408
Begich vs Palin101,217 Begich63,618 Palin37,599
Peltola vs Palin91,266 Peltola86,026 Palin5,240
2 of 2

Begich wins both of his pairwise contests. That makes him the Condorcet winner under these recorded reduced rankings, even though instant runoff eliminated him first. Peltola still had the most first choices and won under Alaska's adopted rule.

Layer two · more support, then a loss

Move votes toward the winner

Take ballots from the published Palin-only group and place Peltola first, Palin second. Nobody on those altered ballots ranks Peltola lower than before. Drag the control through the point where the first elimination changes.

0 movedlast-place tie at 5,1636,000 moved
Begich first53,810
Palin first52,973
Peltola first81,799

Begich wins. Palin is eliminated; Begich defeats Peltola 87,859 to 85,451. Peltola gained 6,000 first choices and changed from winner to loser.

Source-pattern guard: the published Palin-only group contains 21,272 ballots, so the control can never move more ballots than exist.

Layer two · a ballot that hurts its voter

Now let those ballots stay home

This time, remove ballots that rank Palin first, Begich second, and Peltola last. With all of them present, Peltola wins. Remove enough, and their second choice wins instead of their last choice. This is a no-show witness, not a prediction that anyone would actually abstain.

0 removedlast-place tie at 5,1636,000 removed
Begich first53,810
Palin first52,973
Peltola first75,799

Begich wins. Palin is eliminated; Begich defeats Peltola 81,859 to 79,451. The absent ballots rank Begich above Peltola, so their preferred finalist wins only when they do not participate.

Source-pattern guard: the published Palin › Begich › Peltola group contains 34,049 ballots, so the control can never remove more ballots than exist.

Independent route, fixed anchors, broken instrument

The check has teeth

The moving page expands the profile into ballot rankings and simulates elimination. This panel separately adds named published columns. It then checks outside anchors and feeds the checker a profile known to be wrong.

Pinned profile SHA-256

a7b9842897f81faa150aece1e71fa5bc24860c8fe4889376d929575208ffdc32

The poison control changes the 27,053 ballot Begich › Palin › Peltola column to 27,054. A checker that accepts it is broken. This checker rejects both its total and Begich first-round count.

Running checks…

What the apparatus does not know

The record has edges

Scanned ballots are not every physical ballot

Alaska says its CVR contains ballots that were scanned and excludes ballots counted only by hand. It shows write-in rankings but not the names written. This page uses the paper's published reduction of that CVR, not the large raw JSON export and not a census of every physical ballot cast.

Parsing choices are substantive

Alaska excludes unresolved write-ins for this tabulation, stops at an overvote and lower ranks, and makes a ballot inactive after two consecutive skipped ranks. The paper says it disregards write-ins and groups complete with incomplete rankings when that does not change the instant-runoff winner. Its last six columns may differ slightly under another faithful parsing. That is why the official round report, not an asserted raw-CVR replay, is the anchor here.

The exact tie is a refusal, not a guess

At exactly 5,163 altered ballots, Begich and Palin each have 53,810 first choices. Alaska resolves a necessary tie by lot. The instruments refuse to pick an elimination at that setting, and the verifier counts that refusal as a tested result.

These witnesses are counterfactual, not behavioral

Moving or removing 6,000 records proves sensitivity of this fixed profile under this rule. It says nothing about campaign strategy, voter persuasion, turnout response, or the result of a new election. A Condorcet winner is also a defined pairwise fact, not a claim of unique political legitimacy and not a general verdict on ranked choice voting.

Alaska Division of Elections

CVR scope, exclusions, and ballot-handling definitions.

Official RCV detailed report

Published round totals, transfers, exhausted ballots, and overvotes.

Graham-Squire and McCune, arXiv:2209.04764v3

Nine-pattern reduction, pairwise counts, monotonicity witness, and no-show witness.

research/the-six-thousand-votes-that-hurt/README.md

Data digest, independent algorithms, uncertainties, and rerun command.