Alaska 2026 ranked-choice explorer
Explore ranked choice election preliminary data from the Alaska Division of Elections.
Preliminary and unofficial. Counting isn’t finished, and this page’s count isn’t guaranteed to match the Division of Elections’ or to be correct. Not affiliated with the State of Alaska or the Division of Elections. Read the caveats.
The sections below count real ballots from the CVR and nothing else. The one exception is the simulated projection near the end of the page, which adds copied ballots to estimate the ballots still to be counted.
CVR ballots only
Ranked-choice count of CVR ballots
Rounds run left to right, and each candidate keeps a row. Ribbon thickness is ballots, and percentages are of that round’s continuing ballots. Ribbons stay with a candidate, or split off when their candidate is eliminated. Gray is ballots that stopped counting. Tap or hover for numbers. Blanks and first-round overvotes aren’t drawn; they’re in the table.
Preliminary. CVR ballots only, so it isn’t the official count.
Blanks never counted for any candidate: they have no rankings, skip the first two ranks, or rank only write-ins. Overvotes stopped counting at a rank with more than one candidate marked, even if those candidates had been eliminated. Exhausted ballots counted for a candidate who was eliminated, then had nowhere to go: every candidate they ranked next was out too, or they skipped two ranks in a row.
Blanks have no mark on this measure, and overvotes marked both yes and no. Neither counts as a vote.
CVR ballots only
By house district
The map and the list both show the contest’s two finalists head to head: the two candidates left after statewide RCV tabulation leaves two candidates. Each ballot goes to whichever candidate it ranks higher, as in the final round, so these are the final round’s numbers split by district, not first choices. Districts don’t get their own eliminations, so a district’s own top two could differ. On a ballot measure, they show yes against no. A statewide race split by district is informational, and many rural ballots aren’t in the CVR at all: hollow dots mark districts with under 90% of their counted ballots in the CVR.
Hover or tap a district for its numbers, and tap it again to clear it. District lines: from the Alaska Division of Elections’ results map, with coordinates rounded to about 1 km, and clipped to the U.S. Census Bureau’s outline of Alaska.
The outlines on the state map mark the insets.
Every district, ranked
“Ballots left” are ballots these numbers don’t include yet, rounded to the nearest 10: those counted by hand, which the CVR leaves out, plus an estimate of early and absentee ballots received but not counted. The Division only publishes that second number statewide, so this page splits it using each district’s share of early and absentee ballots in the 2024 CVR. They also include an estimate of questioned ballots, described under “How the simulation works” below. They leave out ballots still to arrive and precincts yet to report.
Every district as a table, with CVR coverage
Ballots here are all ballots in a district, by the Division’s district labels, not just those in the selected contest. “Summary” is ballots the Division’s summary counted. The 90% line for hollow dots uses these numbers.
CVR ballots and the Division's summary
The CVR against the Division's summary
The summary counts first choices on all ballots counted so far, including ballots counted by hand. The CVR only has scanned ballots. If they disagree, trust the summary over anything on this page.
CVR ballots only
Explore ballot rankings
Use this tool to explore the rankings recorded on ballots. Pick a first choice candidate to see where those ballots would go next if that candidate were eliminated.
How many candidates each ballot ranked
Simulated
Simulated projection
The sections above count real ballots, but we can also simulate how the remaining ballots could look and predict the race outcome. Each simulation fills in the ballots still to be counted with copies of real ballots from the CVR, then runs the same ranked-choice count as above. The table says how often each candidate wins a simulation. That’s a share of simulations under the assumptions below, not a chance of winning: different assumptions give different numbers. Simulations always cover the whole contest, whatever district is selected.
One dot is one simulation, placed at the winner’s share of the continuing ballots in the round that decided it, and shifted up or down only so dots don’t overlap. Round 1 and round 3 aren’t the same kind of win: a round 1 win is a majority of first choices, and a round 3 win is a majority of the ballots still counting after eliminations.
How the simulation works, and what it assumes
Every simulated ballot is a real ranking copied from the CVR, and every simulation is counted by the same tabulator as the rest of this page. Races with one candidate aren’t simulated. Each district gets:
- CVR ballots, as they are.
- Ballots counted by hand. The Division’s summary gives their first choices but not their later rankings. For each first choice the summary has beyond the CVR, the simulation copies a ballot with that same first choice from the district’s Election Day ballots, then its other ballots. If those come to fewer than 30, the rest come from the same first choice elsewhere in the race.
- Early and absentee ballots not yet counted, drawn from the district’s early and absentee ballots. The number per district is the estimate described under “Ballots left” above.
- Questioned ballots, drawn from the district’s Election Day and hand-counted ballots. The Division hasn’t given a full count of them yet, so each district gets its number from the 2024 CVR, scaled by how 2026’s Election Day turnout compares. That’s an assumption. Many questioned voters vote outside their own district, and then only their statewide races count: in 2024, 11,031 questioned ballots had a U.S. Representative race and 4,041 had a State House race. State Senate races use the State House numbers, because 2024’s Senate races were in other districts.
Where few ballots are known. Some rural districts have almost nothing in the CVR because their ballots are counted by hand: House District 37 has 82 CVR ballots out of about 2,800 counted so far. The simulation does four things about this. If a district has fewer than 300 ballots to draw early, absentee or questioned ballots from, it fills the pool up with the district’s other ballots, including the hand-counted ones, whose first choices are published. Each simulation resamples all of these ballots, so a small pool varies more. The number of uncounted and questioned ballots also varies between simulations, as far as district turnout varied from 2024 to 2026. A larger lean applies to local races, described next. None of this can show what’s unknown about rankings: in House District 37, where each candidate’s later rankings rest on 15 to 24 ballots, the transfers are close to a guess. The page warns when under half of a race’s simulated ballots have known rankings.
The lean is a way to say that uncounted ballots might not vote like the ballots they’re copied from. Each simulation picks a number for each candidate, the same in every district of the race, from a normal distribution, and multiplies that candidate’s share of the uncounted and questioned ballots by e raised to it. Blanks and overvotes keep their share. Hand-counted ballots don’t lean, because their first choices are known. A district with no ballots for a candidate can’t lean to them.
The size of the lean comes from testing the same method on 2024. Copying each district’s early ballots to predict its absentee ballots was off by 7.1 points for President and 6.8 for U.S. Representative. A standard deviation of 0.2 moves a 50-50 race by about 7 points, so statewide races and ballot measures use 0.2. In 2024’s State House races the same test was off by about 10 points once sampling noise is taken out, so House and Senate races use 0.28. The win counts depend on these numbers. Those two statewide misses both ran toward the Democrat, and the lean doesn’t assume that repeats. Predicting questioned ballots from Election Day ballots missed toward the Republican, by 7.0 and 3.9 points.
It leaves out ballots yet to arrive and precincts yet to report. Districts are a fixed set, so the simulation doesn’t resample across them.
Read this before quoting anything here
- These results are preliminary. The Division of Elections keeps counting absentee, early and questioned ballots after election night, and it won’t publish any ranked-choice results until November 18, 2026. Numbers here will change, and a candidate leading here may not lead once all votes are counted.
- No guarantee. I wrote this tabulation myself. It isn’t guaranteed to match the Division’s, or to be correct at all. The Division’s count is the only official one.
- No affiliation. This page isn’t affiliated with, endorsed by or checked by the State of Alaska or the Division of Elections. Please don’t contact them about it. Send questions to will@dummthings.dev.
- This page leaves out hand-counted ballots. The Cast Vote Record published by the Division of Elections does not include hand-counted ballots, and those ballots are concentrated in rural areas. In this update, on the CVR alone Nick Begich has a first-round majority for U.S. Representative, while the Division’s summary has him under 50%, and Dan S. Sullivan leads the CVR’s first round for U.S. Senate while Mary Peltola leads the summary’s.
- Individual district numbers do not determine statewide races. A statewide race broken out by house district is informational: it’s not a district result, and Alaska doesn’t count statewide races by district. District coverage ranges from everything to nothing (see the table under the district list), so hollow dots mean the CVR is missing much of a district. Early and absentee ballots are grouped under the districts the Division’s reports give them, which can differ from where voters live: compare the Early column between districts before reading much into one.
- Filtered views aren’t official in any sense. Filtering by district re-runs the count on just those ballots, with the same rules. No such count exists officially, and small groups can swing widely.
What the Division says about the cast vote record
From the Division’s 2026 results page:
The Cast Vote Record (CVR) contains the votes and rankings on all the election ballots that were scanned. It does not include ballots that were only counted by hand. It also does not indicate the rankings that each individual write-in candidate received. The CVR shows how voters ranked write-in candidates, but it does not show the names the voters wrote in.
The CVR is a JSON file, used by the ranked-choice software to tabulate the results. The CVR is not tabulation, it is a record of ballots. The division is providing the CVR to increase transparency and provide voters the data necessary to confirm the results. The division cannot help voters access or analyze the data.
The rules this page applies
The rules are AS 15.15.350, 6 AAC 25.071, and the tabulation options printed on the Division’s round-by-round reports:
- If a candidate has more than half the continuing ballots in round 1, the count stops there. Otherwise the candidate with the fewest votes is eliminated, one per round, until two remain. The Division’s option is “Declare by threshold: False”, so a majority in a later round doesn’t stop the count. It does decide it: continuing ballots only shrink and that candidate’s pile only grows, so no one can catch up. This page marks that round.
- A ballot goes to its highest-ranked continuing candidate. Percentages are of continuing ballots.
- A ballot stops counting at an overvote (two candidates at one rank), even if those candidates have been eliminated, or at two or more skipped ranks in a row. A ballot with no rankings, or whose first two ranks are skipped, is a blank.
- Write-ins are passed over, and don’t count as a skipped rank. Under 6 AAC 25.071 write-in rankings only count if the write-ins together are first, or second within half a percent. The CVR has no write-in names, so this page can’t resolve them, and it warns you in a contest where the rule could apply.
- A tie for last is decided by lot under AS 15.15.350(e), which only the Division can do. The page flags a tie and eliminates the candidate listed later as a placeholder.
Ballot measures aren’t ranked, so this page just adds up yes and no on the CVR’s ballots.
How I verified these tabulation rules
The same code that runs on this page reproduces all 9 of the Division’s certified round-by-round reports from the 2024 general election, from the 2024 CVR. That should not be taken as proof that 2026 will match.
The tabulator and the scripts that prepare the data are plain JavaScript and
Python, so you can read them. The prepared data is at /data/alaska-rcv-2026/, and the simulations
are in its projection.json, regenerated from a seeded script.
Sources
- Ballots: the Division’s cast vote record
- First choices and precincts reporting: the Division’s election summary and its results page
- District boundaries: the shapes behind the Division’s results map, whose coordinates are rounded to about 1 km, so edges between districts are approximate
- Questioned ballots, early and absentee shares, and the simulation’s lean: the Division’s 2024 general election cast vote record
- Alaska’s outline: the U.S. Census Bureau’s 1:500,000 state boundaries, 2024
- Charts and map: d3
Prepared by Will Dumm. Questions and corrections are welcome at will@dummthings.dev.