✶ Precinct protocols · 2016 · 2021 · 2024 · 2026
Reading
the numbers.
Every Russian polling station publishes its count. Laid side by side, nearly 90,000 precinct results form shapes that statisticians have studied for two decades. This page shows those shapes and what researchers say about them. It describes; it does not judge any single result, and we make no fraud estimate of our own.
a · Turnout
How many
came to vote.
Each bar counts the ballots cast at precincts whose turnout fell in that bin. Where turnout varies for ordinary reasons, the result is a single hump. Russia's has a second feature: a long shelf towards 100%, and, at the finest resolution, spikes at round numbers (70.0%, 75.0%, 80.0%…).
Kobak, Shpilkin & Pshenichnikov (2016) showed that Russian precincts report whole-number turnouts and results far more often than chance would allow, and read this as a fingerprint of results being written to targets. Switch to 0.1 pt bins to see the spikes.
Precinct turnout →
b · The leader's share
How the winner's
vote spreads.
Votes for United Russia in 2016, by the share of valid votes United Russia won at each precinct.
Votes for United Russia in 2021, by the share of valid votes United Russia won at each precinct.
Votes for Vladimir Putin in 2024, by the share of valid votes Vladimir Putin won at each precinct.
Votes for United Russia in 2026, by the share of valid votes United Russia won at each precinct.
The same integer effect appears here: peaks at round shares. The bar at 100% counts precincts where every valid vote went to the leader.
Leader's share of valid votes at the precinct →
c · Turnout × share
The
comet.
Put both on one chart: turnout across, the leader's share up, each cell shaded by the ballots cast at precincts that land there. The colour scale is logarithmic so thin tails stay visible.
Klimek and colleagues (2012) compared such "fingerprints" across countries. In most elections they form one compact cluster: turnout and the winner's share vary, but not together. In the recent Russian elections they studied, the cluster trails off in a tail towards 100% turnout and 100% for the winner, which they found "well-explained by systematic ballot stuffing" — the "comet" that Kobak, Shpilkin and Pshenichnikov also describe.
A tail is not proof by itself. Precincts in closed institutions, small villages or some republics can legitimately vote with high turnout and near-unanimity. Researchers weigh the size and shape of the tail, not its existence.
Hover or tap a cell
c · The comet in relief
The same shape,
with height.
The heatmap above as a landscape: the main cluster is the mountain, the tail towards the far right corner its ridge. The wall behind it is the turnout histogram at 0.1-point resolution, where the round-number spikes stand up like fence posts. Switch years in the bar above; drag to turn it.
This 3D view needs WebGL. The heatmap and histograms above show the same data.
d · Four descriptive numbers
Summaries,
not verdicts.
- Covered 92,804 precincts and 52,443,398 ballots · State Duma (party list).
- At ≥ 90% turnout 12.89% of all ballots were cast at precincts reporting 90% turnout or more.
- Leader, all vs low-turnout 55.21% vs 42.53% United Russia's share of valid votes at all covered precincts, and at precincts under 50% turnout. If turnout and preference were unrelated, these would be close.
- Integer turnouts 1.29× Precincts with a whole-number turnout (±0.05 pt, from 50% to 95%), compared with neighbouring non-integer values. 1.00× would mean no excess.
- Covered 92,116 precincts and 55,550,998 ballots · State Duma (party list).
- At ≥ 90% turnout 13.66% of all ballots were cast at precincts reporting 90% turnout or more.
- Leader, all vs low-turnout 50.83% vs 34.71% United Russia's share of valid votes at all covered precincts, and at precincts under 50% turnout. If turnout and preference were unrelated, these would be close.
- Integer turnouts 1.20× Precincts with a whole-number turnout (±0.05 pt, from 50% to 95%), compared with neighbouring non-integer values. 1.00× would mean no excess.
- Covered 88,530 precincts and 74,974,867 ballots · President.
- At ≥ 90% turnout 28.28% of all ballots were cast at precincts reporting 90% turnout or more.
- Leader, all vs low-turnout 88.19% vs 81.95% Vladimir Putin's share of valid votes at all covered precincts, and at precincts under 50% turnout. If turnout and preference were unrelated, these would be close.
- Integer turnouts 1.28× Precincts with a whole-number turnout (±0.05 pt, from 50% to 95%), compared with neighbouring non-integer values. 1.00× would mean no excess.
- Covered 86,310 precincts and 56,635,897 ballots · State Duma (party list).
- At ≥ 90% turnout 10.63% of all ballots were cast at precincts reporting 90% turnout or more.
- Leader, all vs low-turnout 58.75% vs 42.62% United Russia's share of valid votes at all covered precincts, and at precincts under 50% turnout. If turnout and preference were unrelated, these would be close.
- Integer turnouts 1.25× Precincts with a whole-number turnout (±0.05 pt, from 50% to 95%), compared with neighbouring non-integer values. 1.00× would mean no excess.
Computed by us from the precinct counts, following the definitions in Kobak et al. (2016). Shares here are of valid votes at covered precincts, so they differ slightly from the CEC's national figures, which use all ballots and include remote e-voting. A gap between the two leader shares or an integer excess has innocent contributors as well as suspicious ones; the numbers describe the distribution and are not an estimate of fraud.
e · Other people's estimates
What analysts
have estimated.
Sergey Shpilkin's method assumes that in precincts with ordinary turnout the leader's share is roughly stable, and treats votes above that pattern at high-turnout precincts as anomalous. These are his and others' published figures, attributed; we have not reproduced them.
| Election | Official | Estimate | By |
|---|---|---|---|
| 2021 · State Duma | United Russia 49.8%, turnout 51.7% | United Russia ≈31–33%, turnout ≈38%; about 14 million votes added | Sergey Shpilkin (statistician), via RFE/RL ↗ |
| 2024 · President | Putin 87.28%, turnout 77.49% | At least 22 million of Putin's polling-station votes anomalous | Novaya Gazeta Europe, Shpilkin's method ↗ |
| 2026 · State Duma | United Russia 58.2% of ballots | United Russia's 'clean' share ≈34% (excluding Moscow and e-voting) | Novaya Gazeta Europe (Alesya Sokolova), Shpilkin's method ↗ |
These are outside analysts' estimates. The legal result is the CEC's protocol, which is what the rest of this site reports as official.
f · 2026
How the 2026
vote was run.
- No OSCE observers Russia did not invite OSCE/ODIHR, which called it 'a serious breach of the commitments it has undertaken'. Source ↗
- The anti-war party removed On 10 Aug 2026 the Supreme Court cancelled Yabloko's federal list on a suit by Rodina; the appeal failed on 17 Aug. Its district candidates stayed on the ballot and won no seats. Source ↗
- Independent monitoring gone Golos, which observed Russian elections from 2000, ceased work on 8 Jul 2025 after co-chair Grigory Melkonyants was convicted of running an 'undesirable organisation'. Source ↗
- Foreign agents barred Since May 2024 anyone designated a 'foreign agent' must shed the status before registering as a candidate; designated people may not campaign, donate or sit on commissions. Source ↗
- Observers turned away More than 1,300 observers from the CPRF and Yabloko were not admitted to polling stations on 18 Sep, according to Echo. Source ↗
- Voting in occupied Ukraine The vote was held in the occupied Donetsk, Luhansk, Zaporizhzhia and Kherson oblasts and in Crimea. The EU called the results there inapplicable; Australia, Canada, New Zealand, Norway and the UK called them a sham. Source ↗
Remote electronic voting (DEG)
7.1Mvoters, our sum
Voters in about 32 regions could vote online through the federal system: some 3.65 million did, according to RBC. Moscow ran its own system, used by about 3.45 million.
Online votes have no paper ballot to recount and no precinct commission whose count observers can watch; they are tallied by the system's operators and reported as separate electronic "precincts". Precinct-based methods like those above cannot be applied to them, which is why our charts leave them out and why the 2026 estimate in the table excludes Moscow and e-voting.
RBC, 18 Sep 2026 ↗g · About the data
What these
charts leave out.
- Source. Precinct (UIK) protocols collected in Dmitry Kobak's dkobak/elections repository: scraped from the CEC by Sergey Shpilkin (2000–2021), Ivan Shukshin (2024) and Petr Zhizhin (2026).
- Licence. The repository states no licence, so we publish only aggregated histograms, never precinct rows.
- Excluded. Remote e-voting pseudo-precincts, and precincts with fewer than 100 registered voters. For 2024 and 2026 there are no precinct data for the Donetsk, Luhansk, Zaporizhzhia and Kherson oblasts annexed in 2022.
- Coverage in 2026. Our charts hold 56,635,897 of the 67,202,227 ballots in the CEC total (84.3%). For scale, the gap of 10,566,330 compares with 7,100,000 remote e-voters (RBC) and 3,466,372 ballots in the four oblasts annexed in 2022 (CEC district protocols). Our arithmetic.
- Basis. Party-list ballots for the Duma; the presidential ballot for 2024. Turnout is ballots in boxes ÷ registered voters, capped at 100%.