The count from start to finish
Lula and Flávio in valid votes, from the first ballot boxes on: what the TSE counted, IPSE's projection of the final result, and the expected path of the count to 100%.
Runoff map
Choose the scale (municipalities or states) and the indicator. Hover over (or tap) to see the numbers; click a municipality to open its profile, or a state to open its count.
By region
Ballot boxes and valid votes counted in each region, with IPSE's projection of the final result.
By state
Result in each state. The border color shows who won.
Where the runoff votes came from
Who voted for whom, from the first round to the runoff. Hover over the bands to see how many voters took each path. The diagram and the waterfalls follow the state chosen in the selector below.
For each candidate: first-round votes, what he lost, what he gained and from whom, up to the runoff total. In millions of votes; for Brazil, the 90% margin in parentheses.
Todos os cargos
Presidente, governador, senado e deputados, no 1º turno, e o 2º turno de presidente e de governador, por estado, município e bairro.
Forecasts and polls against the count
What pollsters measured in the final week, what aggregators and IPSE forecast before the vote, and where the projection stands, for Lula (red) and Flávio (blue), in valid votes. The vertical lines are IPSE's projection with the count; once counting ends, they are the result. The “dif.” column shows how far each estimate was above or below the projection for Lula (Flávio's is the same with the opposite sign).
What IPSE published before the vote
The final predictive study (Oct 24) and the election-morning addendum (Oct 25, 8 am), in Portuguese, at ipse.global/publications. The count is the test of the model, and the comparison will be made in public.
Open the full study (PDF, 105 slides, in Portuguese) ↗Open the Oct 25, 8 am addendum (PDF) ↗
How it works
The data
During the count, every minute, we read from the TSE the votes for Lula and Flávio in each state and abroad, and how many sections of each municipality had been counted. The baseline is the 2026 first round, summed section by section from the TSE's open data (499,000 sections), and the 2022 runoff. The start of the series, from 5:07 pm to 8:06 pm, when this panel went live, reproduces the minute-by-minute record of the TSE count published on Estadão's projection panel.
Projecting the result
For each municipality we know how Lula and Flávio did in the first round. In the 150 largest cities, about 40% of the electorate, we measured every minute the city's own change between rounds, on a logistic scale, and applied it to what was still to be counted there. Elsewhere we used the change measured in the rest of the state, and a state with almost nothing counted used the national change. The number of votes still to come follows each place's electorate, adjusted by the turnout already observed. Late-counting regions therefore enter the projection with the vote they showed in the first round, not with the average of what had been counted so far.
Uncertainty and the chance of winning
Uncertainty comes from three sources, all about what is still to be counted: (1) the error in the change estimated for each state, measured by the actual spread of changes across large cities tonight (– point); (2) the order of sections within each large city, which does not count centre and periphery at the same time (2 points on the remaining share of each); and (3) an error common to the whole country, from the uneven redistribution of the roughly 8% of votes for other first-round candidates (1 point). Together they give the deviation σ. The 95% range is the projection ± 1.96σ, and the chance of winning is Φ((projection − 50) / σ). Tested on the 2022 runoff with simulated counting orders, the method erred by less than 0.15 point.
How the projection performed
At 8:06 pm, with 27% of boxes counted and Lula still behind in the count (49.5%), the projection gave him 51.4%, with a 95% range of 50.1% to 52.7% that already contained the final result. At 45% counted, the central value was within 0.1 point of the result and, at 57%, it read 51.78%, the final figure. The range contained the result in every update of the night. At 8:34 pm, with 57.7% counted, it switched to the spread measured across cities, which is why it widens at that point of the chart.
The path to 100%
The dashed line shows how the count should evolve to the end, assuming the slowest states stay slow. It explains why the count can change leader in the middle of the night, since the Northeast, São Paulo and Rio usually report late. This time the order was unusual, with the North and Paraíba early and the South later, and the Lula and Flávio curves crossed more than once.
Where the votes came from
The diagram and the waterfalls estimate who voted for whom between rounds by ecological inference. Using the results of all 5,716 municipalities and abroad in both rounds, we fit the transition matrix with Goodman's method under constraints (each row sums to 100% and no flow is negative), weighted by the electorate, state by state; states with few municipalities are pooled with their region. The 90% margin comes from 100 resamples of municipalities and measures sampling variation, not aggregation bias. The flow from blank and spoiled votes to Lula is the component most exposed to that bias.
The municipal map
For each municipality, the transition matrix is fitted on its 80 nearest municipalities, with weights that fall with distance and rise with the electorate, and applied to that municipality's own first round. The origin of the votes thus varies across the territory, showing where Lula gained most from abstention and where most from blank and spoiled votes. The electorate's profile and first-round abstention by sex and age come from the TSE's open data. Added up, the municipal estimates do not match the national one, and the gap between the two is part of the method's uncertainty.
Box by box in the territories
In the favelas, quilombos and Indigenous lands tab, the runoff comes from the ballot-box reports published by the TSE, read one by one. Each section is linked to a territory through its polling place, when it lies inside the polygon or within 50 metres, with the option of widening to 150 metres (favelas) or 300 metres (quilombos and Indigenous lands). Territories are those of the IBGE 2022 Census and, in the city of Rio, the city's 2022 favela boundaries from the Instituto Pereira Passos. A box shows who votes at that place, which is not exactly who lives in the territory. Open the tab.