Polls were due to close at 17:00 Brasília time. Minutes after that, Justice Cristiano Zanin of the Supreme Federal Court (STF) granted the PT's request to keep voting open until 19:00 nationwide, the first extension in Brazilian history. Lula's campaign cited bus shortages; Flávio's had asked earlier for two extra hours in 17 rain-hit cities. By then most sections had shut, and closed electronic ballot boxes cannot be reopened, so an hour later the order was narrowed to sections still open. Some regional courts said they had kept the 17:00 close. The ballot-box logs show only 34,330 votes after 17:00, 29,001 of them in the first five minutes, and none at all in 94% of sections. Lula won by 4,430,191 votes.
2026: the TSE's final results files for the runoff (resultados.tse.jus.br, election 6258), one per state, per city and for voters abroad: electorate, turnout, valid, blank and spoiled votes, and each candidate's votes. 2018 and 2022: the TSE's open-data file detalhe_votacao_munzona for each year, summed over every municipality and zone for the presidential runoff (round 2). Those sums reproduce the official national turnout exactly: 78.70% in 2018, 79.42% in 2022 and 80.78% in 2026.
Each state keeps its 2026 electorate. Its turnout is set to the rate it had in the 2018 or 2022 runoff, and every 2026 figure there (Lula, Flávio, blank, spoiled) is scaled by the same factor. So a state that turned out more in 2026 loses voters, and those voters are assumed to have split exactly like everyone else in that state. Voters abroad are treated as a 28th state. This only moves the national result through where turnout changed, not who turned out inside a state. Data by city would show more, but the question was asked by state.
The TSE's open-data portal has not yet published the runoff's ballot-box bulletins, but the ballot-box logs it publishes were analysed the day after (Setup News for CNN Brasil, Agora RN, G1, Jovem Pan). They show 34,330 votes after 17:00 (35,521 in G1's count), 29,001 of them between 17:00 and 17:05, and no vote at all after 17:00 in 468,043 of the 497,897 sections, so only 29,854 sections (6%) took any. Only 1,667 sections took their last vote after 18:00. Which sections missed the extension: every place reported closed counts at its reported share: the states whose courts kept 17:00 in full, São Paulo at 80%, Paraná at 87.4%, and the reported cities outside those states in full. What two more hours brought: the 29,854 sections with any late vote are taken as the ones that stayed open. That overstates the late votes per open section, because closed sections also served their 17:00 queue. The central estimate counts only the 5,329 votes cast after 17:05, since people already queuing at 17:00 may vote at any section, open or closed: about 0.18 votes per section. The high estimate counts all 34,330, as if every one came from the extension: about 1.15 per section. Each closed section in the reported places then gets that many extra votes. Logs by state: the logged late votes for ten states (Jovem Pan) give a direct check. States whose courts kept 17:00 averaged 0.058 per section, mostly the queue, against 0.090 where open sections ran on to 19:00, so the extension's measurable effect is about 0.03 votes per section. Both estimates above are well beyond it. Until this search the page assumed one section in five had stayed open; the logs showed the real share was far smaller. Turnout check: if the closures had kept many people from voting, the reported places should show weaker turnout growth from round 1 (4 October, same 17:00 close everywhere) to the runoff than the rest of Brazil.
No official list of closed sections has been published. The places on this page are the ones reported on the night. Kept the 17:00 close: Santa Catarina, Rio Grande do Sul and Ceará (Gazeta do Povo, which also says the Federal District closed nearly everything); Jornal Razão lists Santa Catarina, the Federal District, Rio Grande do Sul, Minas Gerais, Mato Grosso do Sul, Amazonas and Rio Grande do Norte. The two disagree on Rio Grande do Norte: Gazeta says its open sections ran to 19:00. Mostly shut: TRE-SP said about 80% of São Paulo's ballot boxes had closed; Gazeta says 87.4% of Paraná's voting places had. Those states count 80% and 87.4% of their electorate. Cities: the 17 in Flávio's rain request (as listed by ND+), plus Cuiabá and Goioerê (Jornal Razão), Porto Alegre (PUC-RS rooms shut, TRE-RS), Belo Horizonte (a closed section reported after the order), Rio de Janeiro (Colégio Pensi in Flamengo and all 28 boxes at Ciep Tancredo Neves in Catete shut at 17:00; two schools in Barra da Tijuca; Agência Brasil, Brasil 247), Duque de Caxias (Brasil 247), Teresina (poll workers had no word of the extension, Agência Brasil) and Campinas (three schools shut, Agência Brasil). Agência Brasil says Flávio's request also covered Minas Gerais and Rio Grande do Sul; ND+'s list of 17 cities has none there. A city takes its state's closed share where the state's court gave one (São Paulo 80%, Paraná 87.4%) and counts in full otherwise.
The slider adds turnout, in points of the electorate, to the closed part of each place, and stops at 100% turnout. Three reference points: What the extension added is the 34,330 votes logged after 17:00, spread over the 6% of sections that took any vote after 17:00, which gives about 0.36 points. That is generous, since 29,001 of those votes came in the first five minutes from people already queuing, and people in the queue at 17:00 may vote at every section, closed or not. Round 1's 17:00 queue is the 3,724,999 people who voted after 17:00 on 4 October, about 2.35 points: a whole extra wave of voters at once. Enough to flip it finds the smallest rise that would have put Flávio ahead if every extra voter had backed him.
All for Flávio: every extra voter casts a valid vote for Flávio Bolsonaro. This is the most he could possibly gain. Like the area: the extra voters split exactly as that place voted in 2026, blank and spoiled votes included: a city by its own result, a state by the state's.