In one sentence: since the pro-Lula rallies of Oct 11 to 13, the pooled poll margin went from −4.2 to −2.0 points in valid votes, and the 7 pollsters with a wave before and after, same method, together record +2.9 points; it is not possible to separate the effect of the streets from the end of the 6x1 schedule and the natural wear on the first-round winner, but the channel through which the streets can decide the election is turnout, not conversion.
This is the weightiest theme, because it acts on who turns out and on voters without a side. In the Oct 16–18 BTG/Nexus, 54% think Flávio will win and 39% Lula (it was 59% and 36% on Oct 10–12); 13% of Lula's own voters are betting on Flávio (it was 19%). In PoderData, the expectation of a Lula win rose from 37% (Oct 5–7) to 42% (Oct 13–15), stood at 41% on Oct 16–18 and returned to 42% on Oct 19–21; Flávio's went from 54% to 51%, 52% and 52%. Expectations improved for Lula after the first week but stopped rising, and the gap is still 10 to 15 points.
The runoff math rests on 9.3 million votes from the eliminated candidates and 33.5 million absentees. In the Oct 16–18 PoderData, Flávio takes 56% of Cury's voters (Lula, 31%), 36% of Renan's (29%) and 46% of Caiado's (19%); Lula leads among those who voted blank or spoiled (33% to 21%), but among those who did not vote the lead shifted to Flávio (41% to 31%; it was 43% to 30% for Lula on Oct 13–15). The BTG/Nexus from the same dates measures the opposite among absentees (Lula 49%, Flávio 44%): these are small subsamples and the signal is uncertain. In the model, which starts from the counted vote by state, turnout still matters more for Lula: in the model, 2 points less abstention in the Northeast alone raise his chance from 17% to 21%; 3 points more, to 11%.
Passed by the Senate 75 to 1 on Oct 14 and promulgated on Oct 16 (Amendment 141). Lula has made it a banner since the first round; Flávio voted in favor after pressure and promised to revisit the issue if elected. The issue stopped being "for or against" and became "who guarantees it." In the Oct 16–18 BTG/Nexus, the first to ask after promulgation, 66% approve of ending the 6x1 schedule and 26% disapprove; among Flávio's runoff voters, 41% approve and 49% disapprove. In polls with fieldwork entirely after promulgation, the margin did not move toward Lula consistently: in PoderData it went from a tie to Flávio +3.2 in valid votes, in BTG/Nexus from Flávio +3.2 (Oct 10–12) to +0.6 and in Futura from Flávio +1.6 to +0.8. The issue is popular but has not yet visibly turned into votes.
The left returned to the streets on a scale not seen in years: Paulista on Oct 11 (≈ 261,000 by occupied area), Copacabana on Oct 12, the Esplanada on Oct 13, more than 70 cities on the first day and a new round on Oct 16 and 17, under the slogan "It's time for the comeback." Online, the "Exército do Lula" channel (3.3 million followers) became a fight at the TSE. Flávio responded with motorcades and, on Oct 18, with the "Bloco do 22" on Paulista, alongside Tarcísio and Ricardo Nunes: 71,900 people at the peak by USP's count, more than three times what the PL expected. The streets are no longer the left's alone. The effect on the polls is measured below.
Flávio began advocating an end to reelection with longer terms, constitutional changes to "redemocratize" the country, judicial reform with impeachment of STF justices (Alexandre de Moraes first) and suspension of the tax reform. According to Folha, this caused unease in moderate sectors. It matters above all among those who reject both: in the Oct 16–18 BTG/Nexus, the "anti-Lula and anti-Bolsonaro" (7% of the sample) vote 48% for Lula and 36% for Flávio when they choose (it was 50% and 37%); among the non-polarized, who are 21%, Flávio pulled ahead (41% to 40%; it was Lula 44% to 37%).
Lula is using the office: he zeroed federal taxes on fuel, launched a package for micro-entrepreneurs and promised free city buses on weekends nationwide (Oct 15). Approval did not firm up: in PoderData, 46% approved and 47% disapproved of Lula's job on Oct 13–15 and, on Oct 16–18, 44% and 50%, back to the Oct 10–12 level, and on Oct 19–21, 46% and 49%; in the Oct 16–18 BTG/Nexus, 47% approve and 49% disapprove; in Alfa's Oct 15–19 poll (in person), 46% and 50%; in the Oct 17–20 Quaest, 46% and 47% (it was 45% and 48% on Oct 10–13), with rejection of Lula falling from 53% to 47%. Weighing against Lula are the still-negative assessment of the government and Flávio's alliances with PP, Republicanos and Caiado.
Each dot is a national poll taken after Oct 4 (57 through Oct 24), on the date its fieldwork ended; above zero, Lula ahead. The bars are the pooled margin, without house-effect correction, of polls ending before, during and after the first wave of pro-Lula rallies (Oct 11 to 13), with a 95% interval. The model's adjusted trend is in "The race over time."
| Date | Theme | What happened | Expected effect |
|---|---|---|---|
| 4/10 | Result | Flávio wins the first round with 47.03% of valid votes to Lula's 45.16%; 33.5 million voters did not vote. source | tends to favor Flávio |
| 6/10 | Alliances | Flávio receives the backing of PP, Republicanos and Ronaldo Caiado; the PSD sides with neither. source | tends to favor Flávio |
| 7/10 | Lula's campaign | Lula calls his own campaign "the most disorganized" since 1989; allies trade accusations over the first-round defeat. source | tends to favor Flávio |
| 8/10 | Rules of the game | Flávio begins advocating an end to reelection with longer terms, constitutional changes to the political system and the judiciary, and suspension of the tax reform. source | tends to favor Lula |
| 9/10 | TV | Free electoral airtime for the runoff begins. source | neutral |
| 10/10 | Streets | Flávio holds a motorcade in São Gonçalo (RJ) and asks supporters to bring those who did not vote: "get one more little vote". source | tends to favor Flávio |
| 11/10 | Streets | Pro-Lula rally on Av. Paulista, with Lula, fills 7 blocks: about 261,000 people based on the area's capacity (Poder360); 1 million according to the campaign. There were rallies in more than 70 cities. source | tends to favor Lula |
| 11/10 | Electoral courts | TSE president Nunes Marques suspends the "Exército do Lula" channel (3.3 million followers) at Flávio's request; the electoral inspector general reinstates it the same day; on Oct 13, a new 48-hour suspension and the campaign registers the channel. source | neutral |
| 12/10 | Streets | Pro-Lula rally in Copacabana (RJ), no crowd estimate. source | tends to favor Lula |
| 13/10 | Streets | Students go to the Esplanada dos Ministérios; the campaign announces rallies in BH, Salvador, Belém, Fortaleza and Recife under the slogan "It's time for the comeback" to reduce abstention. source | tends to favor Lula |
| 14/10 | Labor | The Senate approves the end of the 6x1 schedule 75 to 1, in both rounds of voting. Flávio votes in favor, after pressure, and promises to revisit the issue if elected. source | tends to favor Lula |
| 15/10 | Rules of the game | Impeachment of STF justices, starting with Alexandre de Moraes, emerges as a banner for Flávio, who spoke of changing the Constitution to "redemocratize" the country. source | tends to favor Lula |
| 15/10 | Government deliverables | Lula promises free city buses on weekends nationwide; the government had already zeroed federal taxes on fuel and launched a package for micro-entrepreneurs (MEIs). source | tends to favor Lula |
| 15/10 | Streets | The PL requests Av. Paulista for Oct 18, with a crowd it estimates at 20,000, in response to the pro-Lula rally. source | tends to favor Flávio |
| 15/10 | Markets | On Polymarket, Lula returns to a 25% chance; it was near 10% the previous week. source | tends to favor Lula |
| 16/10 | Labor | Congress promulgates Constitutional Amendment 141, which ends the 6x1 schedule; the maximum workweek falls from 44 to 42 hours within 60 days. source | tends to favor Lula |
| 16/10 | Streets | New round of pro-Lula rallies in 12 states on Friday and Saturday, plus Barcelona. source | tends to favor Lula |
| 17/10 | Scandals | Roberta Luchsinger, under investigation in Operation Sem Desconto (INSS fraud) and a friend of Lulinha, tells CNN she witnessed meetings between the president's son and government officials; the defense denies wrongdoing. source | tends to favor Flávio |
| 18/10 | Streets | Flávio's "Bloco do 22" rally on Av. Paulista, with Tarcísio and Ricardo Nunes, draws 71,900 people at the peak, according to USP's Political Debate Monitor with Cebrap and More in Common (from 63,300 to 80,500); the PL expected 20,000. source | tends to favor Flávio |
| 18/10 | Debate | Band hosts the first runoff debate between Lula and Flávio, with clashes over INSS and Lulinha, Banco Master, the pandemic (with a mea culpa from Flávio on health) and privatization of the SUS. source | neutral |
| 18/10 | Electoral courts | Justice Marco Aurélio Bellizze (TSE) gives Meta 24 hours to prove it treats both campaigns equally, as decided on Oct 15, under penalty of a higher daily fine. source | neutral |
| 19/10 | Debate | A Quaest qualitative assessment with 30 voter groups gives Flávio 82 points and Lula 79 for their performance in the Band debate: a "slightly superior" edge for the senator, without measuring votes. source | tends to favor Flávio |
| 19/10 | Electoral courts | Lula's campaign files a new action at the TSE accusing Meta of "algorithmic manipulation" in Flávio's favor and asks for the PL ticket's registration to be revoked. source | neutral |
| 19/10 | Electoral courts | The electoral inspector general, Ricardo Villas Bôas Cueva, bars Nikolas Ferreira and Pablo Marçal from reposting content already removed; the list includes an excerpt of a Flávio post about Lula's remark on street sweepers, which the full TSE had deemed lawful on Oct 15. source | tends to favor Flávio |
| 20/10 | Scandals | Lula asks the TSE to lift his own bank, tax and phone secrecy and that of family members, covering 8 years, and challenges Flávio to do the same by Oct 21; the rapporteur is André Mendonça, and the PL calls the request "bait". source | tends to favor Lula |
| 20/10 | Electoral courts | The TSE holds a closed-door meeting with Meta and lawyers from both campaigns; Meta says it treats the candidates "equally" and reports having made 720 posts unavailable. source | neutral |
| 20/10 | Government deliverables | Lula says he will send Congress the Tarifa Zero constitutional amendment (PEC), for free public transit, on Oct 21. source | tends to favor Lula |
| 20/10 | Markets | On Polymarket, Lula falls to 21.5% (1 p.m. Brasília time), from 23.5% on the night of Oct 19. Kalshi: Flávio 79% x Lula 22%. source | tends to favor Flávio |
| 20/10 | Scandals | At a rally in Juazeiro (BA), Lula says Flávio "won't have the courage" to waive his secrecy and that "his life is rotten"; Flávio's allies say he will not take up the challenge. source | tends to favor Lula |
| 20/10 | Electoral courts | Flávio's campaign asks the TSE (via an AIJE electoral-abuse suit) to annul the Lula–Alckmin ticket and bar Lula and minister Sidônio Palmeira from office over mass GovBR WhatsApp blasts about federal programs. source | tends to favor Flávio |
| 20/10 | Electoral courts | The full TSE unanimously upholds the injunction ordering the removal of posts by Gleisi, Lindbergh, Boulos and others linking Flávio to Operation Unha e Carne, to militias and to the "rachadinha" salary-skimming scheme. source | tends to favor Flávio |
| 20/10 | Rules of the game | Itamaraty accuses Milei of an "unacceptable attempt at foreign interference" after the Argentine president called Lula a "thief" and insinuated fraud in the runoff. source | neutral |
| 21/10 | Markets | On Polymarket, Lula stands at 20.5% (7:20 a.m. Brasília time), after 19.5% overnight; he was at 21.5% on the afternoon of Oct 20. Kalshi: Flávio 82% x Lula 19%. source | tends to favor Flávio |
| 21/10 | Lula's campaign | Quaest (Oct 17–20): Lula's rejection falls from 53% to 47% in one week and Flávio's from 54% to 52%; government approval rises to 46% (disapproval 47%). source | tends to favor Lula |
"Expected effect" is the panel's reading of whom the event tends to help, not a measurement. The list is updated with each new event (noticias_2turno.csv).
| Date | Side | Where | Crowd | Source |
|---|---|---|---|---|
| Oct 10 | pro-Flávio | Flávio's motorcycle convoy in São Gonçalo (RJ) | no estimate | O Hoje |
| Oct 11 | pro-Lula | Av. Paulista (SP), with Lula; rallies in more than 70 cities the same day | ≈ 261,000 based on the capacity of the occupied area (7 blocks, 4 people/m²); 1 million according to the campaign; no official count | Poder360 |
| Oct 12 | pro-Lula | Copacabana (RJ) | thousands, no numerical estimate | Comprova |
| Oct 13 | pro-Lula | Esplanada dos Ministérios (DF), student movement | thousands, no numerical estimate | Vermelho |
| Oct 16–17 | pro-Lula | New round in 12 states and in Barcelona; BH, Salvador, Belém, Fortaleza and Recife announced by the campaign | to be confirmed | Leia Sempre Brasil / CNN |
| Oct 18 | pro-Flávio | Flávio's "Bloco do 22" rally on Av. Paulista, with Tarcísio and Ricardo Nunes | 71,900 at peak (63,300 to 80,500), drone count by the Monitor do Debate Político (USP/Cebrap/More in Common); the PL expected 20,000. Not comparable to the Oct 11 rally, which had no drone count | Exame / USP |
The 7 polls with fieldwork completed by Oct 10 had, pooled, a margin of −4.2 ± 1.6 points in valid votes. The 5 completed between Oct 11 and 13, with part of the fieldwork before the rallies, −3.7 ± 1.9. The 45 completed afterward, −2.0 ± 0.6 (−1.7 without Gerp, which systematically measures Flávio higher). The narrowing of about 2 points is larger than the pooled sampling error.
The cleanest comparison is a pollster repeating its own method. AtlasIntel −5.6 → −1.5; DataTrends −6.4 → −4.2; Datafolha −4.3 → −1.1; Futura −3.5 → −0.3; Ipespe −5.4 → −1.1; PoderData −5.4 → −2.1; Vox Brasil +1.7 → +1.2. Taken together, the shift is +2.9 ± 2.2 points of margin. In PoderData, Lula went from 44% to 47% and Flávio from 49% to 47% in total votes through Oct 13–15, but the Oct 16–18 round went back to 46% to 49% and the Oct 19–21 round came in at 48% to 49%; in Datafolha, Flávio fell from 49% to 47% and Lula stayed at 45% through Oct 13–14, and on Oct 20–22 Lula rose to 46% with Flávio at 47% (50% to 50% in valid votes). Folha noted that the rallies "did not translate into voting intention" in the Oct 13–14 Datafolha: the shift there came from Flávio losing, not Lula gaining.
The rallies coincided with the vote to end the 6x1 work schedule, the fuel tax cut and the campaign's new digital strategy. Besides, whoever is behind usually recovers in the runoff: in 2014 and 2018, the first and last Datafolha and Ibope polls showed the runner-up gaining 6 to 10 points of margin in two weeks (in 2022, 2 points). What the data allow us to say is that Lula's recovery began on the weekend of the rallies and shows up in both pollsters with a comparable series. What the rallies changed more clearly was expectations: +5 points for Lula in PoderData through Oct 13–15 (+4 on Oct 16–18, +5 on Oct 19–21) and +3 in BTG/Nexus.
Street rallies persuade few; they mobilize those already on the same side. That is why their effect on the result runs through turnout. In the model, abstention equal to 2022 leaves Lula with a 17% chance; 2 points less abstention in the Northeast alone, the explicit goal of the rallies in state capitals, brings it to 21% (+0.4 points of margin). Today's chance, 18% in the ensemble, already incorporates the post-rally polls. Sociologist Alberto Carlos Almeida sums up the condition for the streets to turn into votes: the movement would have to grow "spontaneously" (CNN, Oct 13).
In one sentence: the comeback had two stages. In September, the consolidation of the anti-PT vote took Flávio from 35.7% to 43.0% of valid votes in the polls; on election day, a last mile of 4.1 points, which the polls did not capture, put him ahead. Lula stayed where the polls measured him: at his ceiling.
This is the mechanism of the comeback. The third way had 22.8% of valid votes in the polls of the first week of September, 12.4% in the last and 7.8% at the ballot box. Between the eve and the vote, Flávio received 4.2 of the 4.8 points the others lost, a conversion rate above the stated second choice (in Quaest, 46% of Cury's voters, 58% of Renan's and 59% of Zema's were going to Flávio; 4% to 25% to Lula). The campaign openly asked for a tactical vote from Sep 21 on, focusing on the Southeast. Zema finished with 0.27%, Caiado with 2.18%.
This is the necessary condition. Lula got 45.16%, practically what the polls measured (45.5), and never topped 46% in the weekly average since August. The government reached the election with 34% excellent/good ratings (Datafolha, Sep 22–23), exactly at the break-even point in the historical series of incumbent presidents running (33.7%), and with rejection of 45% to 55%, the same as Flávio's. With similar ceilings, whoever inherits the third way wins, and the 2026 third way was opposition. Compared with 2022, Lula lost 3.4 million votes (57.26 → 53.88 million) and the right-wing candidate gained 5.0 million (51.07 → 56.10 million).
This is where the election flipped. Compared with the 2022 runoff, Lula's margin fell 7.8 points in the South, 6.4 in the Center-West, 4.3 in the Southeast and 3.9 in the Northeast. The Northeast still gave Lula a net 11.1 million votes, but the Southeast, South and Center-West gave back 12.9 million. Minas Gerais, which decided 2022 for Lula, went to Flávio. The right reached election day with strong governors in the two largest electorates: Tarcísio de Freitas re-elected in São Paulo with 62.85% and Cleitinho elected in Minas with 55.41%, both in the first round.
Part of the surprise was in the instrument, not the electorate. In-person polls missed the margin by +5.2; Palver had Flávio at 47.5% on the eve, and the polling average, 42.8%. State polls overestimated Lula by 7.3 points in the South and 6.3 in the Southeast, where Flávio's voters are the majority. The pattern is consistent with differential nonresponse: right-wing voters are less likely to accept the interview, and weighting by 2022 vote does not correct for that. For those who read only the benchmark polls, Lula led until the end; for those who discounted the historical error, a tie with an edge for Flávio was the central scenario.
The events of the final week reinforced a trend that already existed, without creating a new one. The Globo debate, scheduled for Oct 1, was canceled after Flávio pulled out and the legal fight over the empty podium: the rising candidate avoided the only direct confrontation that could have interrupted the movement. Both sides faced negative news in the final stretch (Lulinha's trip revealed by the Federal Police on Sep 29; the Dark Horse earmarks case and the suspension of the "Foi o Flávio" jingle), and polls from Sep 27 to Oct 3 kept moving toward Flávio. There is no way to isolate the effect of each event with the available data.
Abstention was 21.1%, versus 20.9% in 2022. It fell in the Northeast (18.4%, from 19.5%) and rose in the South, Southeast and Center-West, a geography that favored Lula. The correlation between the change in abstention and the change in the vote by state is −0.15. The low-turnout scenario discussed on the eve did not happen and does not explain the result.
The other candidates totaled 7.8% of valid votes. To reach 50% with them alone, with no change in turnout, Lula would need 62% of those votes; Flávio, 38%. The stated second choice before the first round (Quaest, Sep) gave Lula about a quarter; polls taken after the first round show more: PoderData (Oct 10–12) gives Lula 43% of those who choose one of the two, and Datafolha (Oct 6–7) 32% versus 46% for Flávio overall, 41% of the decided. Adding the official result and that transfer, the runoff starts at −2.7 for Lula (Flávio ahead).
The tactical vote already brought the most anti-PT third-way voters to Flávio in the first round; the 7.8% who resisted him are likely to transfer less to Flávio than the stated matrix assumes. Lula's path runs through that remainder, through people who did not vote in the first round (33.5 million) and through the Southeast, where every 3 points of margin are worth about 1.2 nationally. In 2014 and 2018, the left's candidate gained 6 to 10 points in the polls over the course of the runoff.
In today's run, the model gives Lula a 17% chance in the runoff; the betting market (Polymarket, Oct 25, 8 a.m.), 20%. The figure already combines the counted vote with 57 polls taken after Oct 4.
In one sentence: the polls measured Lula well and everything else badly: in the final week Lula came out with an average error of −0.6 points, Flávio −4.2 and the other candidates +4.8; the voter the polls still placed in the third way voted for Flávio.
In the week to Sep 6, Flávio had 35.7% of valid votes, Lula 41.5% and the others 22.8% (Pablo Marçal was still on the lists until his candidacy was rejected on Sep 11). In the final week, Flávio stood at 43.0%, Lula at 44.7% and the others at 12.4%. The average margin fell from +5.8 to +1.7: the polls recorded 7.2 points of growth for Flávio, nearly all of it from the third way.
Between the final week of polling and the ballot box, Flávio gained another 4.1 points and the other candidates lost 4.6; Lula moved +0.5. Aggregate data cannot separate a last-minute tactical vote from Flávio voters who did not show up in the polls (hidden among the undecided, the third way or nonrespondents). Both explanations produce the same picture, and both point to Flávio.
| Pollster | Method | Fieldwork through | n | Valid L × F | Margin error | Lula error | Flávio error | Others error | Model correction | Error after correction |
|---|---|---|---|---|---|---|---|---|---|---|
| Palver | online | 03/10 | 5,000 | 43.4 × 47.5 | −2.2 | −1.7 | +0.4 | +1.3 | +4.4 | +2.2 |
| Futura | phone | 29/09 | 2,000 | 41.3 × 44.2 | −1.1 | −3.9 | −2.8 | +6.7 | +2.4 | +1.3 |
| Gerp | phone | 28/09 | 2,400 | 42.1 × 44.2 | −0.2 | −3.1 | −2.8 | +5.9 | +5.2 | +4.9 |
| Vox Brasil | not reported | 01/10 | 2,100 | 45.0 × 45.9 | +1.0 | −0.2 | −1.1 | +1.3 | +1.8 | +2.8 |
| Meio/Ideia | phone | 28/09 | 2,000 | 40.1 × 39.1 | +2.9 | −5.1 | −8.0 | +13.0 | +1.5 | +4.4 |
| PoderData ★ | phone | 02/10 | 4,000 | 44.7 × 43.6 | +2.9 | −0.5 | −3.4 | +3.9 | +2.4 | +5.3 |
| Datafolha ★ | in-person | 03/10 | 4,006 | 45.7 × 43.5 | +4.0 | +0.5 | −3.5 | +3.1 | +0.5 | +4.5 |
| DataTrends | phone | 30/09 | 2,000 | 45.1 × 42.9 | +4.1 | −0.1 | −4.2 | +4.3 | +0.6 | +4.7 |
| Quaest ★ | in-person | 03/10 | 3,702 | 46.0 × 43.7 | +4.2 | +0.8 | −3.3 | +2.5 | −0.9 | +3.3 |
| AtlasIntel ★ | online | 28/09 | 5,005 | 46.3 × 43.1 | +5.0 | +1.1 | −3.9 | +2.8 | +0.3 | +5.3 |
| Real Time Big Data | phone | 30/09 | 2,000 | 45.7 × 41.5 | +6.1 | +0.6 | −5.5 | +5.0 | −0.3 | +5.9 |
| BTG/Nexus | phone | 27/09 | 2,000 | 44.2 × 38.9 | +7.1 | −1.0 | −8.1 | +9.0 | +1.2 | +8.3 |
| CNT/MDA ★ | in-person | 02/10 | 2,007 | 47.8 × 42.2 | +7.5 | +2.7 | −4.9 | +2.2 | −1.9 | +5.6 |
| Indexa | phone | 29/09 | 2,000 | 45.9 × 40.0 | +7.7 | +0.7 | −7.0 | +6.3 | 0.0 | +7.8 |
| Method | Pollsters | Average margin error | Absolute error | Lula error | Flávio error | Others error |
|---|---|---|---|---|---|---|
| not reported | 1 | +1.0 | 1.0 | −0.2 | −1.1 | +1.3 |
| online | 2 | +1.4 | 3.6 | −0.3 | −1.7 | +2.0 |
| phone | 8 | +3.7 | 4.0 | −1.5 | −5.2 | +6.8 |
| in-person | 3 | +5.2 | 5.2 | +1.3 | −3.9 | +2.6 |
The classic error of Brazilian first-round polls repeated itself for the fifth election in a row (2010: +5.2; 2014: +10.5; 2018: +2.4; 2022: +5.8; 2026: +4.7 in the anchor). The shape repeated too: the PT vote is measured precisely, and the error lies in the opposing camp, which the polls show as more fragmented and smaller than it is. Weighting by stated 2022 vote, a practice several pollsters adopted, did not fix it, probably because the right-wing voter who does not respond does not report a past vote either.
The pollsters that the market and the model itself treated as pro-Flávio outliers (Palver, Gerp, Futura, Vox; house effects of −2 to −5) missed the margin by 1.1 points on average. The most established in-person pollsters (Datafolha, Quaest, CNT/MDA) missed by 5.2. Palver, online, was the only one to measure Flávio's level (47.5%) almost exactly. One election is not enough to change the anchor, but it is enough to stop treating these pollsters' divergence as noise.
| Region | Share of valid votes | Lula × Flávio | Two-party margin | 2022 runoff | Change | State-poll error | Net votes (Lula − Flávio) | Abstention (2022) |
|---|---|---|---|---|---|---|---|---|
| Southeast | 40.7% | 39.7 × 51.4 | −12.8 | −8.5 | −4.3 | +6.3 | −5.66 million | 22.7% (22.0%) |
| Northeast | 28.3% | 63.8 × 30.9 | +34.8 | +38.7 | −3.9 | −4.5 | +11.10 million | 18.4% (19.5%) |
| South | 14.7% | 31.3 × 60.1 | −31.5 | −23.7 | −7.8 | +7.3 | −5.04 million | 20.5% (19.3%) |
| North | 8.6% | 44.6 × 49.2 | −4.8 | −2.0 | −2.8 | −4.8 | −0.46 million | 19.4% (20.9%) |
| Center-West | 7.6% | 32.6 × 56.5 | −26.8 | −20.4 | −6.4 | +3.6 | −2.17 million | 21.7% (21.4%) |
Lula lost more than 1 point of margin relative to 2022 in 20 of the 27 states: the shift was national, strongest in the South and Center-West. Minas Gerais flipped (+0.4 in 2022, −5.4 now). State polls erred with a regional sign: they overestimated Lula in the South, Southeast and Center-West and underestimated him in the Northeast and North. The change in abstention does not track the change in the vote (correlation −0.15 across states).
| State | Lula × Flávio | Others | Two-party margin | 2022 runoff | Change | State-poll error (no.) | Abstention (vs. 2022) |
|---|---|---|---|---|---|---|---|
| TO | 43.4 × 50.4 | 6.1 | −7.5 | +2.7 | −10.2 | −21.6 (1) | 18.3% (−0.2) |
| GO | 31.1 × 53.6 | 15.3 | −26.6 | −17.4 | −9.2 | +1.8 (2) | 21.8% (+0.1) |
| RS | 35.7 × 55.6 | 8.6 | −21.8 | −12.7 | −9.1 | +10.7 (3) | 21.4% (+1.6) |
| MT | 29.2 × 65.1 | 5.7 | −38.1 | −30.2 | −8.0 | +6.9 (2) | 22.2% (−1.2) |
| MA | 64.0 × 30.9 | 5.1 | +34.9 | +42.3 | −7.4 | −5.9 (2) | 19.7% (−2.5) |
| SC | 25.0 × 66.7 | 8.3 | −45.4 | −38.5 | −6.8 | +10.6 (1) | 18.9% (+0.5) |
| PR | 31.2 × 59.9 | 8.9 | −31.5 | −24.8 | −6.7 | +1.7 (4) | 20.6% (+1.1) |
| MS | 34.7 × 58.6 | 6.7 | −25.6 | −19.0 | −6.7 | +3.9 (1) | 23.9% (+1.8) |
| CE | 63.3 × 31.3 | 5.4 | +33.9 | +39.9 | −6.1 | +0.2 (4) | 15.5% (−1.9) |
| MG | 43.3 × 48.2 | 8.4 | −5.4 | +0.4 | −5.8 | +4.2 (7) | 22.8% (+0.5) |
| SP | 38.2 × 51.9 | 9.9 | −15.2 | −10.5 | −4.8 | +7.6 (7) | 22.5% (+0.9) |
| BA | 66.2 × 28.5 | 5.3 | +39.7 | +44.2 | −4.5 | −8.2 (4) | 20.0% (−1.4) |
| PI | 71.0 × 24.1 | 4.9 | +49.3 | +53.7 | −4.4 | −9.7 (2) | 17.2% (−0.5) |
| RN | 59.8 × 34.8 | 5.5 | +26.4 | +30.2 | −3.8 | −1.6 (2) | 18.4% (+0.2) |
| PA | 49.9 × 44.5 | 5.6 | +5.7 | +9.5 | −3.8 | −8.6 (2) | 19.1% (−2.1) |
| PB | 61.3 × 33.1 | 5.6 | +29.9 | +33.2 | −3.3 | −3.3 (3) | 17.4% (+0.1) |
| RO | 25.9 × 67.5 | 6.7 | −44.5 | −41.3 | −3.2 | +9.1 (1) | 21.7% (−2.9) |
| AL | 54.7 × 40.4 | 4.8 | +15.0 | +17.4 | −2.4 | −0.1 (2) | 20.6% (−1.7) |
| ES | 37.8 × 54.8 | 7.5 | −18.4 | −16.1 | −2.3 | +10.1 (1) | 21.4% (+0.6) |
| RJ | 39.4 × 53.0 | 7.6 | −14.7 | −13.1 | −1.6 | +4.8 (7) | 23.3% (+0.6) |
| SE | 62.7 × 30.6 | 6.6 | +34.4 | +34.4 | 0.0 | −9.2 (2) | 17.3% (−1.0) |
| PE | 63.5 × 31.0 | 5.5 | +34.3 | +33.9 | +0.5 | −2.7 (8) | 18.2% (0.0) |
| RR | 22.9 × 71.1 | 6.1 | −51.3 | −52.2 | +0.8 | +5.1 (3) | 17.1% (+0.4) |
| AM | 48.2 × 45.0 | 6.8 | +3.5 | +2.2 | +1.3 | −1.1 (2) | 20.2% (0.0) |
| AC | 28.7 × 64.6 | 6.7 | −38.4 | −40.6 | +2.2 | +8.0 (1) | 20.4% (−2.0) |
| AP | 45.7 × 45.7 | 8.6 | +0.1 | −2.7 | +2.8 | +2.5 (2) | 17.4% (−2.1) |
| DF | 38.1 × 51.3 | 10.6 | −14.8 | −17.6 | +2.9 | +3.5 (6) | 18.9% (+1.4) |
In one sentence: the line that sums up all the polls now stands at -0.9 points for Lula, versus -2.0 two weeks ago — a shift toward Lula, still within the margin of uncertainty.
Over the last two weeks the adjusted margin went from -2.0 to -0.9 (1.1 pp toward Lula); over the last week, stable. The slide continues, but more slowly: the second week moved less than the first. Comparing each poll with the same pollster's previous wave, 20 moved toward Lula, 10 toward Flávio and 7 stayed stable — the movement also shows up within each pollster's series. The band around the line (95%) shows that none of these changes takes the race out of a statistical tie.
| Pollster | Fieldwork through | Previous wave (Lula × Flávio) | Now | Δ valid margin | Movement |
|---|---|---|---|---|---|
| Gerp | 24/10 | 45 × 50 (21/10) | 46 × 49 | +2.1 | → Lula |
| Palver (v3 calibration) | 24/10 | 48 × 50 (20/10) | 49 × 50 | +1.0 | → Lula |
| Datafolha | 24/10 | 46 × 47 (22/10) | 47 × 48 | 0.0 | stable |
| Quaest | 24/10 | 44 × 45 (20/10) | 45 × 46 | 0.0 | stable |
| BTG/Nexus | 24/10 | 45.9 × 46.5 (18/10) | 46 × 47 | -0.4 | stable |
| PoderData | 24/10 | 48 × 49 (21/10) | 47 × 49 | -1.1 | → Flávio |
| Futura (education from PNAD) | 24/10 | 46.8 × 48.9 (20/10) | 47.9 × 48.2 | +1.9 | → Lula |
| AtlasIntel | 23/10 | 47.6 × 49.8 (18/10) | 48.4 × 49.9 | +0.7 | → Lula |
| Veritá | 23/10 | 47.5 × 50.2 (22/10) | 46.5 × 49.4 | -0.3 | stable |
| Instituto França | 23/10 | 39.6 × 41.0 (20/10) | 43.1 × 45.6 | -1.2 | → Flávio |
| CNT/MDA (weighted by first round) | 23/10 | 45.9 × 46.8 (15/10) | 45 × 46.3 | -0.5 | → Flávio |
| Datafolha | 22/10 | 45 × 47 (14/10) | 46 × 47 | +1.1 | → Lula |
| Vox Brasil | 22/10 | 46.4 × 44.6 (17/10) | 45.4 × 44.3 | -0.8 | → Flávio |
| Real Time Big Data | 22/10 | 45 × 48 (19/10) | 45 × 49 | -1.0 | → Flávio |
| Alfa Inteligência | 22/10 | 45.9 × 48.8 (19/10) | 46 × 49.9 | -1.0 | → Flávio |
| Ipespe | 22/10 | 46 × 48 (15/10) | 47 × 48 | +1.1 | → Lula |
| Veritá | 22/10 | 48.2 × 49.7 (19/10) | 47.5 × 50.2 | -1.2 | → Flávio |
| Gerp | 21/10 | 45 × 50 (19/10) | 45 × 50 | 0.0 | stable |
| DataTrends | 21/10 | 46 × 49 (17/10) | 46 × 50 | -1.0 | → Flávio |
| PoderData | 21/10 | 46 × 49 (18/10) | 48 × 49 | +2.1 | → Lula |
| Indexa | 21/10 | 42 × 44 (14/10) | 44.1 × 45.2 | +1.1 | → Lula |
| Quaest | 20/10 | 44 × 46 (13/10) | 44 × 45 | +1.1 | → Lula |
| Palver (v3 calibration) | 20/10 | 49 × 50 (13/10) | 48 × 50 | -1.0 | → Flávio |
| Veritá | 19/10 | 47.7 × 49.7 (16/10) | 48.2 × 49.7 | +0.5 | → Lula |
| Gerp | 19/10 | 45 × 50 (16/10) | 45 × 50 | 0.0 | stable |
| PoderData | 18/10 | 47 × 47 (15/10) | 46 × 49 | -3.2 | → Flávio |
| AtlasIntel | 18/10 | 45.7 × 51.1 (08/10) | 47.6 × 49.8 | +3.3 | → Lula |
| BTG/Nexus | 18/10 | 45 × 48 (12/10) | 45.9 × 46.5 | +2.6 | → Lula |
| Futura Inteligência | 17/10 | 47.5 × 49 (15/10) | 47.5 × 48.3 | +0.7 | → Lula |
| DataTrends | 17/10 | 44 × 50 (09/10) | 46 × 49 | +3.2 | → Lula |
| Vox Brasil | 17/10 | 44.2 × 42.7 (07/10) | 46.4 × 44.6 | +0.3 | stable |
| Gerp | 16/10 | 44 × 53 (12/10) | 45 × 50 | +4.0 | → Lula |
| PoderData | 15/10 | 46 × 48 (12/10) | 47 × 47 | +2.1 | → Lula |
| Futura Inteligência | 15/10 | 46.6 × 50 (10/10) | 47.5 × 49 | +2.0 | → Lula |
| Ipespe | 15/10 | 44 × 49 (08/10) | 46 × 48 | +3.2 | → Lula |
| Datafolha | 14/10 | 45 × 49 (07/10) | 45 × 47 | +2.1 | → Lula |
| PoderData | 12/10 | 44 × 49 (07/10) | 46 × 48 | +3.2 | → Lula |
Quaest · 26 states · fieldwork ended between Sep 21 and Sep 28 · 804 to 1,800 interviews per state · commissioned by: Globo, Inter TV, NSC, RBS, RPC, Rede Amazonica, TV Anhanguera, TV Bahia, TV Cabo Branco/TV Paraiba, TV Centro America, TV Gazeta, TV Gazeta ES, TV Liberal, TV Mirante, TV Morena, TV Sergipe, TV Verdes Mares
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 25–28/09 | 1,800 | 30 | 36 | −6 | Cury 5 | 9 / 12 | 21.6 / 21.1 | 21.6% |
| MG | 25–28/09 | 1,506 | 35 | 31 | +4 tie | Cury 5 | 6 / 14 | 22.3 / 21.0 | 10.4% |
| RJ | 25–28/09 | 1,302 | 31 | 38 | −7 | Cury 4 | 11 / 11 | 22.7 / 22.2 | 8.1% |
| BA | 20–23/09 | 900 | 58 | 23 | +35 | — | 5 / 6 | 21.4 / 20.5 | 7.2% |
| PR | 20–23/09 | 804 | 26 | 40 | −14 | Cury 5 | 6 / 13 | 19.5 / 18.6 | 5.5% |
| RS | 20–23/09 | 900 | 31 | 35 | −4 tie | — | — / 16 | 19.8 / 19.3 | 5.4% |
| PE | 25–28/09 | 1,302 | 58 | 20 | +38 | Cury 3 | 5 / 11 | 18.2 / 17.3 | 4.6% |
| CE | 19–22/09 | 900 | 55 | 23 | +32 | Cury 5 | 7 / 6 | 17.4 / 17.1 | 4.4% |
| PA | 22–25/09 | 804 | 40 | 35 | +5 tie | Cury 7 | 4 / 10 | 21.2 / 22.6 | 4.0% |
| SC | 20–23/09 | 804 | 23 | 50 | −27 | Cury 5 | — / — | 18.4 / 17.4 | 3.6% |
| MA | 21–24/09 | 900 | 59 | 22 | +37 | — | — / — | 22.2 / 23.5 | 3.3% |
| GO | 21–24/09 | 804 | 27 | 33 | −6 tie | Caiado 23 | — / — | 21.7 / 20.5 | 3.2% |
| PB | 18–21/09 | 804 | 54 | 23 | +31 | Cury 6 | — / — | 17.3 / 16.7 | 2.1% |
| ES single source | 21–24/09 | 804 | 31 | 37 | −6 tie | — | — / 13 | 20.8 / 20.5 | 1.9% |
| AM | 20–23/09 | 804 | 38 | 32 | +6 tie | Cury 6 | 6 / 14 | 20.1 / 21.9 | 1.8% |
| RN | 21–24/09 | 804 | 52 | 25 | +27 | Cury 6 | — / — | 18.2 / 17.4 | 1.7% |
| MT | 21–24/09 | 804 | 25 | 50 | −25 | Cury 5 | 3 / 11 | 23.4 / 22.5 | 1.7% |
| AL | 21–24/09 | 804 | 44 | 32 | +12 | Cury 5 | 6 / 8 | 22.4 / 23.2 | 1.5% |
| DF | 25–28/09 | 1,104 | 32 | 38 | −6 | Caiado 8 | 6 / — | 17.5 / 16.7 | 1.4% |
| MS | 21–24/09 | 804 | 27 | 42 | −15 | — | — / 13 | 22.1 / 22.4 | 1.3% |
| SE | 20–23/09 | 804 | 54 | 26 | +28 | — | — / — | 18.4 / 18.9 | 1.1% |
| RO | 20–23/09 | 804 | 19 | 53 | −34 | Cury 4 | 6 / 12 | 24.6 / 24.7 | 0.8% |
| TO | 21–24/09 | statistical tie reported without the numbers | 18.5 / 20.4 | 0.7% | |||||
| AC | 21–24/09 | 804 | 24 | 50 | −26 | Cury 4 | — / — | 22.4 / 28.4 | 0.4% |
| AP single source | 21–24/09 | 804 | 42 | 35 | +7 | — | — / — | 19.5 / 27.2 | 0.4% |
| RR | 21–24/09 | 804 | 20 | 59 | −39 | — | — / — | 16.7 / 21.9 | 0.3% |
Quaest · 19 states · fieldwork ended between Sep 21 and Oct 20 · 804 to 1,800 interviews per state · commissioned by: Globo, Rede Amazonica, TV Anhanguera, TV Bahia, TV Cabo Branco/TV Paraiba, TV Centro America, TV Gazeta, TV Globo, TV Liberal, TV Mirante, TV Sergipe
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 22/09 | 1,800 | 36 | 42 | −6 | — | — / — | 21.6 / 21.1 | 21.6% |
| MG | 28/09 | 1,506 | 41 | 40 | +1 tie | — | 15 / 4 | 22.3 / 21.0 | 10.4% |
| RJ | 28/09 | 1,302 | 34 | 44 | −10 | — | 15 / 7 | 22.7 / 22.2 | 8.1% |
| BA | 23/09 | 900 | 62 | 27 | +35 | — | 7 / 4 | 21.4 / 20.5 | 7.2% |
| PE | 28/09 | 1,302 | 61 | 24 | +37 | — | — / — | 18.2 / 17.3 | 4.6% |
| CE | 22/09 | 900 | 59 | 27 | +32 | — | — / — | 17.4 / 17.1 | 4.4% |
| PA | 25/09 | 804 | 44 | 40 | +4 tie | — | — / — | 21.2 / 22.6 | 4.0% |
| MA | 24/09 | 900 | 63 | 26 | +37 | — | 8 / 3 | 22.2 / 23.5 | 3.3% |
| PB | 21/09 | 804 | 60 | 26 | +34 | — | 10 / 4 | 17.3 / 16.7 | 2.1% |
| ES | 13/10 | 1,098 | 38 | 51 | −13 | — | 6 / 5 | 20.8 / 20.5 | 1.9% |
| AM | 12/10 | 1,098 | 47 | 43 | +4 tie | — | 5 / 5 | 20.1 / 21.9 | 1.8% |
| RN | 12/10 | 1,098 | 59 | 33 | +26 | — | 5 / 3 | 18.2 / 17.4 | 1.7% |
| MT | 24/09 | 804 | 30 | 56 | −26 | — | — / — | 23.4 / 22.5 | 1.7% |
| AL | 24/09 | 804 | 48 | 34 | +14 | — | — / — | 22.4 / 23.2 | 1.5% |
| DF | 20/10 | 1,104 | 39 | 50 | −11 | — | 6 / 5 | 17.5 / 16.7 | 1.4% |
| SE | 23/09 | 804 | 56 | 30 | +26 | — | — / — | 18.4 / 18.9 | 1.1% |
| TO | 13/10 | 1,098 | 42 | 48 | −6 | — | 6 / 4 | 18.5 / 20.4 | 0.7% |
| AC | 12/10 | 1,098 | 28 | 60 | −32 | — | 5 / 7 | 22.4 / 28.4 | 0.4% |
| RR | 24/09 | 804 | 22 | 67 | −45 | — | — / — | 16.7 / 21.9 | 0.3% |
Veritá · 26 states · fieldwork ended between Sep 15 and Sep 19 · 1,030 to 3,025 interviews per state · sponsors: self-funded
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP single source | 14–19/09 | 3,025 | 38.6 | 47.5 | −8.9 | Renan 5.1 | 3 / — | 21.6 / 21.1 | 21.6% |
| MG single source | 12–17/09 | 2,030 | 35.2 | 38.5 | −3.3 tie | Cury 7.2 | 11.9 / — | 22.3 / 21.0 | 10.4% |
| RJ single source | 13–18/09 | 2,030 | 37.2 | 50.3 | −13.1 | Cury 4.7 | 2.6 / — | 22.7 / 22.2 | 8.1% |
| BA single source | 10–15/09 | 2,020 | 52.3 | 35.7 | +16.6 | Cury 5.3 | 1.3 / — | 21.4 / 20.5 | 7.2% |
| PR single source | 13–18/09 | 2,010 | 28.2 | 51.5 | −23.3 | Cury 4.7 | 5.6 / — | 19.5 / 18.6 | 5.5% |
| RS single source | 14–19/09 | 2,020 | 35.8 | 50.8 | −15 | Cury 3.9 | 2 / — | 19.8 / 19.3 | 5.4% |
| PE single source | 13–18/09 | 2,010 | 51.1 | 36.2 | +14.9 | Cury 6.1 | 2.6 / — | 18.2 / 17.3 | 4.6% |
| CE single source | 10–15/09 | 1,525 | 51.7 | 31.5 | +20.2 | Cury 6.9 | 4.6 / — | 17.4 / 17.1 | 4.4% |
| PA single source | 12–17/09 | 1,525 | 36.4 | 45.1 | −8.7 | Cury 7.5 | 5.5 / — | 21.2 / 22.6 | 4.0% |
| MA single source | 12–17/09 | 1,525 | 51.3 | 34.2 | +17.1 | Cury 7.2 | 1.7 / — | 22.2 / 23.5 | 3.3% |
| GO single source | 12–17/09 | 1,525 | 30.4 | 42.3 | −11.9 | Caiado 14.9 | 3.7 / — | 21.7 / 20.5 | 3.2% |
| PB single source | 13–18/09 | 1,220 | 47.8 | 36.6 | +11.2 | Cury 7.6 | 2.2 / — | 17.3 / 16.7 | 2.1% |
| ES single source | 12–17/09 | 1,220 | 41.4 | 45.1 | −3.7 tie | Cury 8.3 | 0.7 / — | 20.8 / 20.5 | 1.9% |
| AM single source | 10–15/09 | 1,220 | 40.1 | 40.6 | −0.5 tie | Cury 7.8 | 7.3 / — | 20.1 / 21.9 | 1.8% |
| PI single source | 13–18/09 | 1,220 | 54 | 25.6 | +28.4 | Cury 5.9 | 10.9 / — | 17.7 / 18.7 | 1.7% |
| RN single source | 13–18/09 | 1,220 | 53.5 | 33.1 | +20.4 | Cury 6.6 | 2.8 / — | 18.2 / 17.4 | 1.7% |
| MT single source | 12–17/09 | 1,220 | 26 | 50.6 | −24.6 | Cury 4.3 | 12.2 / — | 23.4 / 22.5 | 1.7% |
| AL single source | 10–15/09 | 1,220 | 48.6 | 36.4 | +12.2 | Cury 5.1 | 4.9 / — | 22.4 / 23.2 | 1.5% |
| DF single source | 10–15/09 | 1,220 | 47.3 | 38.3 | +9 | Cury 6 | 0.3 / — | 17.5 / 16.7 | 1.4% |
| MS single source | 12–17/09 | 1,220 | 32.2 | 49.3 | −17.1 | Cury 5.5 | 8.9 / — | 22.1 / 22.4 | 1.3% |
| SE single source | 14–19/09 | 1,220 | 53.9 | 32.6 | +21.3 | Cury 7.5 | 1.8 / — | 18.4 / 18.9 | 1.1% |
| RO single source | 14–19/09 | 1,220 | 28.8 | 60.4 | −31.6 | Cury 4.3 | 1.9 / — | 24.6 / 24.7 | 0.8% |
| TO single source | 14–19/09 | 1,220 | 31.1 | 55.7 | −24.6 | Cury 6.2 | 2.5 / — | 18.5 / 20.4 | 0.7% |
| AC single source | 10–15/09 | 1,030 | 30.2 | 53.1 | −22.9 | Cury 9.8 | 0.9 / — | 22.4 / 28.4 | 0.4% |
| AP single source | 10–15/09 | 1,030 | 40.5 | 41.2 | −0.7 tie | Cury 8.4 | 1.3 / — | 19.5 / 27.2 | 0.4% |
| RR single source | 14–19/09 | 1,030 | 21.8 | 65.3 | −43.5 | Cury 5.7 | 4 / — | 16.7 / 21.9 | 0.3% |
Veritá · 26 states · fieldwork ended between Sep 15 and Sep 19 · 1,030 to 3,025 interviews per state · sponsors: self-funded
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 19/09 | 3,025 | 40.6 | 52.3 | −11.7 | — | 7.2 / — | 21.6 / 21.1 | 21.6% |
| MG | 17/09 | 2,030 | 38.4 | 45.6 | −7.2 | — | 16 / — | 22.3 / 21.0 | 10.4% |
| RJ | 18/09 | 2,030 | 39.3 | 53.4 | −14.1 | — | 7.3 / — | 22.7 / 22.2 | 8.1% |
| BA | 15/09 | 2,020 | 55.1 | 40.5 | +14.6 | — | 4.4 / — | 21.4 / 20.5 | 7.2% |
| PR | 18/09 | 2,010 | 30.6 | 58.9 | −28.3 | — | 10.4 / — | 19.5 / 18.6 | 5.5% |
| RS | 19/09 | 2,020 | 37.9 | 55.7 | −17.8 | — | 6.4 / — | 19.8 / 19.3 | 5.4% |
| PE | 18/09 | 2,010 | 52.9 | 40.1 | +12.8 | — | 7 / — | 18.2 / 17.3 | 4.6% |
| CE | 15/09 | 1,525 | 54.1 | 36 | +18.1 | — | 9.9 / — | 17.4 / 17.1 | 4.4% |
| PA | 17/09 | 1,525 | 39.6 | 48.8 | −9.2 | — | 11.6 / — | 21.2 / 22.6 | 4.0% |
| MA | 17/09 | 1,525 | 53.7 | 38.6 | +15.1 | — | 7.8 / — | 22.2 / 23.5 | 3.3% |
| GO | 17/09 | 1,525 | 33.4 | 53.4 | −20 | — | 13.2 / — | 21.7 / 20.5 | 3.2% |
| PB | 18/09 | 1,220 | 51.6 | 40.9 | +10.7 | — | 7.5 / — | 17.3 / 16.7 | 2.1% |
| ES | 17/09 | 1,220 | 43.1 | 50.9 | −7.8 | — | 6 / — | 20.8 / 20.5 | 1.9% |
| AM | 15/09 | 1,220 | 42.2 | 43.1 | −0.9 tie | — | 14.7 / — | 20.1 / 21.9 | 1.8% |
| PI | 18/09 | 1,220 | 57.6 | 28.7 | +28.9 | — | 13.6 / — | 17.7 / 18.7 | 1.7% |
| RN | 18/09 | 1,220 | 55.4 | 37.4 | +18 | — | 7.1 / — | 18.2 / 17.4 | 1.7% |
| MT | 17/09 | 1,220 | 28.1 | 54.9 | −26.8 | — | 17.1 / — | 23.4 / 22.5 | 1.7% |
| AL | 15/09 | 1,220 | 52.8 | 40.8 | +12 | — | 6.4 / — | 22.4 / 23.2 | 1.5% |
| DF | 15/09 | 1,220 | 49.7 | 45.4 | +4.3 tie | — | 4.9 / — | 17.5 / 16.7 | 1.4% |
| MS | 17/09 | 1,220 | 34.4 | 53.5 | −19.1 | — | 12.1 / — | 22.1 / 22.4 | 1.3% |
| SE | 19/09 | 1,220 | 56.7 | 37.6 | +19.1 | — | 5.7 / — | 18.4 / 18.9 | 1.1% |
| RO | 19/09 | 1,220 | 30.5 | 63.9 | −33.4 | — | 5.6 / — | 24.6 / 24.7 | 0.8% |
| TO | 19/09 | 1,220 | 33.6 | 61.1 | −27.5 | — | 5.3 / — | 18.5 / 20.4 | 0.7% |
| AC | 15/09 | 1,030 | 32.7 | 61.3 | −28.6 | — | 6 / — | 22.4 / 28.4 | 0.4% |
| AP | 15/09 | 1,030 | 43.9 | 46.2 | −2.3 tie | — | 9.9 / — | 19.5 / 27.2 | 0.4% |
| RR | 19/09 | 1,030 | 23.5 | 70.8 | −47.3 | — | 5.6 / — | 16.7 / 21.9 | 0.3% |
AtlasIntel · 18 states · fieldwork ended between Sep 1 and Sep 28 · 1,185 to 2,000 interviews per state · sponsors: A Tarde, Estadão, Focus Comunicação e Mídia, Grupo ND, TV Meio Norte, own funds
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 15–20/09 | 1,800 | 39.8 | 45.1 | −5.3 | Renan 7.6 | 0.7 / 1.2 | 21.6 / 21.1 | 21.6% |
| MG single source | 30/08–04/09 | 1,804 | 44.3 | 35.4 | +8.9 | Cury 8.9 | — / — | 22.3 / 21.0 | 10.4% |
| RJ single source | 15–20/09 | 1,829 | 41.1 | 40.3 | +0.8 tie | Renan 4.7 | — / — | 22.7 / 22.2 | 8.1% |
| BA | 24–28/09 | 2,000 | 57.6 | 27.8 | +29.8 | Cury 3.9 | 2.1 / — | 21.4 / 20.5 | 7.2% |
| PR | 11–16/09 | 1,794 | 34 | 53.7 | −19.7 | Cury 4.1 | 0.4 / 1.8 | 19.5 / 18.6 | 5.5% |
| RS | 15–20/09 | 1,819 | 36.2 | 41.6 | −5.4 | Cury 9.5 | 0.1 / 0.9 | 19.8 / 19.3 | 5.4% |
| PE | 15–20/09 | 1,836 | 58.3 | 29.9 | +28.4 | Cury 3.2 | 2.3 / 0.7 | 18.2 / 17.3 | 4.6% |
| CE | 28/08–02/09 | 1,834 | 51.6 | 25.7 | +25.9 | Cury 12.3 | 1 / 3.7 | 17.4 / 17.1 | 4.4% |
| SC | 04–09/09 | 1,211 | 25 | 48 | −23 | Cury 9.9 | 1.5 / 0.3 | 18.4 / 17.4 | 3.6% |
| GO single source | 27/08–01/09 | 1,214 | 26.9 | 38.2 | −11.3 | Caiado 19.2 | 2.1 / 0.3 | 21.7 / 20.5 | 3.2% |
| PB single source | 28/08–02/09 | 1,207 | 51.8 | 28.1 | +23.7 | Cury 8.9 | 2.4 / 1.5 | 17.3 / 16.7 | 2.1% |
| AM single source | 29/08–03/09 | 1,185 | 39.2 | 41.3 | −2.1 tie | Cury 10 | 0.9 / 0.2 | 20.1 / 21.9 | 1.8% |
| PI | 23–28/09 | 1,218 | 68.5 | 24.6 | +43.9 | Cury 2.4 | 0.6 / 0.6 | 17.7 / 18.7 | 1.7% |
| MT single source | 04–09/09 | 1,209 | 31.2 | 53.8 | −22.6 | Cury 5.5 | 0.2 / 0.4 | 23.4 / 22.5 | 1.7% |
| AL single source | 04–09/09 | 1,208 | 47.7 | 37 | +10.7 | Cury 7.1 | 1.3 / 0.6 | 22.4 / 23.2 | 1.5% |
| DF single source | 27/08–01/09 | 1,193 | 36.8 | 42.4 | −5.6 tie | Cury 9.3 | 0.5 / 0.7 | 17.5 / 16.7 | 1.4% |
| MS single source | 27/08–01/09 | 1,204 | 32.6 | 46.6 | −14 | Cury 9.7 | 0.8 / 1 | 22.1 / 22.4 | 1.3% |
| RR | 04–09/09 | 1,200 | 20.7 | 58.6 | −37.9 | Cury 9.2 | 1.2 / 1 | 16.7 / 21.9 | 0.3% |
AtlasIntel · 19 states · fieldwork ended between Sep 1 and Oct 2 · 800 to 2,189 interviews per state · sponsors: A Tarde, Estadão, Focus Comunicação e Mídia, Grupo ND, TV Meio Norte, own funds
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 20/09 | 1,800 | 42.8 | 48.8 | −6 | — | 8.3 / 0 | 21.6 / 21.1 | 21.6% |
| MG | 02/10 | 2,189 | 47.4 | 50.2 | −2.8 tie | — | 2.4 / — | 22.3 / 21.0 | 10.4% |
| RJ | 20/09 | 1,829 | 44.7 | 44.9 | −0.2 tie | — | — / — | 22.7 / 22.2 | 8.1% |
| BA | 28/09 | 2,000 | 61.3 | 31.4 | +29.9 | — | 7.3 / — | 21.4 / 20.5 | 7.2% |
| PR | 16/09 | 1,794 | 38.8 | 57.5 | −18.7 | — | 3.7 / — | 19.5 / 18.6 | 5.5% |
| RS | 20/09 | 1,819 | 44.4 | 47.9 | −3.5 tie | — | — / — | 19.8 / 19.3 | 5.4% |
| PE | 20/09 | 1,836 | 60.5 | 32.7 | +27.8 | — | 6.9 / 0 | 18.2 / 17.3 | 4.6% |
| CE | 02/09 | 1,834 | 54.6 | 30.1 | +24.5 | — | 15.3 / — | 17.4 / 17.1 | 4.4% |
| SC | 09/09 | 1,211 | 27.5 | 61.5 | −34 | — | 11 / — | 18.4 / 17.4 | 3.6% |
| GO | 01/09 | 1,214 | 32.9 | 54.6 | −21.7 | — | 12.5 / — | 21.7 / 20.5 | 3.2% |
| PB | 02/10 | 1,268 | 61.4 | 33.9 | +27.5 | — | 4.7 / — | 17.3 / 16.7 | 2.1% |
| AM | 03/09 | 1,185 | 41.3 | 48 | −6.7 | — | 10.7 / — | 20.1 / 21.9 | 1.8% |
| PI | 28/09 | 1,218 | 70 | 26 | +44 | — | 4 / — | 17.7 / 18.7 | 1.7% |
| MT | 09/09 | 1,209 | 35.2 | 59.1 | −23.9 | — | 5.7 / — | 23.4 / 22.5 | 1.7% |
| AL | 09/09 | 1,208 | 50.6 | 40 | +10.6 | — | 9.4 / — | 22.4 / 23.2 | 1.5% |
| DF | 02/10 | 1,200 | 39.6 | 55.3 | −15.7 | — | 5.1 / — | 17.5 / 16.7 | 1.4% |
| MS | 01/09 | 1,204 | 36.1 | 57.1 | −21 | — | 6.8 / — | 22.1 / 22.4 | 1.3% |
| AP | 02/10 | 800 | 50.7 | 43.5 | +7.2 | — | 5.8 / — | 19.5 / 27.2 | 0.4% |
| RR | 02/10 | 800 | 29.3 | 65.3 | −36 | — | 5.4 / — | 16.7 / 21.9 | 0.3% |
Real Time Big Data · 12 states · fieldwork ended between Sep 9 and Oct 2 · 1,600 to 2,000 interviews per state · sponsors: own funds
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 24–28/09 | 2,000 | 36 | 43 | −7 | Renan 6 | 3 / 3 | 21.6 / 21.1 | 21.6% |
| MG | 28/09–02/10 | 2,000 | 42 | 40 | +2 tie | Renan 4 | 2 / 2 | 22.3 / 21.0 | 10.4% |
| RJ | 25–29/09 | 2,000 | 35 | 44 | −9 | Cury 5 | 3 / 3 | 22.7 / 22.2 | 8.1% |
| BA | 23–26/09 | 1,600 | 61 | 24 | +37 | Cury 3 | 4 / 2 | 21.4 / 20.5 | 7.2% |
| PR | 24–28/09 | 1,600 | 30 | 49 | −19 | Cury 4 | 5 / 3 | 19.5 / 18.6 | 5.5% |
| RS single source | 24–28/09 | 1,600 | 41 | 44 | −3 tie | Cury 4 | 2 / 1 | 19.8 / 19.3 | 5.4% |
| PE | 23–26/09 | 1,600 | 60 | 26 | +34 | Cury 4 | 2 / 2 | 18.2 / 17.3 | 4.6% |
| CE single source | 21–24/09 | 1,600 | 60 | 26 | +34 | Cury 4 | 2 / 2 | 17.4 / 17.1 | 4.4% |
| PA single source | 23–26/09 | 1,600 | 45 | 36 | +9 | Cury 4 | 6 / 2 | 21.2 / 22.6 | 4.0% |
| SC | 12–16/09 | 1,600 | 26 | 52 | −26 | Cury 5 | 2 / 2 | 18.4 / 17.4 | 3.6% |
| GO | 17–21/09 | 1,600 | 26 | 30 | −4 tie | Caiado 31 | 2 / 1 | 21.7 / 20.5 | 3.2% |
| MS single source | 05–09/09 | 1,600 | 33 | 43 | −10 | Renan 5 | 4 / 2 | 22.1 / 22.4 | 1.3% |
Real Time Big Data · 12 states · fieldwork ended between Sep 9 and Oct 2 · 1,600 to 2,000 interviews per state · sponsors: own funds
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 28/09 | 2,000 | 43 | 51 | −8 | — | — / — | 21.6 / 21.1 | 21.6% |
| MG | 02/10 | 2,000 | 46 | 46 | 0 tie | — | 4 / 4 | 22.3 / 21.0 | 10.4% |
| RJ | 29/09 | 2,000 | 41 | 52 | −11 | — | 4 / 3 | 22.7 / 22.2 | 8.1% |
| BA | 26/09 | 1,600 | 64 | 28 | +36 | — | 5 / 3 | 21.4 / 20.5 | 7.2% |
| PR | 28/09 | 1,600 | 30 | 61 | −31 | — | 5 / 4 | 19.5 / 18.6 | 5.5% |
| RS | 28/09 | 1,600 | 43 | 53 | −10 | — | 2 / 2 | 19.8 / 19.3 | 5.4% |
| PE | 26/09 | 1,600 | 61 | 35 | +26 | — | 2 / 2 | 18.2 / 17.3 | 4.6% |
| CE | 24/09 | 1,600 | 63 | 28 | +35 | — | 4 / 5 | 17.4 / 17.1 | 4.4% |
| SC | 16/09 | 1,600 | 30 | 62 | −32 | — | 5 / 3 | 18.4 / 17.4 | 3.6% |
| MA | 09/09 | 1,600 | 65 | 30 | +35 | — | 2 / 3 | 22.2 / 23.5 | 3.3% |
| GO | 21/09 | 1,600 | 33 | 57 | −24 | — | 6 / 4 | 21.7 / 20.5 | 3.2% |
| MS | 09/09 | 1,600 | 38 | 52 | −14 | — | 5 / 5 | 22.1 / 22.4 | 1.3% |
Datafolha · 5 states · fieldwork ended between Sep 30 and Oct 1 · 910 to 1,610 interviews per state · sponsors: Globo and Folha de S.Paulo
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 28–30/09 | 1,610 | 38 | 39 | −1 tie | Cury 5 | 4 / 2 | 21.6 / 21.1 | 21.6% |
| MG | 28/09–01/10 | 1,204 | 42 | 36 | +6 | Caiado 4 | 3 / 4 | 22.3 / 21.0 | 10.4% |
| RJ | 28/09–01/10 | 1,204 | 38 | 45 | −7 | Cury 3 | 4 / 2 | 22.7 / 22.2 | 8.1% |
| PE | 28/09–01/10 | 1,204 | 61 | 25 | +36 | Cury 3 | 4 / 2 | 18.2 / 17.3 | 4.6% |
| DF | 28–30/09 | 910 | 35 | 44 | −9 | Caiado 9 | 3 / 2 | 17.5 / 16.7 | 1.4% |
Datafolha · 5 states · fieldwork ended between Sep 30 and Oct 8 · 910 to 1,610 interviews per state · sponsors: Folha de S.Paulo; TV Globo, Globo and Folha de S.Paulo
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| SP | 30/09 | 1,610 | 45 | 48 | −3 tie | — | 7 / 1 | 21.6 / 21.1 | 21.6% |
| MG | 01/10 | 1,204 | 48 | 44 | +4 tie | — | 7 / 1 | 22.3 / 21.0 | 10.4% |
| RJ | 01/10 | 1,204 | 42 | 51 | −9 | — | 6 / 1 | 22.7 / 22.2 | 8.1% |
| PE | 01/10 | 1,204 | 65 | 29 | +36 | — | 6 / 1 | 18.2 / 17.3 | 4.6% |
| DF | 08/10 | 910 | 43 | 49 | −6 tie | — | 6 / 2 | 17.5 / 16.7 | 1.4% |
DataTempo · 1 state · fieldwork ended between Sep 27 and Sep 27 · 1,000 to 1,000 interviews per state · sponsors: Sempre Editora (O Tempo)
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| MG single source | 24–27/09 | 1,000 | 40.2 | 30.9 | +9.3 | Cury 5.8 | — / — | 22.3 / 21.0 | 10.4% |
no poll in this selection
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate | |
|---|---|---|---|---|---|---|---|---|---|---|
| No poll in this selection. | ||||||||||
no poll in this selection
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate | |
|---|---|---|---|---|---|---|---|---|---|---|
| No poll in this selection. | ||||||||||
Paraná Pesquisas · 1 state · fieldwork ended between Oct 17 and Oct 17 · 1,500 to 1,500 interviews per state · sponsors: FIETO
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| TO | 17/10 | 1,500 | 41.9 | 52.9 | −11 | — | 3.1 / 2.1 | 18.5 / 20.4 | 0.7% |
no poll in this selection
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate | |
|---|---|---|---|---|---|---|---|---|---|---|
| No poll in this selection. | ||||||||||
PoderData · 1 state · fieldwork ended between Oct 11 and Oct 11 · 1,200 to 1,200 interviews per state · sponsors: RealtimeAM
| State | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|
| AM | 11/10 | 1,200 | 49 | 49 | 0 tie | — | 2 / 1 | 20.1 / 21.9 | 1.8% |
AtlasIntel, Datafolha, Quaest, Real Time Big Data · 27 states · fieldwork ended between Sep 21 and Oct 2 · 804 to 2,000 interviews per state · sponsors: A Tarde, Globo and Folha de S.Paulo, Inter TV, NSC, Rede Amazonica, TV Anhanguera, TV Cabo Branco/TV Paraiba, TV Centro America, TV Gazeta, TV Gazeta ES, TV Meio Norte, TV Mirante, TV Morena, TV Sergipe, own funds
| State | Pollster | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|---|
| SP | Datafolha | 28–30/09 | 1,610 | 38 | 39 | −1 tie | Cury 5 | 4 / 2 | 21.6 / 21.1 | 21.6% |
| MG | Real Time Big Data | 28/09–02/10 | 2,000 | 42 | 40 | +2 tie | Renan 4 | 2 / 2 | 22.3 / 21.0 | 10.4% |
| RJ | Datafolha | 28/09–01/10 | 1,204 | 38 | 45 | −7 | Cury 3 | 4 / 2 | 22.7 / 22.2 | 8.1% |
| BA | AtlasIntel | 24–28/09 | 2,000 | 57.6 | 27.8 | +29.8 | Cury 3.9 | 2.1 / — | 21.4 / 20.5 | 7.2% |
| PR | Real Time Big Data | 24–28/09 | 1,600 | 30 | 49 | −19 | Cury 4 | 5 / 3 | 19.5 / 18.6 | 5.5% |
| RS single source | Real Time Big Data | 24–28/09 | 1,600 | 41 | 44 | −3 tie | Cury 4 | 2 / 1 | 19.8 / 19.3 | 5.4% |
| PE | Datafolha | 28/09–01/10 | 1,204 | 61 | 25 | +36 | Cury 3 | 4 / 2 | 18.2 / 17.3 | 4.6% |
| CE single source | Real Time Big Data | 21–24/09 | 1,600 | 60 | 26 | +34 | Cury 4 | 2 / 2 | 17.4 / 17.1 | 4.4% |
| PA single source | Real Time Big Data | 23–26/09 | 1,600 | 45 | 36 | +9 | Cury 4 | 6 / 2 | 21.2 / 22.6 | 4.0% |
| SC | Quaest | 20–23/09 | 804 | 23 | 50 | −27 | Cury 5 | — / — | 18.4 / 17.4 | 3.6% |
| MA | Quaest | 21–24/09 | 900 | 59 | 22 | +37 | — | — / — | 22.2 / 23.5 | 3.3% |
| GO | Quaest | 21–24/09 | 804 | 27 | 33 | −6 tie | Caiado 23 | — / — | 21.7 / 20.5 | 3.2% |
| PB | Quaest | 18–21/09 | 804 | 54 | 23 | +31 | Cury 6 | — / — | 17.3 / 16.7 | 2.1% |
| ES single source | Quaest | 21–24/09 | 804 | 31 | 37 | −6 tie | — | — / 13 | 20.8 / 20.5 | 1.9% |
| AM | Quaest | 20–23/09 | 804 | 38 | 32 | +6 tie | Cury 6 | 6 / 14 | 20.1 / 21.9 | 1.8% |
| PI | AtlasIntel | 23–28/09 | 1,218 | 68.5 | 24.6 | +43.9 | Cury 2.4 | 0.6 / 0.6 | 17.7 / 18.7 | 1.7% |
| RN | Quaest | 21–24/09 | 804 | 52 | 25 | +27 | Cury 6 | — / — | 18.2 / 17.4 | 1.7% |
| MT | Quaest | 21–24/09 | 804 | 25 | 50 | −25 | Cury 5 | 3 / 11 | 23.4 / 22.5 | 1.7% |
| AL | Quaest | 21–24/09 | 804 | 44 | 32 | +12 | Cury 5 | 6 / 8 | 22.4 / 23.2 | 1.5% |
| DF | Datafolha | 28–30/09 | 910 | 35 | 44 | −9 | Caiado 9 | 3 / 2 | 17.5 / 16.7 | 1.4% |
| MS | Quaest | 21–24/09 | 804 | 27 | 42 | −15 | — | — / 13 | 22.1 / 22.4 | 1.3% |
| SE | Quaest | 20–23/09 | 804 | 54 | 26 | +28 | — | — / — | 18.4 / 18.9 | 1.1% |
| RO | Quaest | 20–23/09 | 804 | 19 | 53 | −34 | Cury 4 | 6 / 12 | 24.6 / 24.7 | 0.8% |
| TO | Quaest | 21–24/09 | statistical tie reported without the numbers | 18.5 / 20.4 | 0.7% | |||||
| AC | Quaest | 21–24/09 | 804 | 24 | 50 | −26 | Cury 4 | — / — | 22.4 / 28.4 | 0.4% |
| AP single source | Quaest | 21–24/09 | 804 | 42 | 35 | +7 | — | — / — | 19.5 / 27.2 | 0.4% |
| RR | Quaest | 21–24/09 | 804 | 20 | 59 | −39 | — | — / — | 16.7 / 21.9 | 0.3% |
AtlasIntel, Datafolha, Paraná Pesquisas, Quaest, Real Time Big Data, Veritá · 27 states · fieldwork ended between Sep 16 and Oct 20 · 800 to 2,000 interviews per state · sponsors: A Tarde, FIETO, Globo and Folha de S.Paulo, TV Centro America, TV Gazeta, TV Globo, TV Liberal, TV Meio Norte, TV Mirante, TV Sergipe, self-funded, own funds
| State | Pollster | Fieldwork | n | Lula | Flávio | Difference | 3rd place | Blank/spoiled / undecided | 2022 abstention (1st / 2nd round) | Electorate |
|---|---|---|---|---|---|---|---|---|---|---|
| SP | Datafolha | 30/09 | 1,610 | 45 | 48 | −3 tie | — | 7 / 1 | 21.6 / 21.1 | 21.6% |
| MG | Real Time Big Data | 02/10 | 2,000 | 46 | 46 | 0 tie | — | 4 / 4 | 22.3 / 21.0 | 10.4% |
| RJ | Datafolha | 01/10 | 1,204 | 42 | 51 | −9 | — | 6 / 1 | 22.7 / 22.2 | 8.1% |
| BA | AtlasIntel | 28/09 | 2,000 | 61.3 | 31.4 | +29.9 | — | 7.3 / — | 21.4 / 20.5 | 7.2% |
| PR | Real Time Big Data | 28/09 | 1,600 | 30 | 61 | −31 | — | 5 / 4 | 19.5 / 18.6 | 5.5% |
| RS | Real Time Big Data | 28/09 | 1,600 | 43 | 53 | −10 | — | 2 / 2 | 19.8 / 19.3 | 5.4% |
| PE | Datafolha | 01/10 | 1,204 | 65 | 29 | +36 | — | 6 / 1 | 18.2 / 17.3 | 4.6% |
| CE | Real Time Big Data | 24/09 | 1,600 | 63 | 28 | +35 | — | 4 / 5 | 17.4 / 17.1 | 4.4% |
| PA | Quaest | 25/09 | 804 | 44 | 40 | +4 tie | — | — / — | 21.2 / 22.6 | 4.0% |
| SC | Real Time Big Data | 16/09 | 1,600 | 30 | 62 | −32 | — | 5 / 3 | 18.4 / 17.4 | 3.6% |
| MA | Quaest | 24/09 | 900 | 63 | 26 | +37 | — | 8 / 3 | 22.2 / 23.5 | 3.3% |
| GO | Real Time Big Data | 21/09 | 1,600 | 33 | 57 | −24 | — | 6 / 4 | 21.7 / 20.5 | 3.2% |
| PB | AtlasIntel | 02/10 | 1,268 | 61.4 | 33.9 | +27.5 | — | 4.7 / — | 17.3 / 16.7 | 2.1% |
| ES | Quaest | 13/10 | 1,098 | 38 | 51 | −13 | — | 6 / 5 | 20.8 / 20.5 | 1.9% |
| AM | Quaest | 12/10 | 1,098 | 47 | 43 | +4 tie | — | 5 / 5 | 20.1 / 21.9 | 1.8% |
| PI | AtlasIntel | 28/09 | 1,218 | 70 | 26 | +44 | — | 4 / — | 17.7 / 18.7 | 1.7% |
| RN | Quaest | 12/10 | 1,098 | 59 | 33 | +26 | — | 5 / 3 | 18.2 / 17.4 | 1.7% |
| MT | Quaest | 24/09 | 804 | 30 | 56 | −26 | — | — / — | 23.4 / 22.5 | 1.7% |
| AL | Quaest | 24/09 | 804 | 48 | 34 | +14 | — | — / — | 22.4 / 23.2 | 1.5% |
| DF | Quaest | 20/10 | 1,104 | 39 | 50 | −11 | — | 6 / 5 | 17.5 / 16.7 | 1.4% |
| MS | Veritá | 17/09 | 1,220 | 34.4 | 53.5 | −19.1 | — | 12.1 / — | 22.1 / 22.4 | 1.3% |
| SE | Quaest | 23/09 | 804 | 56 | 30 | +26 | — | — / — | 18.4 / 18.9 | 1.1% |
| RO | Veritá | 19/09 | 1,220 | 30.5 | 63.9 | −33.4 | — | 5.6 / — | 24.6 / 24.7 | 0.8% |
| TO | Paraná Pesquisas | 17/10 | 1,500 | 41.9 | 52.9 | −11 | — | 3.1 / 2.1 | 18.5 / 20.4 | 0.7% |
| AC | Quaest | 12/10 | 1,098 | 28 | 60 | −32 | — | 5 / 7 | 22.4 / 28.4 | 0.4% |
| AP | AtlasIntel | 02/10 | 800 | 50.7 | 43.5 | +7.2 | — | 5.8 / — | 19.5 / 27.2 | 0.4% |
| RR | AtlasIntel | 02/10 | 800 | 29.3 | 65.3 | −36 | — | 5.4 / — | 16.7 / 21.9 | 0.3% |
Each figure was checked against at least two outlets, except those marked as single source. The Quaest round (Globo and affiliates, in person, 804 to 1,800 interviews per state) had fieldwork from Sep 18 to 28; SP, MG, RJ, PE and DF were in a second wave, Sep 25 to 28. A Band compilation from Oct 1 mixed August and September waves and was not used. "—" = figure not published in the article consulted. The 2022 abstention comes from the TSE (via CNN Brasil) and serves as a reference for turnout; state polls do not measure abstention, only those who say they will vote blank, spoil their ballot or not show up.
In one sentence: adding up the states with recent polls (8% of the electorate), the estimated national margin is +3.6 points for Lula, versus -0.9 in the national polls — the state reading is 4.5 points more favorable to Lula. It does not enter the simulation yet (coverage 8% < 30%).
| State | 2022 margin | 2026 margin (adjusted) | Raw swing | Adjusted swing | Share of electorate | Polls | Pollsters · fieldwork |
|---|---|---|---|---|---|---|---|
| ES | -16.1 | -12.8 | +3.3 | +2.8 | 1.9% | 1 | Quaest · through Oct 13 |
| AM | +2.2 | +5.0 | +2.8 | +2.1 | 1.8% | 2 | PoderData, Quaest · through Oct 12 |
| RN | +30.2 | +30.3 | +0.1 | +0.5 | 1.7% | 1 | Quaest · through Oct 12 |
| DF | -17.6 | -8.7 | +8.8 | +7.8 | 1.4% | 3 | Datafolha, Quaest · through Oct 20 |
| TO | +2.7 | -8.0 | -11.0 | -8.2 | 0.7% | 2 | Paraná Pesquisas, Quaest · through Oct 17 |
| AC | -40.6 | -34.4 | +6.2 | +3.8 | 0.4% | 1 | Quaest · through Oct 12 |
Margins in points of valid votes (Lula − Flávio; in 2022, Lula − Bolsonaro, official TSE result by state). Weights by the 2026 registered electorate. Included are state runoff polls of Lula × Flávio with fieldwork in the last 45 days; 13 of the 98 are single-source (the rest were confirmed in two sources). Before the calculation, each state poll gets two adjustments: the pollster's national house effect is subtracted (today, on average, +0.9 point) and the amount the national margin moved between the end of fieldwork and today is added (on average, +0.9 point — the Datafolha waves are from August, when Lula was doing better). Shrinkage works on two levels, state → region → country: states without polls receive their region's average (Northeast +1.2, Southeast +2.1, South 0.0). State polls have larger error than sampling error alone, so the model adds 3 pp of error per poll and 1.5 pp of error common to all of them.
The municipality map (2.2 MB of drawing) loads separately, as soon as this part of the page approaches the screen.
| Municipality | Voters | 2022 margin | 2026 margin | Change |
|---|---|---|---|---|
| Osasco SP | 583,991 | +0.8 | −2.1 | −3.2 |
| Ribeirão das Neves MG | 218,584 | +1.3 | −2.6 | −4.1 |
| Cotia SP | 194,159 | +1.7 | −4.4 | −6.7 |
| Araraquara SP | 170,310 | +1.0 | −1.0 | −2.1 |
| Teixeira de Freitas BA | 105,045 | +0.7 | −4.1 | −5.1 |
| Sapucaia do Sul RS | 102,445 | +2.3 | −3.5 | −6.2 |
| Sabará MG | 94,323 | +1.2 | −1.6 | −3.0 |
| Bagé RS | 87,802 | +7.4 | −0.4 | −7.9 |
| Nova Lima MG | 77,107 | +2.3 | −3.5 | −6.2 |
| Jandira SP | 76,022 | +1.6 | −1.7 | −3.6 |
| Municipality | Voters | 2022 margin | 2026 margin | Change |
|---|---|---|---|---|
| Jequiá da Praia AL | 8,245 | −0.4 | +2.2 | +2.7 |
Margin = Lula minus the opponent, in points of valid votes. The two elections are not alike: 2022 is the runoff, with two candidates; 2026 is the first round, with seven, and the votes for Cury, Renan, Caiado and Zema come from both sides. For the change and for the color intensity of 2026, the 2026 margin is computed between Lula and Flávio only (Lula − Flávio over Lula + Flávio), comparable to 2022's; it does not say where the other candidates' votes go in the runoff. In the table, "Change" follows the same calculation. Boa Esperança do Norte/MT was created after 2022 and is not yet in the IBGE boundary file used here: it counts in the 2026 tallies but does not appear on the map; 4 municipalities tied in 2022 or in 2026 and have no arrow (Crixás do Tocantins/TO, Guará/SP, Trabiju/SP, Irati/SC). Sources: TSE (2026 results release, election 6257, files by municipality; Open Data, 2022 vote by municipality and zone) and IBGE (municipal boundaries, minimum quality).
#exterior- and its name to the page address, as in #exterior-Lisboa.The map (0.5 MB of drawing and data) loads separately, as soon as this part of the page approaches the screen.
| City | Eligible | Abst. % | Absent | Lula % | Flávio % | Third way % | Blank+spoiled % | Margin | Change since 2022 |
|---|---|---|---|---|---|---|---|---|---|
| Portugal | 78,607 | 63.6 | 49,964 | 53.5 | 37.8 | 8.7 | 2.5 | +15.8 | −11.7 |
| United States | 53,358 | 69.3 | 36,968 | 24.4 | 69.8 | 5.9 | 1.8 | −45.4 | +3.3 |
| Portugal | 47,496 | 57.3 | 27,217 | 52.5 | 38.3 | 9.2 | 2.7 | +14.2 | −14.0 |
| Japan | 44,536 | 63.4 | 28,219 | 13.1 | 74.5 | 12.4 | 5.6 | −61.4 | +1.4 |
| United States | 40,053 | 69.3 | 27,775 | 18.8 | 74.2 | 7.0 | 2.1 | −55.4 | +2.8 |
| United States | 38,953 | 74.0 | 28,819 | 46.2 | 47.4 | 6.4 | 2.4 | −1.2 | +5.2 |
| United Kingdom | 37,448 | 60.5 | 22,651 | 56.3 | 36.1 | 7.6 | 2.2 | +20.2 | +1.6 |
| Japan | 30,028 | 55.4 | 16,636 | 18.8 | 67.9 | 13.3 | 7.4 | −49.2 | +4.0 |
| France | 26,957 | 60.9 | 16,417 | 77.9 | 16.0 | 6.1 | 1.9 | +61.9 | +0.0 |
| United States | 26,386 | 65.9 | 17,392 | 15.0 | 78.7 | 6.3 | 1.4 | −63.7 | — |
| Italy | 24,594 | 69.7 | 17,139 | 50.2 | 40.6 | 9.2 | 3.9 | +9.6 | +5.4 |
| Spain | 23,464 | 70.9 | 16,631 | 54.5 | 35.7 | 9.8 | 4.1 | +18.7 | +8.6 |
| Canada | 20,953 | 57.1 | 11,954 | 53.7 | 38.2 | 8.1 | 3.4 | +15.5 | +5.3 |
| United States | 18,262 | 74.8 | 13,653 | 47.4 | 44.4 | 8.2 | 2.5 | +3.0 | +4.8 |
| United States | 18,146 | 73.3 | 13,310 | 51.4 | 39.5 | 9.1 | 2.7 | +11.9 | +12.0 |
| Spain | 17,254 | 69.9 | 12,069 | 73.2 | 19.7 | 7.1 | 2.7 | +53.5 | +6.5 |
| Ireland | 16,720 | 49.7 | 8,303 | 72.6 | 18.8 | 8.6 | 1.4 | +53.8 | −4.7 |
| United States | 16,438 | 72.6 | 11,938 | 29.3 | 59.8 | 10.9 | 4.0 | −30.6 | −3.6 |
| United States | 16,301 | 69.2 | 11,279 | 24.9 | 66.8 | 8.3 | 2.4 | −41.9 | +2.5 |
| United States | 15,492 | 74.9 | 11,596 | 57.5 | 35.0 | 7.5 | 2.9 | +22.6 | +2.5 |
| Switzerland | 15,099 | 60.3 | 9,098 | 50.0 | 39.7 | 10.3 | 3.5 | +10.3 | +4.3 |
| Germany | 15,066 | 45.5 | 6,862 | 81.1 | 11.5 | 7.4 | 2.3 | +69.6 | +3.1 |
| Italy | 14,153 | 67.4 | 9,537 | 57.6 | 33.3 | 9.0 | 4.0 | +24.3 | +7.2 |
| Australia | 14,031 | 59.4 | 8,337 | 55.7 | 32.0 | 12.2 | 2.8 | +23.7 | +2.7 |
| Japan | 13,729 | 49.6 | 6,811 | 13.7 | 72.6 | 13.8 | 6.4 | −58.9 | +2.5 |
| Germany | 13,374 | 59.5 | 7,951 | 63.6 | 24.3 | 12.0 | 2.5 | +39.3 | +5.9 |
| Germany | 12,961 | 56.6 | 7,342 | 66.6 | 24.0 | 9.4 | 2.2 | +42.5 | +3.9 |
| Netherlands | 12,752 | 49.4 | 6,302 | 73.7 | 17.2 | 9.1 | 2.7 | +56.5 | +5.3 |
| United States | 12,452 | 77.7 | 9,673 | 55.1 | 35.4 | 9.5 | 3.1 | +19.7 | +10.0 |
| Canada | 12,082 | 51.1 | 6,170 | 56.4 | 32.8 | 10.8 | 3.4 | +23.6 | −0.5 |
| Switzerland | 11,883 | 55.9 | 6,646 | 45.4 | 44.0 | 10.6 | 3.3 | +1.4 | +0.7 |
| Argentina | 11,748 | 53.7 | 6,313 | 66.1 | 27.3 | 6.6 | 2.3 | +38.8 | +9.3 |
| Canada | 10,953 | 44.3 | 4,852 | 64.9 | 24.7 | 10.4 | 3.5 | +40.2 | +6.9 |
| Belgium | 10,073 | 49.5 | 4,982 | 50.6 | 39.4 | 10.1 | 2.3 | +11.2 | −6.4 |
| United States | 9,181 | 78.7 | 7,230 | 29.4 | 63.7 | 6.9 | 2.9 | −34.3 | −8.9 |
| Lebanon | 8,919 | 54.1 | 4,822 | 65.4 | 29.3 | 5.3 | 5.2 | +36.1 | +19.8 |
| Portugal | 8,271 | 66.6 | 5,510 | 41.1 | 48.9 | 10.0 | 3.3 | −7.8 | −12.7 |
| Australia | 6,243 | 56.7 | 3,538 | 57.0 | 31.0 | 12.0 | 4.1 | +26.0 | +1.3 |
| Paraguay | 5,094 | 49.2 | 2,508 | 8.1 | 89.5 | 2.5 | 3.2 | −81.4 | +0.5 |
| Sweden | 4,866 | 46.4 | 2,259 | 69.7 | 19.9 | 10.4 | 2.6 | +49.8 | +2.2 |
| United Kingdom | 4,447 | 84.4 | 3,753 | 72.1 | 20.5 | 7.4 | 2.4 | +51.6 | — |
| Mexico | 3,958 | 66.8 | 2,643 | 45.5 | 45.8 | 8.6 | 2.3 | −0.3 | +4.4 |
| Chile | 3,551 | 41.7 | 1,482 | 47.1 | 41.0 | 11.9 | 4.7 | +6.0 | +7.8 |
| Israel | 3,442 | 78.8 | 2,712 | 13.1 | 74.3 | 12.6 | 3.6 | −61.2 | −63.2 |
| Austria | 3,134 | 51.4 | 1,612 | 66.6 | 21.1 | 12.3 | 3.4 | +45.4 | +8.4 |
| New Zealand | 3,118 | 89.1 | 2,779 | 70.5 | 19.3 | 10.2 | 2.1 | +51.2 | +16.4 |
| French Guiana | 3,117 | 63.0 | 1,965 | 44.6 | 47.8 | 7.6 | 5.9 | −3.1 | +6.0 |
| United Arab Emirates | 2,988 | 53.3 | 1,592 | 37.8 | 48.6 | 13.6 | 2.9 | −10.8 | +6.4 |
| Denmark | 2,856 | 46.7 | 1,334 | 79.0 | 13.2 | 7.8 | 3.0 | +65.8 | +3.2 |
| Bolivia | 2,765 | 46.7 | 1,292 | 13.6 | 81.4 | 5.0 | 3.1 | −67.8 | +1.2 |
| France | 2,660 | 67.1 | 1,785 | 82.4 | 12.0 | 5.6 | 0.8 | +70.4 | — |
| Uruguay | 2,448 | 45.1 | 1,105 | 54.3 | 36.8 | 8.9 | 6.0 | +17.5 | +3.2 |
| Canada | 2,357 | 39.2 | 923 | 58.3 | 29.0 | 12.6 | 5.1 | +29.3 | +5.4 |
| Paraguay | 2,205 | 47.8 | 1,054 | 23.6 | 69.7 | 6.7 | 7.8 | −46.2 | −0.7 |
| Norway | 2,185 | 58.4 | 1,276 | 67.0 | 22.4 | 10.6 | 2.8 | +44.6 | −2.4 |
| Palestine | 2,135 | 43.7 | 933 | 93.2 | 4.0 | 2.9 | 3.7 | +89.2 | +10.7 |
| Peru | 1,616 | 53.0 | 856 | 34.0 | 52.1 | 14.0 | 3.9 | −18.1 | −5.4 |
| Colombia | 1,499 | 62.9 | 943 | 46.5 | 47.8 | 5.7 | 5.2 | −1.3 | −14.3 |
| Jordan | 1,489 | 44.1 | 656 | 86.2 | 7.5 | 6.3 | 6.0 | +78.7 | +30.9 |
| Suriname | 1,430 | 60.8 | 870 | 22.4 | 67.8 | 9.8 | 3.6 | −45.4 | −4.3 |
| Panama | 1,399 | 51.6 | 722 | 35.9 | 51.1 | 13.1 | 2.8 | −15.2 | +4.3 |
| Finland | 1,278 | 43.8 | 560 | 68.5 | 22.4 | 9.1 | 3.5 | +46.2 | +4.5 |
| Greece | 1,211 | 58.1 | 703 | 39.7 | 46.6 | 13.8 | 2.8 | −6.9 | +3.3 |
| Venezuela | 994 | 73.4 | 730 | 48.9 | 47.7 | 3.4 | 0.8 | +1.1 | — |
| Ecuador | 993 | 65.5 | 650 | 39.6 | 52.1 | 8.2 | 4.4 | −12.5 | +24.2 |
| Argentina | 918 | 59.0 | 542 | 49.6 | 43.3 | 7.1 | 2.4 | +6.3 | +1.2 |
| Paraguay | 899 | 40.3 | 362 | 3.6 | 94.5 | 1.9 | 2.2 | −90.9 | −2.7 |
| Singapore | 864 | 57.9 | 500 | 54.6 | 26.4 | 19.0 | 4.4 | +28.2 | +7.3 |
| Hungary | 810 | 41.9 | 339 | 71.3 | 20.0 | 8.6 | 1.5 | +51.3 | −14.4 |
| Syria | 775 | 84.8 | 657 | 89.8 | 6.5 | 3.7 | 8.5 | +83.3 | — |
| Qatar | 758 | 52.0 | 394 | 41.2 | 48.5 | 10.4 | 1.9 | −7.3 | +10.7 |
| Poland | 748 | 61.1 | 457 | 64.9 | 23.0 | 12.1 | 3.1 | +41.8 | −3.8 |
| China | 739 | 73.3 | 542 | 64.1 | 22.7 | 13.3 | 8.1 | +41.4 | +14.1 |
| Angola | 716 | 62.3 | 446 | 55.7 | 38.3 | 6.1 | 2.2 | +17.4 | +7.3 |
| Costa Rica | 714 | 47.3 | 338 | 46.8 | 43.0 | 10.3 | 1.6 | +3.8 | +13.5 |
| Czechia | 690 | 40.4 | 279 | 70.3 | 17.6 | 12.1 | 3.4 | +52.6 | +0.6 |
| China | 668 | 60.5 | 404 | 40.6 | 43.0 | 16.4 | 7.6 | −2.5 | +5.6 |
| South Africa | 657 | 59.7 | 392 | 23.9 | 66.4 | 9.7 | 2.3 | −42.5 | −4.2 |
| Paraguay | 650 | 50.3 | 327 | 20.0 | 74.2 | 5.8 | 8.7 | −54.2 | −12.3 |
| Taiwan | 622 | 62.9 | 391 | 35.3 | 49.1 | 15.6 | 5.6 | −13.8 | −3.0 |
| Uruguay | 611 | 37.0 | 226 | 28.3 | 61.9 | 9.8 | 4.7 | −33.5 | +2.9 |
| Mozambique | 557 | 59.1 | 329 | 41.6 | 52.8 | 5.6 | 6.1 | −11.2 | −3.4 |
| South Korea | 485 | 53.4 | 259 | 56.9 | 33.3 | 9.7 | 4.4 | +23.6 | −2.4 |
| China | 419 | 76.4 | 320 | 52.2 | 38.0 | 9.8 | 7.1 | +14.1 | +54.3 |
| Bolivia | 418 | 40.0 | 167 | 43.8 | 47.3 | 8.9 | 10.8 | −3.6 | +22.1 |
| Dominican Republic | 415 | 52.5 | 218 | 34.6 | 56.4 | 9.0 | 4.6 | −21.8 | +2.6 |
| Bolivia | 412 | 54.6 | 225 | 30.8 | 64.3 | 4.9 | 1.1 | −33.5 | −10.0 |
| South Africa | 355 | 52.1 | 185 | 56.2 | 36.2 | 7.5 | 5.9 | +20.0 | −1.8 |
| China | 342 | 60.2 | 206 | 77.6 | 16.4 | 6.0 | 1.5 | +61.2 | +30.8 |
| Turkey | 327 | 60.6 | 198 | 70.9 | 23.6 | 5.5 | 1.6 | +47.2 | +23.3 |
| Estonia | 320 | 38.4 | 123 | 74.7 | 10.0 | 15.3 | 3.6 | +64.7 | +2.1 |
| Saudi Arabia | 311 | 73.0 | 227 | 62.7 | 26.5 | 10.8 | 1.2 | +36.1 | +4.8 |
| Thailand | 310 | 67.1 | 208 | 40.4 | 51.5 | 8.1 | 2.9 | −11.1 | −9.0 |
| Guatemala | 284 | 57.7 | 164 | 28.0 | 56.8 | 15.3 | 1.7 | −28.8 | +17.0 |
| Paraguay | 281 | 47.0 | 132 | 9.9 | 86.5 | 3.5 | 5.4 | −76.6 | −2.8 |
| Russia | 252 | 66.3 | 167 | 60.5 | 30.9 | 8.6 | 4.7 | +29.6 | −0.9 |
| Argentina | 235 | 50.6 | 119 | 43.8 | 49.1 | 7.1 | 3.4 | −5.4 | +16.5 |
| Philippines | 229 | 72.5 | 166 | 41.9 | 53.2 | 4.8 | 1.6 | −11.3 | +9.6 |
| Egypt | 203 | 66.0 | 134 | 71.9 | 18.8 | 9.4 | 7.2 | +53.1 | +25.3 |
| Guyana | 178 | 70.2 | 125 | 13.2 | 81.1 | 5.7 | 0.0 | −67.9 | +3.8 |
| Indonesia | 172 | 82.0 | 141 | 25.8 | 61.3 | 12.9 | 0.0 | −35.5 | −4.6 |
| Malaysia | 170 | 51.2 | 87 | 53.7 | 32.9 | 13.4 | 1.2 | +20.7 | +13.6 |
| Slovenia | 162 | 36.4 | 59 | 62.2 | 25.5 | 12.2 | 4.9 | +36.7 | −0.6 |
| El Salvador | 161 | 47.8 | 77 | 29.9 | 62.3 | 7.8 | 8.3 | −32.5 | −11.3 |
| Honduras | 152 | 61.2 | 93 | 28.8 | 57.6 | 13.6 | 0.0 | −28.8 | −9.6 |
| Cape Verde | 146 | 49.3 | 72 | 55.7 | 28.6 | 15.7 | 5.4 | +27.1 | +3.2 |
| India | 141 | 64.5 | 91 | 48.0 | 52.0 | 0.0 | 0.0 | −4.0 | −28.3 |
| Kuwait | 138 | 44.2 | 61 | 74.0 | 16.4 | 9.6 | 5.2 | +57.5 | +112.4 |
| Cyprus | 137 | 57.7 | 79 | 42.1 | 35.1 | 22.8 | 1.7 | +7.0 | −15.9 |
| Croatia | 136 | 61.8 | 84 | 60.4 | 35.4 | 4.2 | 7.7 | +25.0 | −0.7 |
| Nicaragua | 134 | 44.0 | 59 | 47.8 | 50.7 | 1.4 | 8.0 | −2.9 | +2.6 |
| Senegal | 132 | 39.4 | 52 | 64.6 | 30.4 | 5.1 | 1.2 | +34.2 | +19.6 |
| Romania | 130 | 52.3 | 68 | 50.0 | 35.5 | 14.5 | 0.0 | +14.5 | −16.4 |
| Slovakia | 129 | 52.7 | 68 | 60.3 | 22.4 | 17.2 | 4.9 | +37.9 | +7.1 |
| Morocco | 117 | 59.8 | 70 | 71.4 | 14.3 | 14.3 | 10.6 | +57.1 | +1.4 |
| Serbia | 115 | 58.3 | 67 | 62.5 | 27.1 | 10.4 | 0.0 | +35.4 | −10.5 |
| Kenya | 112 | 58.0 | 65 | 56.5 | 37.0 | 6.5 | 2.1 | +19.6 | +4.3 |
| Jamaica | 103 | 81.6 | 84 | 21.1 | 73.7 | 5.3 | 0.0 | −52.6 | −60.3 |
| Bulgaria | 98 | 50.0 | 49 | 72.3 | 17.0 | 10.6 | 4.1 | +55.3 | +7.1 |
| Oman | 97 | 60.8 | 59 | 33.3 | 55.6 | 11.1 | 5.3 | −22.2 | +14.1 |
| Cuba | 92 | 56.5 | 52 | 94.7 | 2.6 | 2.6 | 5.0 | +92.1 | — |
| Côte d'Ivoire | 91 | 51.6 | 47 | 54.5 | 29.5 | 15.9 | 0.0 | +25.0 | +2.9 |
| Timor-Leste | 89 | 47.2 | 42 | 34.8 | 58.7 | 6.5 | 2.1 | −23.9 | −0.2 |
| Guinea-Bissau | 88 | 58.0 | 51 | 45.9 | 45.9 | 8.1 | 0.0 | 0.0 | +11.1 |
| Trinidad and Tobago | 87 | 78.2 | 68 | 21.1 | 52.6 | 26.3 | 0.0 | −31.6 | −46.1 |
| Vietnam | 82 | 61.0 | 50 | 44.8 | 48.3 | 6.9 | 9.4 | −3.4 | −40.5 |
| Nigeria | 82 | 59.8 | 49 | 81.2 | 18.8 | 0.0 | 3.0 | +62.5 | −5.5 |
| Paraguay | 78 | 53.8 | 42 | 5.7 | 91.4 | 2.9 | 2.8 | −85.7 | −5.4 |
| Turkey | 77 | 51.9 | 40 | 64.9 | 18.9 | 16.2 | 0.0 | +45.9 | +12.0 |
| Venezuela | 72 | 100.0 | 72 | — | — | — | — | — | — |
| Bahrain | 69 | 29.0 | 20 | 36.2 | 57.4 | 6.4 | 4.1 | −21.3 | +16.2 |
| Bahamas | 69 | 72.5 | 50 | 31.6 | 52.6 | 15.8 | 0.0 | −21.1 | −3.3 |
| Ukraine | 65 | 64.6 | 42 | 17.4 | 82.6 | 0.0 | 0.0 | −65.2 | — |
| Argentina | 62 | 64.5 | 40 | 31.8 | 68.2 | 0.0 | 0.0 | −36.4 | — |
| Uruguay | 58 | 60.3 | 35 | 33.3 | 42.9 | 23.8 | 8.7 | −9.5 | −30.7 |
| India | 55 | 80.0 | 44 | 45.5 | 9.1 | 45.5 | 0.0 | +36.4 | +20.0 |
| Algeria | 48 | 35.4 | 17 | 87.1 | 9.7 | 3.2 | 0.0 | +77.4 | — |
| Tunisia | 47 | 53.2 | 25 | 54.5 | 40.9 | 4.5 | 0.0 | +13.6 | −19.0 |
| Iran | 45 | 44.4 | 20 | 56.0 | 44.0 | 0.0 | 0.0 | +12.0 | +17.9 |
| Ghana | 43 | 69.8 | 30 | 61.5 | 15.4 | 23.1 | 0.0 | +46.2 | +4.4 |
| Argentina | 41 | 87.8 | 36 | 40.0 | 60.0 | 0.0 | 0.0 | −20.0 | — |
| Namibia | 37 | 64.9 | 24 | 46.2 | 38.5 | 15.4 | 0.0 | +7.7 | +86.2 |
| Tanzania | 34 | 52.9 | 18 | 68.8 | 31.2 | 0.0 | 0.0 | +37.5 | −12.5 |
| Haiti | 33 | 84.8 | 28 | 100.0 | 0.0 | 0.0 | 0.0 | +100.0 | +120.0 |
| Nepal | 32 | 59.4 | 19 | 7.7 | 84.6 | 7.7 | 0.0 | −76.9 | −10.0 |
| Libya | 30 | 93.3 | 28 | 100.0 | 0.0 | 0.0 | 0.0 | +100.0 | — |
| Georgia | 29 | 0.0 | 0 | — | — | — | — | — | — |
| Democratic Republic of the Congo | 27 | 0.0 | 0 | — | — | — | — | — | — |
| Uruguay | 27 | 0.0 | 0 | — | — | — | — | — | — |
| Zambia | 25 | 0.0 | 0 | — | — | — | — | — | — |
| Barbados | 22 | 0.0 | 0 | — | — | — | — | — | — |
| Uruguay | 19 | 0.0 | 0 | — | — | — | — | — | — |
| São Tomé and Príncipe | 18 | 0.0 | 0 | — | — | — | — | — | — |
| Gabon | 18 | 0.0 | 0 | — | — | — | — | — | — |
| Pakistan | 18 | 0.0 | 0 | — | — | — | — | — | — |
| Peru | 17 | 0.0 | 0 | — | — | — | — | — | — |
| Cameroon | 14 | 0.0 | 0 | — | — | — | — | — | — |
| Armenia | 13 | 0.0 | 0 | — | — | — | — | — | — |
| Belize | 13 | 0.0 | 0 | — | — | — | — | — | — |
| Kazakhstan | 13 | 0.0 | 0 | — | — | — | — | — | — |
| Sri Lanka | 12 | 0.0 | 0 | — | — | — | — | — | — |
| Bosnia and Herzegovina | 11 | 0.0 | 0 | — | — | — | — | — | — |
| Iraq | 10 | 0.0 | 0 | — | — | — | — | — | — |
| Botswana | 10 | 0.0 | 0 | — | — | — | — | — | — |
| French Guiana | 10 | 0.0 | 0 | — | — | — | — | — | — |
| Albania | 9 | 0.0 | 0 | — | — | — | — | — | — |
| Benin | 8 | 0.0 | 0 | — | — | — | — | — | — |
| Bolivia | 8 | 0.0 | 0 | — | — | — | — | — | — |
| Saint Lucia | 7 | 0.0 | 0 | — | — | — | — | — | — |
| Togo | 7 | 0.0 | 0 | — | — | — | — | — | — |
| Zimbabwe | 7 | 0.0 | 0 | — | — | — | — | — | — |
| Burkina Faso | 6 | 0.0 | 0 | — | — | — | — | — | — |
| Myanmar | 6 | 0.0 | 0 | — | — | — | — | — | — |
| Malawi | 6 | 0.0 | 0 | — | — | — | — | — | — |
| Equatorial Guinea | 5 | 0.0 | 0 | — | — | — | — | — | — |
| Azerbaijan | 5 | 0.0 | 0 | — | — | — | — | — | — |
| Guinea | 5 | 0.0 | 0 | — | — | — | — | — | — |
| Mali | 5 | 0.0 | 0 | — | — | — | — | — | — |
| Ethiopia | 3 | 0.0 | 0 | — | — | — | — | — | — |
| Nigeria | 3 | 0.0 | 0 | — | — | — | — | — | — |
| Republic of the Congo | 2 | 0.0 | 0 | — | — | — | — | — | — |
| North Korea | 1 | 0.0 | 0 | — | — | — | — | — | — |
| Bangladesh | 1 | 0.0 | 0 | — | — | — | — | — | — |
| Venezuela | 1 | 0.0 | 0 | — | — | — | — | — | — |
| Bolivia | 1 | 0.0 | 0 | — | — | — | — | — | — |
| Antigua and Barbuda | 1 | 0.0 | 0 | — | — | — | — | — | — |
Click a column header to sort and a city name to see it on the map. Lula %, Flávio % and Third way % are shares of valid votes; blank and spoiled, of total votes. Margin = Lula − Flávio, in points of valid votes. Change = the two-way margin (Lula − Flávio over Lula + Flávio) minus Lula's margin over Bolsonaro in the 2022 runoff, in the same city.
Voters abroad vote only for president. The figures are from the 2026 first round (341,766 votes; 41 polling stations were not set up). Third way = Cury, Renan, Caiado, Zema and the five other candidates. The sites are the runoff ones, with the result the same polling stations produced in the first round; each site's position comes from geocoding its official address in OpenStreetMap and may be off by a few blocks; when in doubt, the address prevails. Sources: TSE (Open Data: 2026 vote detail and vote by polling station, 2026 electorate by polling site, 2022 vote by municipality and zone), Natural Earth (country outlines) and OpenStreetMap (street map and, via Nominatim, site coordinates).
The youngest are not that generation. Since the PT's opponent became bolsonarismo, the 16-to-24 group has sat to the left of the country: −5 point(s) in 2018, −5 in 2022 and −2 in 2026 (average of Datafolha's closing-stretch waves). With the PSDB as the opponent, in 2010 and 2014, it was slightly more anti-PT than average (+2 and +2).
The right-leaning generation is the young adults. In 2026, those aged 25 to 34 are +3 and those aged 35 to 44 are +2 relative to the country; in 2018 and 2022 these groups were practically at the average, a point or two above (+1 and +1; +1 and +2). They were born, roughly, between 1982 and 2001, many of them reaching adult political life between the 2013 protests and the 2018 election, and they make up 39% of the electorate. At the other end, those aged 60 or older went from +1 in 2018 to −4 in 2026. The right-wing vote no longer rises with age: it has become an arc, peaking at ages 25 to 44.
The same cohort, getting older. Those who were 16 to 24 in 2018 (born around 1994–2002) were −5 in 2018, −2 in 2022 and +2 in 2026: they moved right as they aged. The 2010 cohort, at the same ages, barely moved (+2, +1, 0 from 2010 to 2018). With five elections and wide age brackets, it is impossible to separate generation from moment, but the movement is concentrated at these ages, not at all of them at once, as was the jump from 2014 to 2018.
Among the youngest, the growing right is not Flávio's. In the first round, Renan Santos (Missão) had 8% among 16-to-24-year-olds in Datafolha's eve-of-election poll, against 3% overall, and 23% against 5% in AtlasIntel. In April, AtlasIntel had Flávio ahead among 16-to-24-year-olds; on the eve, Lula. Pollsters also diverge: Quaest, which groups 16 to 34, puts that bracket to the right of the country. It is there, among the youngest, that the race is open, and the right-wing vote outside bolsonarismo does not show up in the simulated runoff.
Social profile. The vote to the right of the country in 2026 is among men (+5), evangelicals (+17), those with a high-school (+4) or college (+5) education, family income above 2 minimum wages (+8 in the 2-to-5 bracket) and in the South (+13); on the other side, the Northeast (−15), elementary education (−12) and income up to 2 minimum wages (−9). Taken together, the breakdowns sketch a middle-income, educated, evangelical voter outside the Northeast; it can't be said that this is the same voter as the 25-to-44 group, because the published tables don't cross age with any of these breakdowns, nor with sex.
It probably won't end in 2026, but it isn't a wave that grows on its own either. A preference that follows the same cohort across several elections tends to persist, and this cohort is large and in mid-life. But the next generation, entering now, doesn't repeat it: municipalities with more 16-to-24-year-old voters moved right between 2022 and 2026 just as much as the rest, and these voters are shrinking (18.2% of the electorate in 2010, 12.1% in 2026) and turn out less (79% in 2022, against 84% of those aged 25 to 69).
| Age group | 2010 Serra × Dilma | 2014 Aécio × Dilma | 2018 Bolsonaro × Haddad | 2022 Bolsonaro × Lula | 2026 Flávio × Lula |
|---|---|---|---|---|---|
| Ages 16-24 | +2 | +2 | −5 | −5 | −2 |
| Ages 25-34 | −1 | 0 | +1 | +1 | +3 |
| Ages 35-44 | −3 | −3 | +1 | +2 | +2 |
| Ages 45-59 | −2 | −1 | −1 | −1 | +1 |
| Ages 60+ | +3 | +2 | +1 | 0 | −4 |
| Country (% for the PT's opponent) | 45 | 48 | 55 | 48 | 49 |
2026, wave by wave. The 16-to-24 subsample in each Datafolha wave has a margin of 4 to 6 points per candidate, and the group swings from one side of the country to the other between consecutive waves. Those aged 60 or older sit below the country in all of them; those aged 25 to 44, above it in most.
| Year | PT among 16–24 (country) | Main opponent | Right-wing bloc | Bloc deviation |
|---|---|---|---|---|
| 2010 | 46 (50) | 31 (31) | 31 (31) | 0 |
| 2014 | 39 (44) | 26 (26) | 28 (28) | 0 |
| 2018 | 20 (25) | 38 (40) | 51 (54) | −3 |
| 2022 | 56 (50) | 30 (36) | 32 (38) | −6 |
| 2026 | 49 (45) | 34 (42) | 47 (51) | −4 |
| Pollster | Question | Group | Valid votes | Deviation (Flávio × Lula) |
|---|---|---|---|---|
| AtlasIntel Apr 27 | 1st round | 16-24 | Flávio 37 × Lula 29 · right-wing bloc 65 (+13) | +10 |
| AtlasIntel Oct 2 | 1st round | 16-24 | Flávio 27 × Lula 43 · right-wing bloc 54 (+4) | −9 |
| Datafolha Oct 3 | 1st round | 16-24 | Flávio 34 × Lula 49 · right-wing bloc 47 (−4) | −7 |
| Quaest Oct 3 | 1st round | 16-34 | Flávio 45 × Lula 41 · right-wing bloc 45 (−6) | +3 |
| Datafolha Oct 3 | Runoff (simulation) | 16-24 | Flávio 43 × Lula 57 | −6 |
| Quaest Oct 3 | Runoff (simulation) | 16-34 | Flávio 54 × Lula 46 | +3 |
| Group | 2010 | 2014 | 2018 | 2022 | 2026 |
|---|---|---|---|---|---|
| Sex | |||||
| Male | −4 | +1 | +5 | +2 | +5 |
| Female | +3 | −2 | −5 | −3 | −4 |
| Education | |||||
| Primary school | −6 | −9 | −11 | −9 | −12 |
| High school | +2 | +1 | +3 | +3 | +4 |
| College | +11 | +13 | +6 | +3 | +5 |
| Household income | |||||
| Up to 2 min. wages | −6 | −11 | −13 | −9 | −9 |
| Over 2 to 5 MW | +1 | +2 | +6 | +7 | +8 |
| Over 5 min. wages | — | — | — | — | +11 |
| Over 5 to 10 MW | +6 | +12 | +14 | +10 | — |
| Over 10 MW | +13 | +17 | +12 | +13 | — |
| Religion | |||||
| Catholic | −2 | — | — | −7 | −6 |
| Evangelical | — | — | — | +21 | +17 |
| Pentecostal evangelical | +3 | — | — | — | — |
| Non-Pentecostal evangelical | +5 | — | — | — | — |
| No religion | −4 | — | — | — | — |
| Region | |||||
| Southeast | +3 | +8 | +8 | +4 | +4 |
| South | +9 | +7 | +10 | +12 | +13 |
| Northeast | −13 | −18 | −23 | −19 | −15 |
| North | — | −6 | 0 | +4 | — |
| Center-West | — | +8 | +11 | +13 | — |
| North/Center-West | +1 | — | — | — | — |
| Center-West/North | — | — | — | — | +6 |
| Election | Registered at ages 16–17 | 16–24 in the electorate | Turnout 18–24 (1st round) | Turnout 25–69 |
|---|---|---|---|---|
| 2010 | 2,379,580 | 18.2% | — | — |
| 2014 | 1,638,751 | 16.1% | 81.6% | 84.8% |
| 2018 | 1,400,617 | 15.1% | 80.6% | 84.2% |
| 2022 | 2,116,781 | 13.7% | 78.6% | 84.0% |
| 2026 | 1,634,244 | 12.1% | — | — |
Voting is optional at 16 and 17: the number registered at that age measures the interest of the incoming cohort. Turnout by age (2014, 2018, 2022) comes from the TSE turnout-profile file; it doesn't exist for 2010, and the 2026 one has not been published yet.
This is the weakest evidence in the section, for three reasons: it is an association between places, not people (the young people of a young municipality don't necessarily vote like the municipality — the ecological fallacy); the share of young voters varies little across municipalities in the same state, which amplifies any noise; and the registry measures who is registered, not who voted. In each election's cross-sections, without differences, the same calculation produces impossible coefficients (10 of 14 above 100 points), because the share of young voters is confounded with region and income; that is why only the changes between elections appear here, and even they serve only as a check on the polls.
Sources: Datafolha eve-of-election polls (2010 eve of runoff; 2014 eve of runoff; 2018 eve of runoff; 2018 eve of runoff; 2022 eve of runoff; 2026 waves from Sep 17 to Oct 3); other 2026 polls in the table in section 3. TSE, Open Data: electorate profile 2010–2026, turnout and abstention profile 2014–2022, vote by municipality and zone 2010–2022; 2026 results release (election 6257), by zone. Ideological classification of candidates by party, our own. Code: coletar_eleitorado_idade.py, coletar_votos_historico.py, analisar_geracoes.py, geracoes.py.
In one sentence: government approval is what moves the scoreboard most: each additional point of approval is worth about 0.8 point of margin for Lula.
Petrol = effect favorable to Lula; terracotta = unfavorable; dashed = measurement channel (affects the polls, not the vote). Approval is the central mediator: the perceived economy and security feed into it; Bolsonaro's imprisonment and the Banco Master case act both on the cohesion of Flávio's base and on the third way's second choice.
| Variable | Current reading (Sep 2026) | Direction | Leading indicator to watch |
|---|---|---|---|
| Government approval | Real Time Big Data 45/52 (Sep 19–23) · BTG/Nexus 46/49 (Sep 25–27) · Datafolha 47/50 (Sep 22–23) · Quaest 43/50 (Sep 10–13) · CNT/MDA 47/48 (Sep 9–13) · Futura 44.1/53.1 (Sep 11–15) · Atlas 45.4/52.9 · PoderData 42/51; excellent/good rating 32% (Quaest) to 36.5% (Futura) | ↓ Lula | Sep/Oct waves; disapproval among men (50→57) and young people (46→55) — Nunes/Quaest |
| Flávio's base (evangelicals, men, South/Center-West, middle income) | Evangelicals 62×29 Flávio; men 48×43; South 50×38 (Datafolha Aug) | stable/↑ Flávio | Michelle on the campaign stage; perception of "moderate Bolsonaro" vs. "corrupt Bolsonaro" (Sanches/Atlas) |
| Third-way transfer | Zema 59/4 · Renan 58/14 · Marçal 47/19 · Cury 46/24 · Caiado 36/25 (Flávio/Lula) | 2:1 Flávio | If Cury or Caiado stay neutral or split 50/50, +3 pp for Lula; Caiado: "under no circumstances alongside Lula" |
| Undecided / blank / spoiled | Latest waves between 4.8% (AtlasIntel) and 14% (Quaest) | neutral | A drop in the pool to <8% signals crystallization; how late deciders split moves the margin by ±1.2 pp |
| Differential abstention | Abstention 20.95% (2022) with a projection of >22%; Lula's base (lower income/Northeast) stays home more | ↓ Lula | Activist mobilization; weather (rain); free transit in state capitals |
| Perceived economy | 49% say it got worse (Quaest Aug); 64% say income rose, only 10% above the cost of living; consumer confidence at its lowest since 2022 | ↓ Lula | August IPCA (~Sep 10) and Copom (Sep 16); food deflation in Jun–Aug not yet turned into perception |
| Court events / scandals | 40% say Lulinha hurts Lula a lot; 42% say Master hurts Flávio a lot (Quaest Sep 1). Since Sep 10: inquiry into Flávio (Dark Horse) made public; STF crisis over the Federal Police report on Moraes–Vorcaro | symmetric in the polls · pro-Flávio in the markets | Resumption of the ruling on the Moraes–Vorcaro report (Dino's request for review on Sep 15, up to 90 days); Lulinha inquiry (authorized Jul 30); restrictions on Bolsonaro until Oct 11; effect of the betting executive order (Sep 25) and of the Aparecida controversy in the first polls with fieldwork after Sep 25; Oct 12 (patron saint's day) in the runoff |
| Debates and TV | Lula 5'31" vs. Flávio 4'20" of TV time; both absent from the Band debate (Aug 23); SBT/CNN pool debate of Sep 14 canceled on Sep 8 because neither confirmed attendance | neutral | Oct 1 debate (Globo) canceled after Lula and Flávio pulled out: there are no more debates before the first round |
In one sentence: Lula and Flávio are rejected by almost half the electorate (Datafolha Oct 23–24: 46% × 45%), and that ceiling limits how much more each can still grow.
| Pollster | Fieldwork | Rejection — Lula | Rejection — Flávio | Difference |
|---|---|---|---|---|
| Futura (education from PNAD) | Oct 23–24 | 46.2%−1.1 p.p. | 46.6%+1.4 p.p. | Flávio +0.4 pp |
| Datafolha | Oct 23–24 | 46%+1 p.p. | 45%−1 p.p. | Lula +1 pp |
| Ipespe | Oct 20–22 | 47% | 48% | Flávio +1 pp |
| Alfa Inteligência | Oct 18–22 | 45%−1 p.p. | 44%−1 p.p. | Lula +1 pp |
| Datafolha | Oct 20–22 | 45%−2 p.p. | 46%−3 p.p. | Flávio +1 pp |
| Vox Brasil | Oct 20–22 | 44.8%−0.5 p.p. | 47.1%+1.8 p.p. | Flávio +2.3 pp |
| PoderData | Oct 19–21 | 47%−1 p.p. | 45%+1 p.p. | Lula +2 pp |
| Gerp | Oct 19–21 | 49% | 46% | Lula +3 pp |
| Quaest | Oct 17–20 | 47%−6 p.p. | 52%−2 p.p. | Flávio +5 pp |
| Futura (education from PNAD) | Oct 19–20 | 47.3% | 45.2% | Lula +2.1 pp |
| Gerp | Oct 15–19 | 49% | 46% | Lula +3 pp |
| Alfa Inteligência | Oct 15–19 | 46% | 45% | Lula +1 pp |
| PoderData | Oct 16–18 | 48%+2 p.p. | 44%−1 p.p. | Lula +4 pp |
| BTG/Nexus | Oct 16–18 | 47.8%−3.2 p.p. | 46.8%−0.2 p.p. | Lula +1 pp |
| Vox Brasil | Oct 15–17 | 45.3%+0.1 p.p. | 45.3%+1.1 p.p. | Tie |
| Futura Inteligência | Oct 16–17 | 47.4%+0.2 p.p. | 45.8%−0.3 p.p. | Lula +1.6 pp |
| Gerp | Oct 14–16 | 49%−4 p.p. | 46%+4 p.p. | Lula +3 pp |
| PoderData | Oct 13–15 | 46% | 45%+1 p.p. | Lula +1 pp |
| Futura Inteligência | Oct 14–15 | 47.2%+0.3 p.p. | 46.1%+3 p.p. | Lula +1.1 pp |
| Datafolha | Oct 13–14 | 47%+1 p.p. | 49% | Flávio +2 pp |
| Quaest | Oct 10–13 | 53%−2 p.p. | 54%−2 p.p. | Flávio +1 pp |
| Palver (v3 calibration) | Oct 10–13 | 50% | 48% | Lula +2 pp |
| PoderData | Oct 10–12 | 46% | 44%+1 p.p. | Lula +2 pp |
| Gerp | Oct 10–12 | 53%+4 p.p. | 42%+2 p.p. | Lula +11 pp |
| BTG/Nexus | Oct 10–12 | 51%+3 p.p. | 47%−4 p.p. | Lula +4 pp |
| Vox Brasil | Oct 5–7 | 45.2% | 44.2% | Lula +1 pp |
| PoderData | Oct 5–7 | 46%−1 p.p. | 43%−3 p.p. | Lula +3 pp |
| Datafolha | Oct 6–7 | 46%+1 p.p. | 49%+4 p.p. | Flávio +3 pp |
| Futura Inteligência | Oct 2–3 | 46.9% | 43.1% | Lula +3.8 pp |
| PoderData | Sep 30–Oct 2 | 47% | 46% | Lula +1 pp |
| Real Time Big Data | Sep 26–30 | 49%−1 p.p. | 49% | Tie |
| Quaest | Sep 24–27 | 55% | 56% | Flávio +1 pp |
| BTG/Nexus | Sep 25–27 | 48%−1 p.p. | 51%+1 p.p. | Flávio +3 pp |
| Real Time Big Data | Sep 19–23 | 50% | 49%−1 p.p. | Lula +1 pp |
| Datafolha | Sep 22–23 | 45%−2 p.p. | 45%−2 p.p. | Tie |
| Quaest | Sep 17–20 | 55%+2 p.p. | 56%+1 p.p. | Flávio +1 pp |
| BTG/Nexus | Sep 18–20 | 49% | 50% | Flávio +1 pp |
| Gerp | Sep 14–16 | 49%+2 p.p. | 40%−6 p.p. | Lula +9 pp |
| Datafolha | Sep 15–16 | 47%+1 p.p. | 47%+1 p.p. | Tie |
| Datafolha | Sep 8–10 | 46%+1 p.p. | 46%−1 p.p. | Tie |
| Palver | Sep 4–7 | 52% | 51% | Lula +1 pp |
| Quaest | Sep 3–6 | 53%+1 p.p. | 55%+1 p.p. | Flávio +2 pp |
| Datafolha | Sep 1–2 | 45% | 47%+1 p.p. | Flávio +2 pp |
| Real Time Big Data | Aug 27–31 | 50% | 50% | Tie |
| AtlasIntel | Aug 25–30 | 52% | 52.7% | Flávio +0.7 pp |
| Datafolha | Aug 18–20 | 45% | 46% | Flávio +1 pp |
| Quaest | Jul 31–Aug 3 | 52% | 54% | Flávio +2 pp |
| Gerp | Jul 15–17 | 47% | 46% | Lula +1 pp |
While the runoff lead swings 1–6 pp between polls in the same month, rejection varies by at most 1–2 pp between waves of the same pollster — the least noisy series in the whole dashboard.
Datafolha measures rejection about 4 pp lower than Quaest for both candidates — the same pattern as the undecided pool. The difference is in how the question is asked, not in the moment of the race.
From August to early September, Flávio led in rejection in every poll (+1 to +2 pp). In Datafolha's September waves the gap was tied (46×46, 47×47 and 45×45), and Palver (Sep 4–7) already had Lula more rejected. Gerp (Sep 14–16), the poll most favorable to Flávio in the runoff, measured the opposite (Lula 49×40, +9 pp) — probably the same pro-Flávio house effect already identified in the "Bias" section, not a real turnaround. BTG/Nexus (Sep 18–20) returns to a more typical pattern, with the two nearly tied and a slight edge for Flávio (50×49); Quaest (Sep 17–20) goes the same way, with both at the highest rejection in the series (56×55) and Flávio still 1 pp ahead. In the Sep 24–27 waves, Quaest repeats 56×55 and BTG/Nexus opens the pollster's widest gap of the month (Flávio 51 × Lula 48); on the eve, Real Time Big Data (49×49) and PoderData (Lula 47 × Flávio 46) go back to a tie. Quaest (Sep 7): 43% fear the return of bolsonarismo, 43% fear Lula's continuity — a parity of fear.
Mutual rejection of 45–56% caps how far either can grow: neither has obvious room to convert undecided voters into a decisive lead — hence a 30% probability of a result within ±1.5 pp in the ensemble.
In one sentence: between 4.8% and 14% of the electorate has yet to choose or plans to vote blank/spoiled, depending on the pollster — little for a big turnaround, but enough to decide a close election.
| Pollster | Fieldwork | Blank/spoiled | Undecided/DK | Total pool |
|---|---|---|---|---|
| Futura Inteligência | Oct 2–3 | 5.7% | 1.2% | 6.9% |
| Quaest | Oct 2–3 | 13% | 1% | 14%−2 p.p. |
| Datafolha | Oct 2–3 | 6% | 2% | 8%+1 p.p. |
| Palver | Sep 30–Oct 3 | 6% | 1% | 7% |
| CNT/MDA | Sep 30–Oct 2 | 7.6% | 1.9% | 9.5%−3.2 p.p. |
| Gerp | Sep 30–Oct 2 | 6% | 1% | 7% |
| Datafolha | Sep 28–Oct 1 | 6% | 1% | 7%−1 p.p. |
| DataTrends | Sep 28–30 | 7% | 3% | 10% |
| Real Time Big Data | Sep 26–30 | 5% | 4% | 9%−2 p.p. |
| AtlasIntel | Sep 23–28 | 4.8% | — | 4.8%−2.6 p.p. |
| BTG/Nexus | Sep 25–27 | 8% | 1% | 9%+2 p.p. |
| Quaest | Sep 24–27 | 13% | 3% | 16%−1 p.p. |
| Datafolha | Sep 22–23 | 7% | 1% | 8%−2 p.p. |
| Real Time Big Data | Sep 19–23 | 8% | 3% | 11% |
| Quaest | Sep 17–20 | 14% | 3% | 17%−1 p.p. |
| BTG/Nexus | Sep 11–13 | — | — | 7% |
| Indexa | Sep 10–13 | — | — | 15% |
| Quaest | Sep 10–13 | 13% | 5% | 18%+1 p.p. |
| CNT/MDA | Sep 9–13 | 10.2% | 2.5% | 12.7% |
| Datafolha | Sep 8–11 | — | — | 10%−1 p.p. |
| AtlasIntel | Sep 4–9 | 7.4% | — | 7.4%−2.9 p.p. |
| Datafolha | Sep 1–3 | 9% | 2% | 11% |
| Quaest | Aug 30–Sep 1 | 13% | 4% | 17% |
| AtlasIntel | Aug 25–30 | 10.3% | — | 10.3%+2.4 p.p. |
| Datafolha | Aug 18–20 | 9% | 2% | 11% |
| Quaest | Jul 31–Aug 3 | 13% | 4% | 17% |
| AtlasIntel | Jul 22–27 | 7.9% | — | 7.9% |
| Datafolha | Mar 5–7 | — | — | 11% |
Series: 11% → 11% → 11% → 10% → 8% → 7% → 8% (latest: Oct 2–3). It held at 10–11% from March to early September and shrank in the latest waves: a more crystallized electorate under this methodology, with little room for newcomers.
Series: 17% → 17% → 18% → 17% → 16% → 14% (latest: Oct 2–3). Almost double Datafolha's, more because of blank and spoiled votes (13–14%) than undecided ones. Consistent with an in-person household methodology, which historically captures more indecision.
Series: 7.9% → 10.3% → 7.4% → 4.8% (latest: Sep 23–28). The August rise was undone on Sep 4–9, and the pool that left went to Flávio (the margin moved from Lula +4.5 to Flávio +0.2). In the latest wave, almost no voters are left outside the two.
On the eve of the first round, each pollster's latest waves sit between 4.8% (AtlasIntel) and 14% (Quaest). Where the pool shrank, the destination varied: in AtlasIntel it went to Flávio; in Datafolha, Lula rose a point. It enters the model as the tornado's "split of the undecided" variable.
In one sentence: one in five voters usually stays home, and those who stay home are poorer and less educated, groups Lula leads. That is why each additional point of abstention costs Lula about 0.25 point of margin. The effect is real but limited: even 25% abstention, well above the 2022 pattern, would take 1.1 points from Lula (chance from 17% to 7%).
| Scenario (deviation from 2022) | Geographic effect | Profile effect | Total on the margin (range) | Lula | Flávio |
|---|---|---|---|---|---|
| Abstention equal to 2022 (20.6%) | 0.00 | 0.00 | 0.00 (0.0 to 0.0) | 17% | 83% |
| Rises to about 22% nationwide (analysts' projection) | 0.00 | -0.35 | -0.35 (0.0 to -0.6) | 13% | 87% |
| Rises to about 25% nationwide (disengagement scenario) | +0.01 | -1.10 | -1.09 (0.0 to -1.8) | 7% | 93% |
| Northeast stays home more: +3 pp in the Northeast only | -0.38 | -0.21 | -0.59 (-0.4 to -0.7) | 11% | 89% |
| South and Center-West stay home more: +3 pp in the South and Center-West only | +0.20 | -0.17 | +0.03 (+0.2 to -0.1) | 17% | 83% |
| Mobilization in the Northeast: −2 pp in the Northeast only | +0.25 | +0.14 | +0.39 (+0.2 to +0.5) | 21% | 79% |
| Repeats the 2022 first-to-second-round change, state by state | -0.04 | +0.09 | +0.06 (0.0 to +0.1) | 17% | 83% |
Margin in points of valid votes (Lula − Flávio). Chance of winning from the polling model, same simulation as the rest of the dashboard, changing only the level. The profile effect uses K = 0.25 point of margin per point of abstention, calibrated on the estimate that every 4 additional points of abstention take about 1 point off Lula's lead (Adriano Laureno, quoted by InfoMoney); the range in parentheses goes from K = 0 (those who stay home vote like those who turn out) to K = 0.4.
| Region | 2026 electorate | Abstention, 1st round 2022 | Abstention, 2nd round 2022 | 2022 margin | Estimated 2026 margin |
|---|---|---|---|---|---|
| Southeast | 42.0% | 22.0% | 21.2% | -8.5 | -11.5 |
| Northeast | 27.6% | 19.5% | 19.3% | +38.6 | +34.7 |
| South | 14.5% | 19.3% | 18.6% | -23.8 | -27.3 |
| North | 8.3% | 20.9% | 22.9% | -2.1 | -7.2 |
| Center-West | 7.6% | 21.4% | 20.5% | -20.4 | -20.9 |
Abstention weighted by the 2026 electorate from state rates (CNN Brasil, rankings for both 2022 rounds; single source, with the jumps in Amapá and Roraima between rounds still to be checked against the TSE). 2022 margin = official runoff result by region; the reconstruction matches the published figures (Southeast -8.5, Northeast +38.6, South -23.8). 2026 margin = 2022 margin + swing estimated from state polls, aligned with the model's national scoreboard.
| Election | Abstention, 1st round | Abstention, 2nd round |
|---|---|---|
| 2006 | 16.75% | 18.99% |
| 2010 | 18.12% | 21.50% |
| 2014 | 19.39% | 21.10% |
| 2018 | 20.32% | 21.29% |
| 2022 | 20.95% | 20.59% |
| 2026 | ~22% (analysts' projection) | — |
Sources: TSE via CNN Brasil, Gazeta do Povo, Poder360 and Agência Brasil. Abstention rose in the first round in every election since 2006. The projection of about 22% for 2026 comes from analysts quoted in the press, not from a published model; some speak of 25% counting blank and spoiled votes.
In the 2022 first round, abstention was 52% among illiterate voters (optional voting) and 12% among college graduates; about 55% of absentees had at most an elementary education, a group that is 40% of the electorate. By age, 34.5% of voters aged 60 or older and 58.9% of those 70 or older stayed home; among those required to vote, young people aged 18 to 29 turn out less. Men stayed home more than women (21.9% × 19.8%). There is no official data on abstention by income, but the education and regional profile points to poorer voters (sources: Congresso em Foco and InfoMoney, with TSE data; Jairo Nicolau, SciELO).
Lula leads among the less educated, the poorer and in the Northeast: when abstention grows, it tends to grow where his voters are. That is why the profile effect is almost always negative for him, even in scenarios where abstention rises in Flávio's regions: in the "South and Center-West stay home more" scenario, Lula's geographic gain (+0.2 point) is almost entirely canceled by profile (total +0.03). His lever is the reverse, mobilizing the Northeast: two points less abstention in the region are worth +0.4 point of margin (21% chance). For Flávio, the symmetric risk is smaller: his bases (South, Center-West, middle-income voters) have lower and less volatile abstention.
Three findings temper the alarm. Jairo Nicolau (FGV) concluded that the uneven abstention of 2022 did not change the aggregate result. The Ipea study (Pereira et al., 2023) found no effect from free public transit, allowed by the STF in the runoff, on either turnout or the result. And the hypothesis that abstention explained the polling error in the 2022 first round finds no support in published analyses: the recurring explanations are shy voting and last-minute decisions. Repeating in 2026 the change in turnout between the 2022 rounds, state by state, moves the margin by +0.06 point.
The electorate grew to 158.7 million and got older: there are 17 million voters aged 70 or older, for whom voting is optional and abstention exceeded 50% in 2022. About 20 state capitals confirmed free transit on Oct 4 (TSE Res. 23.751/2026), but the 2022 evidence suggests a small effect. The long holiday weekend of Oct 10 to 12 falls between the rounds, not on election day. No 2026 poll has published Lula × Flávio only among those who say they will definitely vote; Datafolha on Sep 5 recorded 57% of voters "more determined to vote" than in 2022.
Limitations: the abstention table by state has a single source (CNN Brasil, from the TSE); the K parameter comes from a single published estimate, which is why the 0 to 0.4 range is shown in every scenario; the model treats abstention as a deviation from the 2022 pattern and does not estimate individual turnout, which would require poll microdata with a vote-certainty question.
In one sentence: Without a targeted campaign, migration takes the margin to −2.7 points (80% between −3.5 and −2.0). A strong person-to-person campaign, with 500,000 initial volunteers and the effectiveness the literature measures, yields a net +32,000 votes and moves the margin by 0.0 point; the maximum one, +0.1. A turnaround would require +5.1 million: with strong reach, conversion per contact would have to be 23 times the literature's figure in the third way. Lula's chance, with poll drift, goes from 0.0% with no campaign to 0.0% with the strong one. The headline doesn't change: this is conditional analysis, not a forecast.
At the end of the strong campaign, among absentees: 3.8% of the 32.9 million pool reached, 0.00% with a voting plan or confirmed. Contacts are few relative to the pool: the bottleneck is reach, not the conversation.
| Scenario | Flávio's response | Volunteers at the end | Names reached | Absent | Third way | Blank/spoiled | Flávio voters | Open call | Flávio's response | Total net (80%) | Margin (80%) | Lula, migration only | Lula with drift |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| No targeted campaign | None | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0k (0k to 0k) | −2.7 (−3.5 to −2.0) | 0.0% | 0.0% |
| Moderate | None | 267k (118k–646k) | 790k | 0 | 0 | 0 | 0 | +16k | 0 | +16k (+5k to +39k) | −2.7 (−3.5 to −2.0) | 0.0% | 0.0% |
| Strong | None | 1.3M (584k–2.5M) | 3.5M | 0 | 0 | 0 | 0 | +32k | 0 | +32k (+10k to +77k) | −2.7 (−3.4 to −1.9) | 0.0% | 0.0% |
| Maximum | None | 2.4M (1.4M–3.8M) | 6.3M | 0 | 0 | 0 | 0 | +48k | 0 | +48k (+15k to +116k) | −2.7 (−3.4 to −1.9) | 0.0% | 0.0% |
| No targeted campaign | Symmetric | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0k (0k to 0k) | −2.7 (−3.5 to −2.0) | 0.0% | 0.0% |
| Moderate | Symmetric | 267k (118k–646k) | 790k | 0 | 0 | 0 | 0 | +16k | −4k | +10k (−2k to +36k) | −2.7 (−3.5 to −2.0) | 0.0% | 0.0% |
| Strong | Symmetric | 1.3M (584k–2.5M) | 3.5M | 0 | 0 | 0 | 0 | +32k | −7k | +20k (−4k to +73k) | −2.7 (−3.4 to −1.9) | 0.0% | 0.0% |
| Maximum | Symmetric | 2.4M (1.4M–3.8M) | 6.3M | 0 | 0 | 0 | 0 | +48k | −11k | +29k (−6k to +109k) | −2.7 (−3.4 to −1.9) | 0.0% | 0.0% |
| Playbook targets met (reference) | None | — | — | +500k | +2.39M | +1.00M | +1.20M | 0 | 0 | +5.09M (+5.1M to +5.1M) | +1.5 (+0.8 to +2.2) | 99.6% | 99.6% |
Net votes (Lula − Flávio), median of the draws; the fronts may not add up to the total because of rounding and the per-state caps on available voters. "With drift" adds to the migration margin the same t5 drift used in the main model (σ = 0.0 pp through Oct 25): it is the chance comparable to the headline (17% main, 23% MCMC), but without the polls. The last row imposes the playbook targets (+5.1M) on the no-campaign scenario: it shows where the margin would go if they were met, not how likely they are to be met.
| Front | Playbook target | Model, strong campaign (80%) | Names reached | Conversion in the literature | Conversion needed | Ratio |
|---|---|---|---|---|---|---|
| Third way split the votes of Cury, Caiado and Renan down the middle | +2.4M | 0k (0k to 0k) | 1.2M | 5% | above 100% | 23× |
| Flávio votes win over 600,000 soft voters | +1.2M | 0k (0k to 0k) | 1.1M | 2% | 59% | 29× |
| Blank and spoiled persuade 1 million to vote for Lula | +1.0M | 0k (0k to 0k) | 1.2M | 5% | 96% | 19× |
| Those who didn't vote bring 1.5 million absentees in Lula strongholds to the polls | +500k | 0k (0k to 0k) | 1.2M | 12% | 79% | 7× |
| Total | +5.1M | +32k (+10k to +77k) | 3.5M |
If all four targets were met, the median margin would reach +1.5 points and Lula's chance, from migration alone, 99.6% (99.6% with drift). Each third-way or Flávio voter who switches counts twice in the net; each blank/spoiled voter or absentee brought in counts once (absentees reached by name vote 80% for Lula, so each yields 0.6). Blank and spoiled voters share the same pool of names with the third way.
| Origin (1st round) | 2nd-round destination, no targeted campaign | Strong campaign: difference (pp) | ||||||
|---|---|---|---|---|---|---|---|---|
| Lula | Flávio | Blank/spoiled | Doesn't vote | Lula | Flávio | Blank/spoiled | Doesn't vote | |
| Voted for Lula 53.9M | 93.2% (90.7–95.2) | 0.4% (0.2–0.9) | 0.9% (0.4–1.7) | 5.3% (3.4–7.7) | 0.00 | 0.00 | 0.00 | 0.00 |
| Voted for Flávio 56.1M | 0.4% (0.2–0.9) | 94.2% (92.0–95.9) | 0.9% (0.4–1.7) | 4.3% (2.8–6.3) | 0.00 | 0.00 | 0.00 | 0.00 |
| Voted for Cury | 47.3% (44.7–49.9) | 41.3% (38.7–43.8) | 6.4% (4.2–8.8) | 4.8% (3.0–7.2) | 0.00 | 0.00 | 0.00 | 0.00 |
| Voted for Renan | 34.6% (31.6–37.8) | 49.2% (46.0–52.5) | 11.0% (7.4–14.8) | 4.8% (3.0–7.2) | 0.00 | 0.00 | 0.00 | 0.00 |
| Voted for Caiado | 31.1% (28.2–34.1) | 52.9% (49.9–56.2) | 10.9% (7.3–14.6) | 4.8% (3.0–7.2) | 0.00 | 0.00 | 0.00 | 0.00 |
| Voted for Zema or others | 21.8% (19.2–24.5) | 64.6% (61.5–67.8) | 8.5% (5.6–11.6) | 4.8% (2.9–7.3) | 0.00 | 0.00 | 0.00 | 0.00 |
| Voted blank or spoiled 6.0M | 15.6% (9.5–23.5) | 15.6% (9.5–23.4) | 60.3% (47.1–71.9) | 7.6% (4.3–12.1) | 0.00 | 0.00 | 0.00 | 0.00 |
| Didn't vote 33.5M | 8.6% (4.0–13.7) | 7.4% (3.5–11.9) | 0.8% (0.3–1.4) | 83.1% (73.2–92.0) | +0.37 | +0.26 | 0.00 | −0.64 |
Gross turnout flows: median rout 5% of those who voted (6.1M), median rin 16% of absentees (5.7M); national net change in abstention +0.5 pp (80% between −0.8 and +1.8); median profile tilt τ 10 pp of margin. With equal inflow and outflow and the same tilt in both directions, the net turnout effect is close to zero; only the deviation from 2022 matters, as in the Abstention section.
| Region | 1st-round absentees | Abstention, 1st round | Expected 2nd-round abstention | No targeted campaign | Strong campaign | 1st-round margin | 2nd-round margin (strong) | ||
|---|---|---|---|---|---|---|---|---|---|
| New voters | Net | New voters | Net (80%) | ||||||
| Northeast | 8.00M | 18.4% | 19.0% | 1.57M | −398k | 1.66M | −355k (−734k to +30k) | +34.8 | +31.9 |
| Southeast | 15.06M | 22.7% | 23.0% | 2.41M | −370k | 2.47M | −371k (−0.96M to +243k) | −12.8 | −12.4 |
| South | 4.68M | 20.5% | 20.6% | 864k | −75k | 880k | −78k (−291k to +133k) | −31.5 | −29.4 |
| North | 2.55M | 19.4% | 20.9% | 327k | −45k | 354k | −42k (−136k to +54k) | −4.8 | −5.0 |
| Center-West | 2.60M | 21.7% | 21.8% | 450k | −123k | 460k | −124k (−237k to −11k) | −26.8 | −25.5 |
| Abroad | 574k | 62.7% | 62.9% | 14k | −5k | 14k | −5k (−15k to +6k) | +4.5 | +2.6 |
Net votes and margin in pp of valid votes, medians. Voters abroad are included in the totals and the margin, but are left out of the campaign and the profile tilt.
Two nested chains. The first takes a single step: each first-round voter, classified by state and by what they did on Oct 4, moves to one of four runoff destinations according to a drawn 8 × 4 matrix; mass is conserved (the destinations add up to each state's electorate). The second is daily and non-homogeneous: in each pool (absentees, third-way and blank/spoiled voters, soft Flávio voters), a voter goes from not reached to reached when they land on a volunteer's list; from reached to converted with probability proportional to the per-contact conversion q, spread across the playbook's phases (the second conversation weighs 1.4; the close and the first, 0.8; the pause, 0); from converted to confirmed at the close and confirmation stages; and falls back to reached at rate δrev per day. Those who confirm vote as agreed in 85% (absentees) or 90% of cases; those with only a voting plan, in 55%. The converted stock is distributed across states in proportion to Lula's vote and applied to the destinations of the first chain, capped at the mass available in each box.
In the "symmetric" column, Flávio builds a network ρ times the size of PT's (median 0.6) with the same effectiveness priors and mirrored pools: absentees reached by name (80% for him), persuadable third-way and blank/spoiled voters, soft Lula voters. The open call to vote (radio, street markets, churches) pulls ω × intensity from the absentees in the states each side led, with the mix of those who stay home (more pro-Lula profile, τ); in the opponent's state, the generic appeal yields 30% and carries the local mix. The net effect of the response has its own column. Because both networks use the same priors, the difference between the columns measures only the asymmetry in size (ρ) and in base (Lula has more absentees in his strongholds; Flávio has more votes to protect).
The priors are in the table below, with the source for each one. The ones that matter most are the number of volunteers (unknown: hence the ladder of scenarios) and the per-contact conversion, taken from the field-experiment literature: about one vote per 14 door-to-door contacts for turnout, a near-zero persuasion effect in general elections, and larger effects when the contact comes from someone close. Quaest's second-choice question is from September; when a runoff wave with the question asked again comes out, it will replace the centers, and the campaign chain will start the day after fieldwork, so as not to double-count conversations already held.
Nothing. Lula's chance on the dashboard still comes from the main model (17%) and the MCMC (23%), which start from the polls and the counted vote. This section answers a different question: given the Oct 4 result, how much can a campaign of personal conversations with the three groups move, and what would have to happen for a comeback. The "with drift" version of the chance exists only to put both calculations on the same scale: with no targeted campaign it gives 0.0%, close to the main model, slightly lower because it does not carry the uncertainty of the polls and house effects; the structural level of the two is the same (reproduction of the transfer model: −2.7 vs. −2.7).
| Parameter | Symbol | Distribution | Mean | SD | Source or rationale |
|---|---|---|---|---|---|
| Direct switching between Lula and Flávio | ε_x | Beta | 0.005 | 0.003 | explicit prior: direct switching between the two is rare |
| Lula or Flávio voter who votes blank/spoiled | ε_bn | Beta | 0.010 | 0.005 | explicit prior |
| Voted in the 1st round and stays home in the 2nd | r_out | Beta | 0.050 | 0.015 | about 6M; in 2022 abstention barely changed between rounds, but different people stayed home |
| National net change in abstention (pp) | δ_nac | Normal | 0.50 | 1.00 | abstention rose in the runoff in 2006–2018 and fell in 2022 |
| Net change by state beyond the 2022 pattern (pp) | δ_u | Normal | 0.000 | 0.70 | 2022 1st→2nd round pattern by state, shrunk 50% and capped at ±3 (abstencao_2022_uf.csv) |
| Profile tilt of those who come in or drop out (pp of pro-Lula margin) | τ | Normal, ≥ 0 | 10.00 | 5.00 | abstention by education 2022 (27/17/13%) × vote by education (Quaest Oct 3) ≈ +6; K = 0.25 from the Abstention section ≈ 20 |
| Third-way "no second choice" voters who pick one of the two | d_3 | Beta | 0.50 | 0.12 | structural model: pool decided at 55% |
| Lula's share among those who pick | s_3 | Beta | 0.50 | 0.080 | 35–65% range from the tornado |
| First-round blank and spoiled voters who pick one of the two | d_bn | Beta | 0.35 | 0.10 | TSE 2022: blank + spoiled votes fell from 7.0M in the first round to 5.7M in the second |
| Blank and spoiled voters who stay home in the runoff | a_bn | Beta | 0.080 | 0.030 | explicit prior |
| Acceptance of the invitation to build one's own list | p_ac | Beta | 0.25 | 0.10 | relational organizing: 1 in 4 invitees sign up |
| Volunteer cap (share of Lula's votes) | ψ | Beta | 0.050 | 0.020 | 2.7M at the center, the order of magnitude of PT's party membership |
| Per-contact conversion: absentees | q_aus | Beta | 0.12 | 0.050 | Gerber & Green (door to door: 1 vote per ~14 contacts); Nickerson & Rogers (voting plan: +4 pp); relational organizing: +5 to +10 pp |
| Per-contact conversion: third way and blank/spoiled | q_tv | Beta | 0.050 | 0.030 | Kalla & Broockman 2018: persuasion ≈ 0 in general elections, nonzero among the undecided; only those with no second choice are persuadable |
| Per-contact conversion: soft Flávio voter | q_fb | Beta | 0.020 | 0.015 | same, with the vote already cast for the opponent |
| Soft voters (share of each candidate's vote) | f_br | Beta | 0.12 | 0.040 | playbook: 600,000 "not very committed" voters is the floor |
| Daily reversion from converted to reached | δ_rev | Beta | 0.030 | 0.010 | voting plan that isn't carried out |
| Lula's share among absentees reached by name | ν | Beta | 0.80 | 0.070 | work by name: only the volunteer's list |
| Open call: drop in absentees in Lula's states (× intensity) | ω | Beta | 0.015 | 0.010 | Ipea 2023: free transit in the 2022 runoff had no measurable effect; low ceiling |
| Flávio's network ÷ PT's network (symmetric response) | ρ | Log-normal | 0.60 | 0.40 | same effectiveness priors, less grassroots reach |
| Initial volunteers (spread around the scenario) | V_0 | Log-normal | scenario | 0.50 | unknown: the scenario ladder runs from 0 to 2M |
Dirichlet row concentrations: κ = 400 (Lula and Flávio), 100 (third way: Quaest subsamples of 100 to 150 interviews, SD of 4 to 5 pp), 150 (blank/spoiled and absentees). Script checks: mass conserved in every draw (ok); rows sum to 1 (ok); chain in [0, 1] and converted ≤ reached (ok); reproduction of the structural model (ok); monotonic ladder and symmetric response ≤ no response (ok); 2 s.
Sources: TSE, official first-round result by state (100% of polling stations); PoderData/Aya Oct 10–12 (stated transfer; before that, Quaest Sep 3–6); TSE 2022 via CNN Brasil (abstention by state in both rounds); TSE, 2022 turnout profile by education; Gerber & Green, Get Out the Vote; Nickerson & Rogers (2010), Psychological Science; Kalla & Broockman (2018), American Political Science Review; Pereira et al. (Ipea, 2023) on the 2022 free transit. Code: migracao_markov.py (calculation, self-checks and this section), called by modelo_eleicao_2026.py (block 5b).
In one sentence: so far, the scandals have not changed the poll numbers — those who say the Dark Horse case pushes them away from Flávio mostly weren't voting for him anyway — but the betting markets reacted to the STF crisis against Lula. In the final stretch (Sep 24–26), three new events tilted slightly toward Lula: the betting ban (supported by 73%), the Nossa Senhora Aparecida controversy and the third phase of Carbono Oculto, which reaches the man handling the money for Flávio's film. The first poll with fieldwork after them (Quaest, Sep 24–27) shows Lula +1 and Flávio steady — 42 × 42, within the margin of error.
The Monitor do Debate Público (LEMEP/IESP-UERJ, supported by INCT ReDem) runs an ongoing panel of asynchronous WhatsApp focus groups — 50 participants in six segments (committed and moderate Bolsonaro supporters, conservative and progressive undecideds, Lula supporters and "disaffected Lula voters"). It is qualitative research: it collects narratives and arguments, with no statistical validity. In the available reports, Bolsonaro supporters reject Lula's reelection en bloc, citing corruption; undecideds defend alternation in power as a democratic principle, not out of affinity with Flávio; Lula supporters back him for "economic stability"; and "disaffected Lula voters" swing between specific frustrations and strategic tactical-voting calculations. Public safety, health and inequality appear as complaints across all six groups.
Quaest adds image items within quantitative polls (they are not focus groups, but they measure perception, not just intention): on Mar 2, 2025, 71% said Lula "is not managing to keep his campaign promises," 47% already considered him "ill-intentioned" and 50% said his public appearances worsened the government's image (31% said they improved it). This research did not locate a symmetric, equally detailed image study of spontaneous perceptions of Flávio — a data gap recorded here as such, not as an absence of scrutiny of him.
The four cases converge on Vorcaro (Banco Master) — even the STF crisis stems from the analysis of his phone — or on schemes that predate the current term (INSS), but the nature of the exposure differs. Master and INSS reach Lula indirectly: an off-the-books meeting, a former minister who fell, an inquiry into influence peddling by his son (Lulinha, authorized on Jul 30). Dark Horse exposes Flávio directly and personally: he is the one who negotiated with Vorcaro a US$24M contract for the film about his father, and he has been formally under investigation since Jul 22 (money laundering, illegal currency transfers, active and passive corruption) — a fact made public only on Sep 10, when André Mendonça released 15 inquiries. The STF crisis, in turn, hits mainly the court itself: a 218-page Federal Police report, requested by Mendonça, records 187 interactions between Vorcaro and Justice Alexandre de Moraes — the central antagonist of Bolsonarism — and it is this report that the full court takes up on Sep 15. For the campaign, this creates a shock of ambiguous sign: it exposes Flávio (inquiry) and, at the same time, gives ammunition to his narrative against Moraes and the STF ("Lula wants to win through the courts").
| Scandal | Origin → stage in Sep 2026 | Reference poll | Result | Most exposed |
|---|---|---|---|---|
| STF crisis: Moraes–Vorcaro report, control of the Federal Police and lifting of secrecy | On Aug 24, Mendonça orders the Federal Police to deepen its analysis of Vorcaro's phone; the report (218 pp.) points to 187 interactions between the banker and Alexandre de Moraes, with messages going back to Dec 2023. The Prosecutor General's Office asks for it to be annulled (investigation of a justice without notifying the full court). On Sep 8, Mendonça removes the Federal Police director-general, Andrei Rodrigues, and the intelligence director, alleging "systematic illegal monitoring" of his actions; Dino reinstates them on Sep 9; Fachin suspends both decisions and all proceedings involving STF justices, and calls an extraordinary session for Sep 15. On Sep 10, Mendonça makes public 15 inquiries in the Master case, confirming Flávio is under investigation; justices accuse him of selectivity. Lula calls for a "complete lifting" of secrecy for everyone involved; Caiado files a similar request; Flávio presses Fachin and accuses the other side of trying to win "through the courts." At the Sep 15 session the STF did not rule on the merits: Gilmar Mendes raised a point of order to merge the petitions in the Moraes and Mendonça cases, Fachin announced a provisional 4–3 tally for keeping them separate, and Flávio Dino — whose vote would have tied it 4–4 — asked for more time to review, with up to 90 days to return the case. The validity of the Federal Police report and any investigation of Moraes remain undecided. | No dedicated opinion poll as of the data date — institutional/legal development | No direct measurement; markets priced the shock against Lula: Polymarket 59% (Sep 8) → 49% (Sep 13) → 45% (Sep 16–17) → 43% (Sep 18) → 40% (Sep 21) → 41% (Sep 23) → 42% (Sep 24) → 43% (Sep 26), a recovery that starts before the betting decree and continues after it | Ambiguous, and now prolonged: with no ruling, the case stays open through the first round. Flávio turns the crisis into a platform ("Lula is Moraes and Moraes is Lula"); Lula took up an explicit defense of Moraes on Sep 16 and hit back ("Flávio is getting nervous... the R$130 million he took from Vorcaro") |
| Banco Master | Private Lula–Vorcaro meeting at the Planalto (Dec 2024), off the official schedule, with Galípolo (then nominee to the Central Bank) present; the Federal Police is investigating a billion-real fraud at the bank. The case resurfaces at the STF on Sep 2, 4 and 9, 2026, with Toffoli denying he received payments from Vorcaro — and the partial lifting of secrecy on Sep 10 and the Moraes–Vorcaro report (see row above) reopen the case publicly. | PoderData, Jun 21–24, 2026 · n=2,400 · MoE ±2 pp (95%) · TSE BR-05722/2026 | 54% blame the Lula government · 29% the Bolsonaro administration · 17% don't know (86% are aware of the case) | Lula (but the same Vorcaro funds Dark Horse — the case also splashes Flávio, now with a confirmed formal investigation) |
| INSS fraud | Improper association fees deducted from pensions expanded under the Bolsonaro government (2019–24, R$6.3B) and uncovered only in Operation Sem Desconto (Apr 2025); minister Lupi resigns on May 2, 2025. The scheme's operator ("Careca do INSS") is linked to an associate of Fábio Luís ("Lulinha"); on Jul 30, 2026, Mendonça authorizes a Federal Police inquiry into influence peddling by Lulinha with the Health Ministry and the Presidency on behalf of World Cannabis, a company linked to the lobbyist. | Quaest/Genial, May 29–Jun 1, 2025 · n=2,000 · MoE ±2 pp (95%) | 31% blame the Lula government · 8% the Bolsonaro government · 14% the INSS itself · 8% the fraudulent entities · 26% don't know | Lula (the scheme started earlier, but the political cost falls on whoever is in office now — and the link to Lulinha reopens the exposure) |
| Dark Horse (Azarão) | Flávio negotiates with Vorcaro a US$24M contract in 14 installments for the biopic about his father; the Federal Police traces US$12.3M (≈R$62.8M) paid between Feb and Sep 2025 via Entre Investimentos → the Havengate fund (Texas). Audio leaks on May 13, 2026; Flávio denies it and later admits the contact ("private sponsorship"). Inquiry requested by the Federal Police on Jul 8 and authorized by Mendonça on Jul 22 (money laundering, illegal currency transfers, active and passive corruption), made public on Sep 10. The film's release is suspended on Sep 11; Flávio questions the "political timing." | Meio/Ideia, Sep 4–7, 2026 · n=1,500 (phone) · MoE ±2.5 pp (95%) · TSE BR-07935/26 | 47.9% say the link to Vorcaro makes them less likely to vote for Flávio · 32.2% say it makes no difference · 9.3% say it makes them more likely · 10.7% don't know | Flávio, by a wide margin — it is the only case in which the candidate himself negotiates, contradicts his own account and is named in an investigation. An open inquiry does not bar his candidacy |
An earlier wave by the same pollster on Dark Horse (Meio/Ideia, May 28, 2026): 57% already said the case hurt Flávio (33% "a lot," 24% "a little") and 44% had a worse opinion of him — the measured damage is not an isolated spike; it deepened from May to September, and the partial lifting of secrecy on Sep 10 adds a new fact (a formal investigation) whose effect has not yet been measured by a dedicated question. In voting-intention polls with fieldwork after Sep 10 (Quaest, Nexus, Indexa, CNT/MDA), Flávio shows no drop: he moves from 41% to 42% in Quaest and holds steady in Nexus — and his rejection ties Lula's in Datafolha (46×46).
| Event | What happened | What the polls already say | Who stands to gain or lose | Expected effect on the race |
|---|---|---|---|---|
| Betting ban (provisional measure of Sep 25) | In São Paulo, Lula signs a provisional measure banning sports betting and online casinos (the "jogo do tigrinho") and their advertising. New deposits are suspended; money can be withdrawn until 11:59 p.m. on Oct 5 and will be refunded between Oct 9 and 14. A bill provides for 4 to 6 years in prison for anyone who operates or promotes betting. Official rationale: household debt. Two days earlier, in a livestream (Sep 23), Flávio promised to shut down online casinos but said he opposed banning sports betting, and blamed Lula for regulating the sector. Fixed-odds betting was legalized in 2018 (Law 13,756, Temer government); the regulations and licenses came under the Lula government, in 2023–24. | AtlasIntel/Bloomberg (Sep 17–22, n=5,015): 73% in favor of the ban, 21% against; 59.9% say betting only does harm; 36% blame the Lula government for its spread. On social media (Palver), 77% of messages support the ban and Lula appears in 51% of them. | Lula, with risk. The measure is popular, speaks to indebted low-income voters and resonates with religious voters, including evangelicals in Flávio's base, who reject gambling; Flávio is left in an in-between position (against casinos, in favor of sports betting) that is hard to explain. Risks: looking like an electoral ploy nine days before the vote, challenges to the measure in the STF and Congress, and the loss of club sponsorships (14 of the 20 Série A clubs display betting companies). | Small and not yet measured: 0 to +1 point for Lula is the plausible range for end-of-campaign announcements. The market rose from 42% to 43% the next day, but it was already rising before. The ban does not change any model parameter; it is listed here as a factor to watch in polls with fieldwork after Sep 25. |
| Flávio and Nossa Senhora Aparecida (Sep 24–25) | On the program Em Pauta (GloboNews), a commentator mentioned a supposed evangelical movement to strip Nossa Senhora Aparecida of the title of patron saint of Brazil. On social media, the clip was tied to Flávio, along with a remark he made on Aug 24 ("one of my first acts will be to decree that Brazil belongs to the Lord Jesus Christ"). The actual bill is PL 2,623/2007, by then-congressman Victório Galli, which was shelved — it is not Flávio's and is not moving. GloboNews corrected the information and apologized. Flávio called the episode fake news from the "PT militia," said he respects all religions and is considering going to the TSE; he attacked Lula for weighing in on priestly celibacy. The president of the CNBB, Dom Jaime Spengler, criticized the "mixing of religion and politics" without naming names. | No dedicated poll. Structural references: Catholics are the largest share of the electorate and voted mostly for Lula in 2022; evangelicals (≈30%) are Flávio's most solid base (62 × 29 in the August Datafolha). | Ambiguous, with slight risk for Flávio. The information is false and was corrected, which limits the damage among undecideds; but the August quote is real and reinforces his image as the evangelical candidate among Catholics. The bigger risk lies in the calendar: Oct 12, the patron saint's day, falls in the middle of the runoff campaign, and the issue may return. | Probably marginal in the first round. If it resurfaces on Oct 12, it weighs on Catholic voters who are still wavering — the same ones who decide close elections. |
| Carbono Oculto, 3rd phase (Sep 24) | The São Paulo prosecutors' Gaeco unit and the Federal Revenue Service, with support from the Federal Police and state police, execute warrants at 29 addresses (13 people, 24 companies) in São Paulo and six other states. Focus: funds, fintechs and shell companies used to buy the fuel distributor Tobras (Terrana), with R$11 billion in identified transactions and a R$370 million ethanol plant. The Tupinambá brothers (Humberto, formerly of Banco Genial) are named as partners of PCC leaders ("Beto Louco" and "Primo"). Also targeted is Antônio Carlos Freixo Júnior, "Mineiro," owner of Entre Investimentos and identified by the Federal Police as Daniel Vorcaro's financial operator — in 2025 the company made seven transfers, totaling US$12.3 million, to the Havengate fund, linked to the film Dark Horse. | No dedicated poll. The closest measurement is Dark Horse's: 47.9% say the link to Vorcaro makes them less likely to vote for Flávio (Meio/Ideia, Sep 4–7), with no corresponding drop in voting intention. | Unfavorable to Flávio. The operation is about PCC money laundering in fuels and does not accuse him of anything, but it places the man handling the film's money inside the same inquiry investigating organized crime — a combination Lula's campaign is likely to exploit. Vorcaro and Master also splash onto the government (the 2024 meeting at the Planalto), which limits its use. | Small. The pattern since Sep 10 is that new facts in the Dark Horse case increase the stated damage without moving votes: those who say they are affected mostly weren't voting for Flávio anyway. |
No voting-intention poll published through Sep 26 has fieldwork after the betting decree (Sep 25) or the Aparecida controversy; Real Time Big Data (Sep 19–23) and Datafolha (Sep 22–23) came before. The first waves with fieldwork from Sep 25 on will show whether any of these events moves the race. Sources: Exame, Brasil de Fato, Gazeta do Povo and Estado de Minas (betting); CNN Brasil, CartaCapital, Terra and Congresso em Foco (Aparecida); Agência Brasil, CNN Brasil, Brasil de Fato and Revista Fórum (Carbono Oculto); AtlasIntel/Bloomberg and Palver (opinion).
| Theme | What's new (Sep 26–28) | Reading |
|---|---|---|
| Impact measurement | Genial/Quaest (Sep 24–27, TSE BR-06520/2026, n=2,004, released today) is the first national poll with all fieldwork after the betting decree (Sep 25) and the Aparecida rumor. Runoff: Lula 42% × Flávio 42% (was 41 × 42 on Sep 17–20); blank/spoiled 13%, undecided 3%. First round: Lula 39% (+2) × Flávio 34% (+1). Quaest did not publish a dedicated question on either topic. | Consistent with the small pro-Lula effect the dashboard anticipated (0 to +1 point), but within the margin of error: the shift cannot be attributed to the betting ban. Flávio has now been at 42% in three straight Quaest waves — the Aparecida episode did not bring him down. |
| Betting ban | The betting industry is organizing a challenge to the measure at the STF; in Congress, review is expected to wait until after the first round. Withdrawals remain allowed until Oct 5. | Nothing that changes the race before Oct 4. The risk for Lula (the measure struck down or suspended) shifts to the runoff. |
| Flávio and Aparecida | On Sep 26 Flávio's campaign issued a formal denial of the rumor. No dedicated poll. | The issue remains latent for Oct 12 (the patron saint's day, in the middle of the runoff). |
| STF / Moraes–Vorcaro | Flávio Dino's review request still has no return date. There are reports that he would return the case only after the runoff (not confirmed by two sources). | The uncertainty carries through the first round, as the dashboard already assumed. |
| Dark Horse / Carbono Oculto | No new verified fact after the third phase (Sep 24). The plea deal of "Mineiro" (Entre Investimentos) with the Prosecutor General's Office, which was circulating as news, was checked on Oct 1: the agreement dates from Sep 3 and was approved on Sep 9, so it predates this week (see the box "Fact check · September 26 to 29"). | An earlier fact, not a new one. The claim that the plea deal names Flávio as the one who asked Vorcaro for the money remains unconfirmed. |
| Master and INSS | No relevant new facts found for Sep 25–28. | — |
| Next trigger | First-round vote on Oct 4. Globo's Oct 1 debate was canceled (see the Sep 30–Oct 1 box). | Final eve-of-election waves (Oct 3) already in the database: PoderData, Gerp, CNT/MDA, Palver, DataTrends, Datafolha, Quaest and Futura. The national AtlasIntel poll of Sep 27–Oct 2 (TSE BR-00999/2026) had not been located as of 9 p.m. |
Sources: O Tempo, Correio Braziliense, TVT News, Imirante and Revista Fórum (Quaest Sep 24–27; Flávio's first-round figure appears as 34% in Imirante and 35% in O Tempo — the dashboard uses 34%). Polymarket checked on Sep 28.
| Theme | What's new | Reading |
|---|---|---|
| Aparecida: the courts | André Mendonça, vice president of the TSE, classified as "notoriously false" the posts claiming that Flávio, if elected, would strip Nossa Senhora Aparecida of the title of patron saint, and ordered more than a thousand posts removed, among them one by comedian Antonio Tabet. At the STF, Flávio Dino suspended the decision on free-speech grounds in electoral debate and sent the case to the full court. It is the second time in less than a month that Dino has overturned a Mendonça decision. Fachin scheduled a meeting between the STF and the Electoral Court to resolve the jurisdictional conflict. | The case is no longer just a rumor and has become another chapter in the institutional conflict that already involves the STF (Moraes–Vorcaro). That keeps the issue in the news through the first round, but also gives Flávio the role of a victim of disinformation. |
| Aparecida: campaigns and the Church | In an interview with the SantoFlow podcast (Sep 28), Flávio called the rumor "a pack of lies," spoke of a "spiritual war" to turn Catholics against evangelicals and compared the uproar to the stabbing his father suffered in 2018. There were also reports of evangelical discomfort with the image of Flávio holding up a statue of Our Lady (Diário Carioca). The CNBB denied any role in spreading the content ("a lie is not fought with another lie"). | Flávio needs to speak to Catholics without losing evangelicals, an equation the episode makes costlier. His gestures toward Aparecida are the visible cost. |
| Aparecida: social media | The Instituto Democracia em Xeque counted 977,600 posts about the episode in 55 hours, with 8.5 million interactions. | High social-media volume is not electoral impact: the institute itself makes that caveat. |
| Measurement in the polls | Three national polls have all or part of their fieldwork after the rumor (Sep 24–25) and the betting decree (Sep 25). Genial/Quaest Sep 24–27: 42 × 42 (Lula +1, Flávio =). BTG/Nexus Sep 25–27: Lula 46 × 44 (Flávio −1). AtlasIntel/Bloomberg Sep 23–28: Flávio 47.7 × Lula 47.6 (was Lula 47.7 × 47.4 on Sep 17–22). None published a question about Aparecida. | No measurable effect: all three shifts are under 1 point and point in different directions. If the episode cost Flávio anything among Catholics, the cost was offset or stayed within the noise. The remaining test is Oct 12, the patron saint's day, in the middle of the runoff. |
Sources: Brasil de Fato, CNN Brasil (Matheus Teixeira's blog), O Hoje, Revista Oeste, The Intercept Brasil and Diário Carioca (Aparecida); Exame and Misto Brasil (AtlasIntel Sep 23–28); Nexus and Poder360 (BTG/Nexus); O Tempo and CNN Brasil (Quaest).
| Theme | What's new | Reading |
|---|---|---|
| Globo debate (Oct 1): canceled | Lula had already said on Sep 30 that he would not attend and gave an interview to Flow at that time slot. On the afternoon of Oct 1, Justice Estela Aranha (TSE), at the request of Lula's coalition, suspended two debate rules: the empty podium labeled with the absent candidate's name and the round of questions directed at him. At 7:26 p.m. Flávio announced on social media that he would no longer attend, "because of the censorship imposed on the event by the TSE." Minutes later, Gilmar Mendes (STF) ordered Globo to include Renan Santos (Missão), after Zema had secured a spot in court. Shortly before 9:30 p.m. Globo canceled the debate, citing "legal uncertainty" and interference with the freedom to organize debates. | Lula and Flávio reach the first round without having debated on TV. The Oct 1 debate was the only remaining opportunity, and it didn't happen. Flávio trades exposure to attacks from Caiado, Zema, Cury and Renan for a narrative of TSE "censorship." Lula avoids the empty chair. There is no poll on the effect of skipping debates in this election, and the model doesn't change. The point to watch is the third way, which loses its biggest showcase and depends on TV airtime and social media to peel tactical votes away from Flávio on Sunday. |
| Debate canceled: reaction | Flávio speaks of TSE "censorship." According to InfoMoney, his campaign was counting on Lula's empty podium as the centerpiece of its participation. Globo criticized "undue interference" by the courts. Renan Santos said Lula and Flávio "ran away once again" and that Flávio pulled out 3 minutes after his own spot was made official. CNN reports the reverse order: Flávio's announcement came minutes before Gilmar's decision. Before the cancellation, Flávio's allies feared he would become the main target (CNN, InfoMoney). There was no public reaction from Lula, Caiado, Zema or Cury as of the last check. | The fight shifted from a clash between candidates to a conflict with the judiciary, which already involves the STF (Moraes–Vorcaro and Aparecida). That reinforces the Bolsonarist narrative that the courts work in Lula's favor. Historically, the effect on votes tends to be small. Renan's reaction shows the third way will exploit the two front-runners' "flight" until Sunday. |
| Latest poll | Real Time Big Data (Sep 26–30, released Oct 1) shows Flávio 46% × Lula 45% in the runoff and Lula 43% × Flávio 39% in the first round, with both up 2 points in the first round from the previous wave. It is the fourth runoff poll with fieldwork after the betting decree and the Aparecida rumor in which the gap is between −1 and +2 points. | Still no measurable effect from the events of Sep 25–28. No poll published a dedicated question about them. |
| Datafolha (Sep 28–Oct 1) | Released on Oct 1 at 7:05 p.m. (TSE BR-08039/2026, n=2,506, in person): runoff Lula 48% × Flávio 45% (was 47 × 45), blank/spoiled 6%, undecided 1%. First round: Lula 42% (+2) × Flávio 38% (+2), Cury 4%, Caiado 3%, Renan 3%, Zema 1%. It is Lula's largest runoff lead in Datafolha since August and the pollster's first wave with all fieldwork after the betting decree and the Aparecida rumor. Against other opponents: Caiado 48 × 42, Zema 49 × 40, Renan 48 × 39, Cury 47 × 42. | One more point for Lula, within the margin of error. In the first round both rise together, taking votes from the third way: Cury and Caiado lose a point each. Polarization closes in before Sunday, without changing the gap between the two. |
| Eve of the vote (Oct 3, morning) | PoderData/Aya (Sep 30–Oct 2, TSE BR-03519/2026, n=4,000, phone): runoff Lula 46% × Flávio 46% (50 × 50 in valid votes); first round Lula 42% × Flávio 41% (45 × 43 in valid votes), Cury 3%, Renan 3%, Caiado 2%; rejection Lula 47%, Flávio 46%. Gerp (Sep 30–Oct 2, TSE BR-00509/2026, n=2,400, phone): runoff Flávio 49% × Lula 44%, blank/spoiled 6%, undecided 1% (was 50 × 43); first round Flávio 46% × Lula 44% in valid votes (no totals published, so left out of the first-round database). MDA/CNT, AtlasIntel, Futura, Palver, Quaest and Datafolha release throughout the day. | Both confirm the tie: PoderData goes from Flávio +1 (46 × 45 on Sep 23) to a tie, and Gerp, the most Flávio-friendly house in the database, narrows his lead from 7 to 5 points. |
| Eve of the vote (Oct 3, afternoon) | CNT/MDA (Sep 30–Oct 2, TSE BR-04756/2026, n=2,007, in person): runoff Lula 47.3% × Flávio 43.1%, blank/spoiled 7.6%, undecided 1.9% (was 47.3 × 40.0); first round Lula 43.1% × Flávio 38.0% (47.8 × 42.1 in valid votes), Cury 3.2%, Caiado 3.0%, Renan 1.5%. Palver (Sep 30–Oct 3, TSE BR-00198/2026, n=5,000, online): runoff Flávio 49% × Lula 44%, blank/spoiled 6%, undecided 1% (was 47 × 45); first round Flávio 47% × Lula 43%, Renan 7%. DataTrends (Sep 28–30, TSE BR-04414/2026, n=2,000, automated phone, released Oct 2): runoff 45% × 45%; first round Lula 41% × Flávio 39%. | The three are split: CNT/MDA shows Flávio up 3 points without taking from Lula, Palver moves to Flávio +5, his biggest lead in the pollster's series, and DataTrends stays tied. The first round remains polarized: Lula and Flávio add up to 80% to 90% in all of them. |
| Eve of the vote (Oct 3, evening) | Datafolha (Oct 2–3, TSE BR-01708/2026, n=4,006, in person): runoff Lula 47% × Flávio 46%, blank/spoiled 6%, undecided 2% (was 48 × 45); first round Lula 42% × Flávio 40% (45 × 42 in valid votes), Caiado 4%, Cury 3%, Renan 3%. Quaest (Oct 2–3, TSE BR-02197/2026, n=3,702, in person): runoff Flávio 44% × Lula 42%, blank/spoiled/won't vote 13%, undecided 1% (was 42 × 42); first round Lula 40% × Flávio 38% (46 × 45 in valid votes), Cury, Renan and Caiado at 3% each. Futura/Apex (Oct 2–3, TSE BR-02431/2026, n=2,000, phone): runoff Flávio 48% × Lula 45.1%, blank/spoiled 5.7%, undecided 1.2% (was 49 × 43.5); first round Flávio 44.8% × Lula 42.6% in valid votes (no totals published, so left out of the first-round database); rejection Lula 46.9%, Flávio 43.1%. The national AtlasIntel poll of Sep 27–Oct 2 was not located. | Datafolha and Quaest move in the same direction: Flávio gains 1 to 2 points in the runoff and narrows the first round to 2 to 3 points. Quaest again shows Flávio numerically ahead in the runoff, as on Sep 13 and 20. In Futura, the movement is the reverse (Flávio's lead falls from 5.5 to 2.9). |
| STF, Master, Dark Horse, Carbono Oculto, INSS | Nothing new confirmed by two sources on Sep 30–Oct 1 about the Moraes–Vorcaro case: Dino's review request still has no return date. The events of Sep 26 to 29 (jingle, Dark Horse forensic report, betting at the STF, Lulinha's trip) are in the next box. | No change in the case; see the Sep 26 to 29 box. |
Sources: O Povo, Poder360 and Correio Braziliense (debate); CNN Brasil, InfoMoney, Estado de Minas and Jovem Pan (Flávio's withdrawal and the cancellation); O Antagonista (Renan Santos); Transmissão Política, AC24horas, Gazeta do Povo and Metrópoles (Real Time Big Data); g1 (Datafolha of Oct 1).
| Theme | What's new (Oct 5) | Reading |
|---|---|---|
| Runoff endorsements | Romeu Zema (Novo, 0.27%) declared his vote for Flávio in a video on the night of Oct 5: "I vote against PT, I vote for Flávio Bolsonaro." Augusto Cury (Avante, 2.89%) is endorsing no one and barred both campaigns from using his name. Renan Santos (Missão, 2.24%) is not asking voters to back either one and freed his party to choose, but says Flávio will be "the next president." Ronaldo Caiado (PSD, 2.18%) will decide after talking to Kassab. PSD freed its local branches and members, and Kassab says the party will stay neutral. Estado de Minas reports that Caiado will announce support for Flávio on Oct 6, in Goiânia (announcement expected, not yet made). | Only the endorsement with the least electoral weight was confirmed. The three who together hold 7.3% of valid votes didn't ask anyone to vote for either candidate. The model already assumes this electorate migrates about 1.3 to 1 toward Flávio, based on the stated transfer in post-first-round polls. A formal Caiado endorsement would reinforce that assumption, but not change it. At the ballot box, Flávio has already converted more of this vote than the second-choice question indicated. |
| Polls | No national runoff poll with fieldwork after Oct 4 had been released as of the night of Oct 5. Announced schedule: Datafolha (TSE BR-02949/2026, fieldwork Oct 6–8, n=2,520) and PoderData (BR-08134/2026, fieldwork Oct 5–7, n=3,000) on Thursday, Oct 8; AtlasIntel (BR-03663/2026, n=5,000, online) and Vox (fieldwork Oct 5–7, n=2,100) on Friday, Oct 9; Ipespe (fieldwork Oct 6–8, n=1,500) on Saturday, Oct 10. For AtlasIntel, ND Mais gives fieldwork starting Oct 3, before the vote (check the registration). | Until Oct 8, the dashboard's runoff estimate still rests on the eve-of-election waves, corrected for the error the anchor showed at the ballot box, and on the counted vote. Waves with fieldwork before Oct 4, as part of AtlasIntel's may be, enter under the rule for old waves. |
| Market | Polymarket had Lula at 16% on the night of Oct 5 (15.5% to 16.5% across three mirrors; Kalshi 18%), with US$173.7 million traded. It was 36% on the eve and 15% on the night of the count. | The market settled at the level of the night of Oct 4, with no new poll to move it. |
| Next trigger | First polls on Oct 8 and 9. First debate between Lula and Flávio in this election: Band/Estadão, Oct 11, 8 p.m. Record scheduled another for the same night, and Globo closes the series on Oct 23. Both campaigns signal they will attend the debates, without formal confirmation. | It will be the two candidates' first direct face-off on TV. In the first round, both skipped every debate. |
Sources: CNN Brasil, Estado de Minas and Midiamax (endorsements); ND Mais and CartaCapital (poll schedule and TSE registrations); CryptoSlate, PredictMarketCap and Rio Times (market); Terra/Estadão, A Crítica and Tribuna do Sertão (debates); CNN Brasil, Jussara Soares's blog (debate attendance).
| Event | What is confirmed | Reading |
|---|---|---|
| TSE suspends the "Foi o Flávio" jingle | In an injunction by the TSE president, Nunes Marques (Sep 26 or 27; outlets differ on the date), new airings of the Lula campaign jingle linking Flávio to the Master case were suspended. The grounds were "serious decontextualization": the material treated the "rachadinhas" (salary kickback) case, already closed, as current. The jingle had aired on Aug 28–31 on four broadcasters (seven spots), and the decision stands until the full court rules. | It takes away from Lula a line of attack linking Flávio to Master, but it targets material from August, not the Dark Horse investigation, which is still ongoing. Probably a small electoral effect: the jingle was no longer on the air. |
| Federal Police forensic review of the Dark Horse material | The Federal Police is redoing the forensic analysis of material seized by the São Paulo Civil Police in Operation Wi-Fi (June), under an order Flávio Dino issued on Sep 13; the news came out on Sep 28 and the material reached the Federal Police on Sep 22. What is being investigated is the diversion of public funds by the film's production company: earmarks from congressman Mario Frias (R$2 million to ICB) and R$750,000 passed on to the production company Go Up, plus the WiFi Livre SP program. | It points to nothing against Flávio: the probe concerns public money and the production company, not Vorcaro's private financing. It keeps the film in the news through the first round. Correction: it was circulating that the R$750,000 was the basis for Dino's order; it is a transfer identified along the chain, and the order is from Sep 13, not Sep 28. |
| "Mineiro's" plea deal | Not news from this week. The Prosecutor General's Office closed the deal on Sep 3 and Justice Mendonça approved it on Sep 9. Antônio Carlos Freixo Júnior made seven transfers, totaling US$12.3 million, to the Havengate fund (Texas), between Feb and Sep 2025, at Vorcaro's request. He said he did not know the final destination. The fund is managed by Paulo Calixto, who is close to Eduardo Bolsonaro. | Not confirmed: a smaller outlet claims the plea deal names Flávio as the one who asked Vorcaro for the money; the full text that could be read does not mention him. A figure of "R$69 million" also circulated, which appears in no source; the confirmed amount is US$12.3 million (≈ R$62.8 million). |
| Betting: petition to the STF | On Sep 28, ANJL and IBJR (industry associations) asked the STF to suspend the betting decree, in a petition filed within the existing cases on the legal framework, addressed to Justice Fux. There is no injunction. Congressional leaders say the issue should only be discussed after the elections, and the decree is valid for up to 120 days. Withdrawals until 11:59 p.m. on Oct 5; the platforms go offline on Oct 6; refunds take place between Oct 9 and 14. | The decree's legal risk, flagged in the previous box, has begun to materialize, but with no effect before the first round: the ban takes effect with the Oct 4 vote. An STF injunction before then would be the biggest possible surprise. |
| Lulinha: trip paid for by a lobbyist | On Sep 29 the Federal Police revealed a R$1.8 million trip to Europe by Fábio Luís ("Lulinha"), paid for by a lobbyist. Gazeta do Povo reports that he is only expected to be called to testify after the runoff. | It adds a concrete fact to the inquiry that already weighed on Lula over the INSS case, in the week of the vote. The expectation that he will only testify after the runoff reduces its immediate electoral use. |
| Debates and the third way | Record canceled its Sep 27 debate after Lula and Flávio said they would not attend. For Globo's Oct 1 debate, Flávio, Caiado, Zema and Cury were confirmed, with Lula absent; the debate ended up canceled (see the Sep 30–Oct 1 box). On Sep 29 Flávio asked Zema and Caiado to drop out: Zema refused immediately and Caiado said "I'm not backing down." No candidate dropped out or made a formal endorsement; Cury and Renan Santos rule out backing either of the two in the runoff. | The pressure on the third way shows Flávio's campaign wants to avoid a split right-wing vote, currently the main difference between the two in the first round. With no withdrawals, vote transfer is left to the runoff, which is what the model already assumes (migrating about 1.3 to 1 toward Flávio, based on the stated transfer in post-first-round polls). |
Fact check done on Oct 1; each row was read in at least two outlets, and unconfirmed claims were flagged as such. Sources: Poder360, Metrópoles (jingle); Gazeta do Povo and CartaCapital (forensics, Go Up); Brasil de Fato and Times Brasil (plea deal); InfoMoney and Tribuna do Sertão (betting at the STF); Agenda do Poder and Diário 360 (Lulinha); Gazeta do Povo, Poder360, O Tempo and CartaCapital (debates and third way). The two most-cited claims on social media about Dark Horse (R$69 million and Mineiro's confession as a new fact) did not survive the check.
In terms of blame assigned, the Banco Master case weighs most heavily on Lula (54%×29%), but it was never measured as "makes you less likely to vote for him." For Flávio, that effect was measured: 47.9% say the Dark Horse case makes them less inclined to support him (Meio/Ideia, Sep 4–7), the largest stated damage among the cases. The new finding in this update is that the stated damage does not show up in the numbers: in the four polls with fieldwork after the inquiry was disclosed (Sep 10), Flávio holds steady or ticks up, and the two candidates' rejection ties. Two readings are consistent with this: (1) those who say the case reduces their vote for Flávio weren't voting for him anyway; (2) the STF crisis, by putting Moraes at the center, reframes the inquiry as part of a "Judiciary × Bolsonarism" fight, which helps Flávio hold his base. INSS remains the most "shared" case: it originated under the previous administration, but the political cost falls on whoever governs (31%×8%), now with a formal inquiry into Lulinha. The Sep 15 STF session, which was the next trigger, resolved nothing: the court stalled on a point of order (4×3 for keeping the cases separate; Dino's vote would have tied it), Dino asked for more time to review and has up to 90 days to return the case — meaning the uncertainty over the validity of the Federal Police report and over investigating Moraes carries through the first round. That is, in itself, an outcome with electoral effect: it keeps the "shielding" narrative alive without producing a new verifiable fact against anyone, and leaves both sides free to use the case as they wish. On Sep 16, Lula chose to defend Moraes by name and to tie Flávio to Vorcaro (the R$130 million for the film); Flávio keeps merging the two figures ("Lula is Moraes and Moraes is Lula"), treating the judiciary as a campaign opponent. The markets, which had been falling for Lula since Sep 8, did not bounce after the session and kept falling until Sep 21 — 45% on Sep 16–17, 43% on Sep 18, 40% on Sep 21 — even with no news from the STF (Dino's review still has no return date); on Sep 23 the series stopped falling for the first time (41%) and on Sep 24 it rose to 42%, with AtlasIntel/Bloomberg showing a statistical tie (Lula 47.7% × Flávio 47.4%) and PoderData/Aya with Flávio 46% × Lula 45% (Sep 20–23), the pollster's narrowest margin of the month. The simplest reading is that the market is reacting to the accumulation of polls (Gerp, Futura, Palver, Quaest, AtlasIntel and PoderData) more than to any single event in the institutional crisis, and the Sep 23–24 recovery suggests the new statistical tie is starting to be priced in.
These polls are one-offs and are not part of the dashboard's weighted aggregator (which covers only the Lula×Flávio race in runoff polls) — they are cited for context, not for the model. Focus-group data (LEMEP/POMT) are not statistically representative by design; they help understand arguments, not magnitude. No TSE registration was found for the Quaest/Genial poll of May–Jun 2025 or for the Meio/Ideia wave of May 2026 (only for the Sep 2026 one, cited above). The STF crisis is treated here as a reported institutional development (CNN Brasil, InfoMoney, O Tempo, Metrópoles, Correio Braziliense, Congresso em Foco, Times Brasil, Rio Times, Revista Oeste, Band; no dedicated opinion poll as of the data date). Two waves with fieldwork through Sep 16 (PoderData/Aya and AtlasIntel) already include days after the Sep 15 STF session, but no question dedicated to the case has been published yet.
In one sentence: against any other right-wing name, Lula does better; Flávio is the most competitive opponent and makes the runoff in 100% of simulations.
| Opponent | Lula | Opponent | Lula's lead | Source |
|---|---|---|---|---|
| Ronaldo Caiado (PSD) | 44–48% | 37–44% | +2.6 to +7 pp | Datafolha (g1, Oct 1: 48×42) / Quaest / PoderData / Atlas |
| Romeu Zema (Novo) | 44–49% | 33–42% | +4 to +14 pp | same |
| Renan Santos (Missão) | 43–48% | 26–39% | +7 to +21.5 pp | same |
| Augusto Cury (Avante) | 42–47% | 39.4–45% | −3 to +6.8 pp | AtlasIntel / PoderData / Datafolha (47×42 on Oct 1) |
| Jair Bolsonaro (PL, hypothetical) | 46.8% | 44.1% | +2.7 p.p. | AtlasIntel |