AFOS Daily · Daily Synthesis

September 3, 2026

Prediction Markets × Polls × News

Synthesis generated from auditable data. Every claim cites its source.

Three national polls in 48 hours disagree on who wins a runoff, and all three margins fit inside the margin of error. Datafolha is the fourth top-tier house to measure Augusto Cury's jump and the only one that leaves him below double digits. And in a presidential book of USD 142.61M, a single one of nineteen contracts moved after 5:28pm.

1. Prediction market

Total accumulated volume in the presidential book adds up to USD 142.61M, in a confirmed reading of 3 September, 23:45 BRT. Since 17:28, a single one of the book's nineteen contracts moved: Flávio Bolsonaro rose 0.35pp to 42.85% (USD 9.64M), while Lula holds at 54.50% (USD 9.67M), flat since 15:22.

The gap between the two fell to 11.65pp, against 12.00pp at 17:28 and 16.55pp on 1 September, a fourth consecutive reading narrowing, always through movement on the runner-up's side. The larger move was in the side contracts. In the first-round runner-up contract, Flávio gave up 2.00pp to 83.50% (USD 527 thousand) and Lula rose 1.80pp to 10.50% (USD 439 thousand). In the third-place contract, Augusto Cury rose 3.60pp to 59.50% (USD 114 thousand) and Renan Santos gave up 4.50pp to 22.00% (USD 269 thousand), the panel's largest fall: the gap between the two there went from 29.40pp to 37.50pp in a few hours.

This round publishes no new price for the contract on a Supreme Court justice leaving through impeachment, the smallest book on the panel: the last confirmed reading is the 17:28 one, at 7.40% (USD 87 thousand). In the Senate, the Liberal Party holds at 82.50% (USD 261 thousand).

2. What the polling institutes recorded

Datafolha came out around 20:20 and is the third national poll in 48 hours. Registration BR-03669/2026, field 1 to 3 September, n=2,002, margin of 2 points. In the first round with Pablo Marçal, Lula has 39% and Flávio 32%; without Marçal, 38% and 33%. In the runoff, Lula 46% and Flávio 44%, and the house also puts the leader ahead against Ronaldo Caiado, by 46% to 41%, and against Romeu Zema, by 48% to 39%.

The comparison that counts is each house against itself: in the Datafolha poll of 21 August the runoff was 47% to 43%, so the gap fell from 4 to 2 points and now fits inside the declared margin.

The new element is Augusto Cury: 7% and 8%, against 2% on 21 August at the same house. It is the fourth top-tier house to measure the jump and the only one that leaves him below double digits, given that BTG/Nexus gave 11% on 31 August, Real Time Big Data 11% on 1 September, Genial/Quaest 10% on 2 September and PoderData 10% on 3 September. Folha de S.Paulo covered the point in an analysis arguing that the campaign wiped out the film's effect.

On government approval, Datafolha reports 45% approval and 51% disapproval. Genial/Quaest of 2 September measured 45% and 48%: approval matches and disapproval differs by 3 points.

Polling calendar — next 7 days

Polls registered with the TSE scheduled for publication between 04/Sep and 10/Sep. Inclusion does not mean confirmed publication, because institutes can delay or cancel. Filters: national scope and sample of at least 1,000. Registration numbers link to the TSE public search.

DateInstituteSampleScopeTSE registrationConf.
06/SepVeritá3,840nationalBR-09426/20260.7
07/SepGenial/Quaest2,004nationalBR-01720/20260.9
08/SepGerp2,400nationalBR-00251/20260.6
08/SepNexus2,000nationalBR-06790/20260.8

Source: public TSE register via the AFOS API. Samples of 3,000 or more in bold. The table lists the 4 NATIONAL polls in the window; 21 state-level registrations were left out by scope. The status is "registered is not published", and confirming actual release requires two primary sources before citing a figure.

3. What the press covered

The harshest act in the Supreme Court crisis so far: Alexandre de Moraes asked Edson Fachin to investigate André Mendonça in the fake news inquiry. The report underpinning the request was commissioned by Moraes from the Federal Police without the court presidency's approval, and Estadão records it as an internal document with no evidentiary value.

The response came the same night: Fachin asked for explanations from Moraes, Mendonça, prosecutor general Paulo Gonet and the Federal Police director general, with a five working day deadline, and postponed to the second half of September the full court session. Mendonça produced tailor's receipts against the suspicion of a gift from Daniel Vorcaro, and Gilmar Mendes argued for removing Federal Police officers from the justices' chambers.

Bolsonaro supporters came out in defence of Mendonça and Marina Silva argued that Moraes should step aside.

Outside the court, Lula declared the INSS queue cleared at a 3pm ceremony at the Planalto Palace. By the social security minister's definition, the queue counts as cleared when fewer people are waiting than the number who filed the previous month, and 261 thousand requests running more than 45 days late remain open in August, against 1.88 million in January. The Prosecutor General's Office closed a plea agreement with the businessman who made payments to the fund tied to the film about Jair Bolsonaro, an agreement closed and not yet ratified.

4. Divergences of the day

Market x poll x poll: three national houses measured the same runoff pairing within 48 hours and reached opposite signs. Genial/Quaest of 2 September put Lula ahead by 42% to 41%, PoderData of 3 September put Flávio ahead by 45% to 44%, and Datafolha the same day puts Lula ahead by 46% to 44%. All three margins are 1 or 2 points and all three houses declare a margin of error of 2 points. None of them asserts a lead, and the panel does not choose between them.

House x house: Augusto Cury's jump shows up in all five readings in the window and the level does not. Four houses put him at 10% or 11%; Datafolha puts him at 7% and 8%. Within Datafolha itself he went from 2% on 21 August, so there is no disagreement about direction. There is disagreement about size, of three to four points, and it is large enough to change the picture for anyone reading a single house.

News x price: the day produced the harshest act in the Supreme Court crisis, with one justice asking for another to be investigated, and the contract that measures a justice leaving is precisely the only one on the panel with no new reading to publish. The panel records the act and does not attribute a price to it.

In summary

  1. The price barely moved on a day dense with news. A single one of the nineteen presidential contracts moved after 17:28, and the gap between the top two fell to 11.65pp.
  2. Declared voting intention is more divided than any single headline suggests. Three national houses in 48 hours, three results in the pairing that decides the election, all inside the margin of error.
  3. The size of the Cury phenomenon depends on which house you read. Five readings agree on direction and disagree on the level by three to four points.

Sources consulted

stories with a direct link to the article (anchor outlets):

secondary stories (Google News redirect URL — the click resolves to the article):

Technical sources: Polymarket (live quotes via the AFOS proxy, fetched 03/Sep 23:45 BRT), registro TSE (official electoral polls).

Sources cited in this text: Polymarket, TSE (registro público), Gazeta do Povo, O Globo, Estadão, VEJA, Poder360, Metrópoles, Folha de S.Paulo, G1, CBN

Method: this synthesis is generated automatically from auditable data on the AFOS Analytics platform, under code-versioned rules in git. All claims can be verified on the platform or in the linked sources. Understand the automated governance.

Integration: for live data and detailed candidate analyses, access the full dashboard. To understand the method in depth, read The Method.

Glossary: Brazilian political terms used in the syntheses (TSE, STF, BolsoMaster, lideranças envelhecidas, etc.), definitions in 3 languages. See the full glossary

Get it by email

Every day, AFOS Daily cross-references prediction markets, polls registered with the Brazilian electoral court, the press and the day’s coverage.

No spam. Unsubscribe anytime.

After signing up you choose whether to receive it in English, Portuguese or Spanish.

AFOS Daily — 3 September 2026 | AFOS Analytics