1. Prediction market

With 67 days to the first round, total accumulated volume in the presidential market is around USD 116.49M (Polymarket via the AFOS proxy, capture of July 29 at 7:50pm BRT). Lula fell 1.00pp to 62.50% (USD 7.73M). Flávio Bolsonaro rose 0.10pp to 23.95% (USD 7.70M), and the gap between the two eased to +38.55pp.

Lula's number needs two sentences, and the order matters. This is not a drop in level: 62.50% is the same value as July 25 and July 27, and the top of the AFOS series marked 63.50% on July 26 and July 28. What happened was the return of the point taken the day before. By the daily close of the series, today's gap is the narrowest since July 24, and the widest gap in the series remains +39.80pp, from July 26.

In the pack, the day's move belongs to Ronaldo Caiado, who rose 0.45pp to 2.35% (USD 5.20M), a value above every daily close of the series since June 20. Renan Santos fell 0.05pp to 8.70% (USD 8.59M), Jair Bolsonaro stayed at 1.05% (USD 5.26M), Romeu Zema rose 0.20pp to 0.75% (USD 4.57M), Geraldo Alckmin rose 0.10pp to 0.35% (USD 4.94M), Fernando Haddad stayed at 0.25% (USD 6.38M) and Tarcísio de Freitas at 0.15% (USD 13.66M), still the highest accumulated volume in the book, a legacy anomaly that measures traded history and not current conviction. Michelle Bolsonaro enters today as a range of 0.35% to 0.55% (USD 9.34M), and the reason is in the capture note below.

In the sub-markets, the signal inverted against the previous day. In first-round second place, Flávio rose 0.50pp to 78.50% (USD 4.34M in the book), Renan fell 0.45pp to 11.70%, and Haddad rose 0.30pp to 1.15%. In third place, in a book of USD 484,000, Caiado fell 1.00pp to 25.50%, and Renan recovered 1.00pp to 61.50%. Yesterday Caiado was rising in the placement book and falling in the winner book; today he did the opposite. The correct reading is that the market moved on the race for the presidency, not on the race for third place.

Among the other markets, impeachment of an STF justice before 2027 rose 0.55pp to 3.35% (USD 83,000), after falling 0.60pp the day before, and it was the only contract that followed the direction of the day's institutional pile-up. The caveat comes before the reading: 0.55pp in a book that size is very little money, so what is on the record is the direction and not the size. In the Senate by seats, the PL stayed at 71.00% (USD 292,000 in the book) and the MDB rose 0.35pp to 17.60%. In the 2026 inflation market, the modal band went back to 5.00% to 5.49%, at 38.30%, while the 4.50% to 4.99% band fell to 34.90%, in a book of USD 81,000. In other words, the shift this brief recorded on July 28 undid itself in a single session, and the inflation market remains outside the double-reading guard.

A capture note, and it is about process. The guard ran three times on July 29. It rejected the first round over a 1.00pp divergence in Renan Santos's third place, approved the second, and on the third, run at 7:50pm to check the market before the editorial close, it rejected again, this time only on Michelle Bolsonaro's book, which swung between 0.35% and 0.55%. Every other price repeated and enters firm; Michelle's enters as a range, with the divergence declared.

2. What the institutes recorded

The national poll with the largest sample in the window came out. AtlasIntel/Bloomberg (n=5,021, field July 22 to 27, 1pp margin, 95% confidence, filing BR-08602/2026, cost of R$ 80,000, self-funded) has Lula 44.9% and Flávio Bolsonaro 35.8% in the first round, with Renan Santos 7.8%, Ronaldo Caiado 3.1%, Romeu Zema 2.8%, Samara Martins 2.1% and Augusto Cury 1.6% (CNN Brasil, JOTA, CartaCapital). In a scenario reduced to five names, Lula goes to 47.4% and Flávio to 37.0%.

In the runoff, Lula wins the six scenarios he appears in, and here is the data point this brief is obliged to highlight: his tightest margin is not against Flávio, it is against Jair Bolsonaro, 49.2% to 43.9%, a difference of 5.3pp against the 6.3pp of the scenario with the son. Since the two scenarios sit 1.0pp apart, at the edge of the margin, this undoes no tie, and instead suggests that, by the same institute and in the same field period, the father's name measures equal or better. In the two scenarios without Lula, Haddad 44.3% x Flávio 43.7%, a statistical tie, and Alckmin 45.8% x Flávio 42.7%. Rejection is the round's hardest number: Flávio 52.9%, Lula 49.4% and Jair 42.6%. On approval, Lula has 47.6% against 51.2% disapproval, with approval rising 1.7pp and disapproval falling 1.1pp against the institute's own July 1 round (Poder360, JOTA).

What the comparison with July 1 shows is stability, not advance. Both front-runners fall, Lula from 46.3% to 44.9% and Flávio from 36.6% to 35.8%, and the first-round gap narrows from 9.7pp to 9.1pp. In the direct runoff, the gap moves from 6.5pp to 6.3pp. With a 1pp margin, each of those moves fits inside the noise. What moves outside the margin is the government rating, and that is precisely where the poll licenses a claim.

Among the state polls, Genial/Quaest published rounds in seven states, and the one that cuts across the country is São Paulo: Flávio 34% x Lula 30% in the first round and 40% x 35% in the runoff, in the largest electoral college (n=1,650, field July 23 to 27, BR-09998/2026), with Tarcísio ahead of Haddad 41% to 26% in the state race (G1). In the Northeast the picture is the reverse: Lula has 52% against 18% in Bahia (G1) and 54% against 23% in Ceará, per Ipsos-Ipec (n=800, field July 23 to 26, BR-07235/2026) (Poder360). None of those enters this brief's national window, and all of them together say the same thing: the national average hides a split country (G1).

📅 Polling calendar — next 7 days

Polls registered with the TSE scheduled for publication between July 30 and August 5. Inclusion in the table does not mean publication is confirmed, because institutes can delay or cancel a release. Filter applied: sample ≥ 1,000. Each filing is linked to the TSE public search.

Date	Institute	Sample	Scope	TSE filing	Conf.
Jul 30	PoderData	2,400	national	BR-07845/2026	0.7
Jul 30	Real Time Big Data	2,000	state	BR-00968/2026	0.9
Jul 31	Alfa Inteligência	2,700	not declared	BR-04488/2026	0.7
Jul 31	Vox Brasil	2,100	national	BR-01084/2026	0.7
Aug 2	Veritá	2,010	state	BR-02249/2026	0.7
Aug 3	Nexus	2,000	national	BR-02874/2026	0.7

Source: public TSE registry via the AFOS API. 🔥 marks samples ≥ 3,000, and this window has none. The window holds 25 filings with a sample ≥ 1,000; the table shows the 3 with national scope, in bold, plus the 3 largest among the rest, and the other 19, almost all state-level, were left out for legibility and not for absence of a filing. Status "filed ≠ published": confirming an actual release requires checking two primary sources before citing numbers.

3. What the press covered

The day's institutional fact was born inside the PL's own convention, and its structure is what matters. On July 28, Alexandre de Moraes gave 48 hours for Jair Bolsonaro's defence to explain the video generated by artificial intelligence carrying the former president's image and voice, and he laid out two mutually exclusive scenarios: if there was authorisation, the release may amount to a fresh breach of the precautionary measures that sustain his house arrest, with a risk of return to closed custody; if there was not, the piece may amount to a deep fake barred by electoral law, with liability for Flávio Bolsonaro and the party (Metrópoles, Terra).

On July 29 the defence replied that he did not authorise the use of his own image and voice, and added that he does not object to relatives producing this kind of content, a practice the filing describes as notorious and continuous, arguing further that he could not have authorised anything because his visits are suspended for thirty days and Flávio is barred from visiting him for ninety (CNN Brasil, Metrópoles, Folha de S.Paulo). It is worth being precise about what that answer is and is not: it is not the father contradicting the son, it is a defence against the allegation of breaching a precautionary measure, and its effect is to move the exposure from the first scenario to the second. The same day, Flávio's campaign told the TSE the video is legal and does not deceive voters, while TSE justices came to see abuse in the use of AI in campaigning and the STF prepared to act should the TSE not punish.

In the Master case, the news is about the quality of the information and it carries a date correction. The authorisation to trace Daniel Vorcaro's assets abroad, which this brief recorded on July 28 as one newsroom's exclusive, appeared in several, and the decision itself was signed in May, becoming public only on July 28. The route is international cooperation under a mutual legal assistance treaty, run by the Justice Ministry's asset recovery department, and among the targets cited is a yacht valued at R$ 500 million (Gazeta do Povo, Imirante, Diario de Pernambuco). On July 28 this brief treated the act as if it were from the previous day, and the correct date goes here.

On the external axis, a correction of what this brief published on July 28 is due. We wrote that the US national emergency is the legal basis of the tariffs. It is not: the 40% tariff based on the IEEPA ceased to apply in February 2026, after the US Supreme Court ruled that the statute does not authorise tariffs, and what is in force today is 25% under Section 301 plus 12.5% announced in July following a forced-labour investigation (InfoMoney, Estado de Minas). The one-year extension, in a document dated July 29, creates no new sanction. What changed is the reading: the Brazilian government came to assess the extension as possible preparation for new sanctions, and the PL campaign leadership expects US measures against Moraes. That is a government assessment, not a settled fact, and it enters as such.

In the electoral race, the day produced three records. The São Paulo electoral court fined Lula R$ 15,000 for early campaigning and ordered the video taken down. Ronaldo Caiado adopted the strategy of attacking Lula and Flávio in equal measure, and on July 28 Tarcísio de Freitas freed allies and base mayors to back his candidacy, while still saying that Tarcísio's own candidate remains Flávio (CNN Brasil). In Minas Gerais, the Republicanos announced that Cleitinho will not run for governor, and the senator denies withdrawing. On the budget, Flávio Dino ordered the government and Congress to specify the responsibility of members of parliament over amendments, with a 10-day deadline. And Lula complained about corporatism among judges and prosecutors and argued for institutional reform.

4. Divergences of the day

Price × polling, and the inverse of the previous day: on July 28 there was no national poll and the favourite's price rose 1.00pp. On July 29 the window's largest-sample poll came out and the price fell 1.00pp. The market's direction matches the polling's, which also narrowed the gap, but the sizes differ: 0.6pp in the poll against 1.10pp in the market. And the same session carried the AI video case at the STF and the shift in reading on the US emergency. This brief records the coincidence of direction and does not claim the poll moved the price.

Convergence that narrowed on the right side: with 7.8% in the polling and 8.70% in price, Renan Santos's distance fell to 0.90pp. What changes is not the level, it is the mechanics: in previous weeks that distance shortened because the price gave way, while the polling stayed at 3%; in this round it shortened because the polling rose, and rose in the largest-sample poll of the window. They remain different quantities, which do not subtract from one another, and convergence is proof of no one being right. The dispersion between institutes, incidentally, did not resolve: 9% at Real Time on July 21, 7.8% at AtlasIntel, 6% at PoderData/Aya on July 16, 5% at BTG/Nexus on July 27 and 3% at Datafolha, Gerp and Indexa all still stand.

Price rising, polling worsening: Caiado rose 0.45pp in the winner contract on the same day AtlasIntel cut him in half, from 6% at BTG/Nexus on July 27 to 3.1%, and had him losing the runoff 48.2% to 38.9%. The divergence this brief recorded on July 27 between Nexus (45% to 43%) and Datafolha (47% to 40%) gained the third measurement it was missing, and it landed on Datafolha's side. Price and polling moved in opposite directions, and both go on the record with no side chosen.

Institutional pile-up × price, with the signal inverted against yesterday: on July 28 the institutional axis was dense and the impeachment market fell 0.60pp. On July 29 the axis was denser still, with the defence's answer to the STF, TSE justices seeing abuse, the STF preparing to act and the President of the Republic criticising corporatism in the judiciary, and the market rose 0.55pp, to 3.35%. Recording the direction is mandatory; building a narrative on it is not, because the whole book adds up to USD 83,000 and moves on very little money.

In summary

The window's largest-sample poll confirmed the lead and did not confirm an advance: against the institute's own previous round, both front-runners fall and the gap narrows 0.6pp, all inside the 1pp margin. What moves outside the margin is the government rating.
The day's method finding is the mechanics of convergence in Renan Santos, not the level. The distance fell to 0.90pp because the polling rose, and that is different from every previous week, when it fell because the price gave way.
The institutional axis inverted the previous day's signal: more fact, and the price rose 0.55pp instead of falling. In a book of USD 83,000, that is a recordable direction and nothing beyond it. This edition also carries two record corrections from this very series: the legal basis of the US tariffs and the date of the decision on tracing Vorcaro's assets.

Fontes consultadas

matérias com link direto para a notícia (veículos âncora):

matérias secundárias (URL Google News redirect — clique resolve à matéria):

Fontes técnicas: Polymarket (live quotes via the AFOS proxy, fetched Jul 29, 7:50pm BRT), TSE registry (official electoral polls).

July 29, 2026