AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
Week in which market and polling moved in opposite directions on the same candidate. On presidential market, Flávio Bolsonaro went from 23.90% on Monday to 27.10% on Friday, a rise of +3.20pp, the largest weekly variation of an individual contract on the panel since Edition №1. Lula opened and closed the week at 60.50%, with a peak of 61.50% on Wednesday, so the gap compressed from +36.60pp to +33.40pp entirely by the opponent's movement. The tradeable point: it was precisely in that window that the two largest national polls of the month took the field and measured Flávio declining within their respective series. [Datafolha](Genial/Quaest (BR-07181/2026, n=2.004, conf. 0.90)) gave 1º turno 40 × 28, below the 29% from June. [Quaest](PoderData/Aya (BR-00059/2026, n=2.400)) gave 40 × 34, below the 36% from June 25. The two disagree with each other by six percentage points at the level and agree on direction. The market repriced upward the candidate that both measured downward. Divergence of level between price and ballot is routine on the panel; divergence of direction in the same window is not. There is, in the audited data, no screenshot or ballot event that explains the rise, and the post-week helps characterize it: between July 18 and 19 Flávio returned 1.45pp of the 3.20pp, with no new fact. And on Sunday night came the structural data: Lula broke through the 60% floor for the first time in the July series, marking 59.50% at 23h30 UTC on July 19, with the gap at +33.85pp. Renan Santos, which Edition №8 signaled as price without support, delivered the correction: ↓2.30pp in the winner (10.10% → 7.80%) and ↓2.50pp in 3rd place of the 1º turno. The presidential contract accumulates USD 113.9M. The next window brings three nationals in two days and Datafolha on July 24, the only one with field after the 25% tariff confirmation by USTR.
1.Executive Summary
Week 13-17/July produced the type of divergence that the panel exists to capture, and it is qualitatively different from previous editions. In Editions №6 and №7 the tension to report was one of level: the market pricing in greater margin than vote margin sustained. Here the tension is one of direction. Flávio rose 3.20pp in the winner contract between Monday and Friday, while the two national polls that were in the field that same week measured him in decline within each house (Quaest from 29% to 28%, PoderData from 36% to 34%). Lula did not move at either end of the window, so all the compression of the gap, from +36.60pp to +33.40pp, came from the other side of the book. We identified no ballot trigger or dated event that explains the rise, and we register this as absence of known cause, not as hidden cause. Post-window behavior reinforces the reading of repricing without informational backing: between 18-19/July the contract returned 1.45pp of the 3.20pp gained, and on Sunday night the Lula side ceded the psychological floor of 60%, closing at 59.50%, the first close below that level in the July series. Market snapshot 19/July 23:21 BRT.
2.Why AFOS does not smooth
The aggregation industry would resolve the week with a single number: the average of the two national polls would give Flávio something close to 31% in the 1º turno. This number was not measured by anyone, does not correspond to any methodology, and would erase the two pieces of information that the window produced: the distance between the polls and their common direction.
Why it matters: divergence in level between implicit probability and stated intent is structural and has already been covered in Editions №6 and №7. An average of the two national polls would have produced ~31% and turned this week into another spread case. What the decomposition shows is something else: Quaest and PoderData disagree by six points in level and agree that Flávio retreated within each series, while the winner contract rose 3.20pp. Methodological caveat, mandatory: market price is probability of victory, not vote margin, so the legitimate comparison is direction and conviction, not pp by pp. It is precisely for this reason that the coincidence of direction, and not the distance in level, is the data of this edition. Second caveat: the Quaest field closed on July 13, before the USTR confirmed the 25% tariff on July 15, so part of the market window is subsequent to their field.
We also record what weakens the thesis, because it is not comfortable for us. The give-back of July 18 and 19 (27.10% → 25.65%) is consistent with the hypothesis that the increase was conviction noise, but it would also be consistent with a legitimate repricing partially reversed. Five trading sessions do not distinguish between the two. The three national polls on July 21 and 22 and the Datafolha of July 24 distinguish, and Edition №10 returns to this point with the result, whatever it may be.
The same framework we apply to Brazil - measuring the distance between the prediction market and polls and checking the signal against the actual election results - has already been validated against ten elections across four continents: France (2024, legislative, the most recent case in our archive), South Korea (2025, the first Asia case), Peru and Colombia (2026), Chile, Germany and Canada (2025), and United Kingdom, Mexico and United States (2024). In some cases the signal pointed to convergence, in others to divergence, and the method reports both without retroactive adjustment. The complete dataset, with methodology and code, is open under Apache 2.0 and has a permanent DOI in the Harvard Dataverse.
3.Weighted scenarios for the week
Scenarios for the July 20-24 window, which brings three national polls in two days (Real Time Big Data and Indexa on July 21, Gerp on July 22) and Datafolha on July 24, plus the opening of the party convention registration period on July 20. It is the first window of this cycle capable of directly arbitrating the divergence in direction of this edition.
As nationals of July 21 and 22 replicate the direction measured by Quaest and PoderData, and Flávio's price consolidates below 26%. In this case the give-back of July 18–19 extends, the rally of July 13–17 becomes characterized as conviction noise without support, and the gap reopens for the +35pp range. Lula's breaking of the 60% floor on July 19 proves to be a Sunday overshoot and is reversed. Net-neutral for BRL: confirms the current regime rather than altering it.
The Datafolha from July 24 (n=2.004, fieldwork July 22-24) validates the PoderData level, measuring Flávio in the 33-34% range and the runoff in Empate técnico. The reading flips completely: the rally of July 13-17 will have brought forward information that older fieldwork surveys had not yet captured, and the Quaest house effect, not the PoderData one, becomes the dominant hypothesis. This is the cleanest test of the cycle for the thesis that the market anticipates polling, and Datafolha is the only one in the window with fieldwork after the 25% tariff confirmation by USTR on July 15. Sustained compression of the gap below +30pp would be the signal. Risk of repricing assets sensitive to electoral risk.
As party conventions, whose deadline opens on July 20 and runs until August 5, produce ticket formation that reprices placement contracts. PL needs to close with Jair Bolsonaro under house arrest and Flávio suspended from visits to his father until mid-October by a July 13 decision. Novo arrives without a defined running mate, after Michelle Bolsonaro ruled out the position on July 19. A campaign rift, unexpected federation, or withdrawal in the third way group moves the second place contract (Flávio at 81.00%) more than any poll in the window. These are contracts with volume two orders of magnitude smaller than the winner contract, therefore rapid repricing with little flow.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| Flávio · winner | 27.10% | ↑3.20pp semana | 7.44M | Largest weekly variation of an individual contract since Edition №1, against the direction measured by the two national polls of the window. Returned 1.45pp on July 18-19 |
| Lula × Flávio gap | +33.40pp | ↓3.20pp semana | 113.9M | 100% compression by opponent; Lula immobile at 60.50% at both ends. Broke to +33.85pp on Sunday with Lula ceding the 60% floor. |
| Lula · winner | 60.50% | 0.00pp semana | 7.41M | Anchored throughout the week, with a peak of 61.50% on Wednesday. Broke below the 60% floor only on July 19 at 23:30 UTC, closing at 59.50% |
| Renan Santos · winner | 7.80% | ↓2.30pp semana | 8.14M | The correction signaled in Issue №8 has materialized; national polls measure it between 3% (Quaest) and 6% (PoderData) |
| Renan · 3rd place 1º turno | 68.00% | ↓2.50pp semana | 144k | Loss of conviction in the placement, not just in the winner. It fell another 4.00pp on July 19, to 64.50%, on the same trading day when Ciro Gomes mentioned him as a voting possibility |
| Flávio · 2nd place 1º turno | 81.50% | ↓2.00pp semana | 181k | Fell back in the return on the same week it rose 3.20pp on the winner: two contracts on the same candidate in opposite directions |
| Michelle Bolsonaro · winner | 2.20% | ↑0.10pp semana | 9.06M | Fourth largest volume in the book without declared presidential pre-candidacy; ruled out being Zema's running mate on July 19 |
| Ronaldo Caiado · winner | 1.30% | ↓0.20pp semana | 4.98M | Measures 4% in the 1º turno in BOTH national polls in the window, the most stable number in the field, and is worth 1.30% in the winner contract |
| Romeu Zema · winner | 0.60% | ↓0.30pp semana | 4.41M | Measures 4% in PoderData, two-thirds of what Renan measures in the SAME poll, and is worth about one-thirteenth of him in price |
| STF impeach < 2027 | 3.55% | faixa de ruído | 83k | Volume three orders of magnitude below the presidential; variations in tenths do not sustain interpretation. Real money continues to price friction, not rupture |
5.Liquidity and market structure
The three largest cumulative order book volumes remain in contracts priced below 0.20%: Tarcísio (USD 13.49M at 0.15%), Carlos Massa (USD 10.36M at 0.05%) and Eduardo Bolsonaro (USD 10.26M at 0.15%). These are legacy positions never unwound, accumulated volume rather than current traction. The reading rule applies to any use of this order book: accumulated volume measures historical interest, not current conviction, and whoever uses it as a proxy for current confidence will read the panel backwards. The relevant change from Edition №8 lies in 5th place: Renan Santos ceased to have double-digit probability (10.25% in №8, now 7.85%) while maintaining the volume lead among living contenders, with USD 8.18M against USD 7.46M for Flávio and USD 7.49M for Lula. He lost price without losing flow, which is the pattern of an orderly correction contract, not capitulation.
The top 5 accumulated volumes account for ~USD 51.4M (~45%) of the presidential market (total USD 113.9M). The cross-reading of price × volume for the week stands out: Flávio's 3.20pp increase was made with real and balanced flow between the two main contracts (USD 7.44M of his against USD 7.41M of Lula at Friday's close), meaning it was not a shallow book movement. This makes the noise hypothesis less trivial and is the main evidence against this edition's thesis, recorded as such. Low-volume spike (USD <500k) in individual contract should continue to be treated as noise until confirmation of recurring flow.
6.Calendar of price-relevant prints
| Date | Sample | Why it matters | |
|---|---|---|---|
| Mon Jul 20 | Opening of party conventions | não é pesquisa · prazo até 05/Ago | Converts pre-candidacy into registered slate; shifts placement contracts, the lowest-volume and highest-sensitivity items on the panel |
| Tue Jul 21 | Real Time Big Data (national) | n=2.000 · BR-09247/2026 · conf. 0.90 | First national poll after the rally; directly tests whether Flávio's 3.20pp increase had ballot support |
| Tue Jul 21 | Instituto Indexa (national) | n=2.000 · BR-02904/2026 · conf. 0.70 | Second national reading on the same day; allows cross-checking of fieldwork between two polling houses |
| Wed Jul 22 | Gerp (national) | n=2.000 · BR-05026/2026 · conf. 0.70 | House that produced a tie in the 1º turno on July 8 and Flávio winning the 2º turno; own series with a history of outlier favorable to him |
| Fri Jul 24 | Datafolha (national) | n=2.004 · BR-01166/2026 · conf. 0.90 · campo 22-24/Jul | The screenshot that determines the edition. Highest reliability of the window and only one with field after the 25% tariff confirmation by USTR on July 15 |
Source: TSE registration via AFOS API. Status 'registered ≠ published' remains in effect: dates are those declared by institutes at registration, and institutes may delay or cancel release. Note on Datafolha, because confusion circulated on the weekend of July 18-19: the institute did not release a presidential poll in July by the time this edition closed. The registration shows fieldwork from July 22-24 and articles from the period anticipate the survey without numbers. Confirmation of actual release requires two primary sources before citing any results.
7.Watch list, week triggers
- Datafolha of July 24 arbitrates this edition. It is the most reliable in the window (0.90, n=2.004) and the only one with a field after the 25% tariff confirmation by USTR on July 15. If it measures Flávio near 33-34%, the rally of July 13-17 will have anticipated information and the noise hypothesis falls. If it measures near 28%, the directional divergence of this edition confirms itself as repricing without backing.
- Lula's floor of 60% has been broken. Closed at 59.50% at 23:30 UTC on July 19, the first close below that level in the July series, with the gap at +33.85pp. If it sustains below 60% after the nationals on July 21 and 22, it ceases to be Sunday's overshoot and becomes a level shift.
- The two Flávio contracts disagree with each other. In the same week, the winner contract rose 3.20pp and the 2º turno 1º turno placement contract fell 2.00pp. The consistent reading is that the market raised the probability of him winning conditional on reaching the runoff and reduced the probability of him being the one who gets there. Placement contracts have volume two orders of magnitude smaller, so it may be a depth artifact; if it persists for another window, it requires structural explanation.
- Renan corrected price without losing momentum. Fell 2.30pp in the winner and 2.50pp in third place, and still maintained the highest cumulative volume among living contenders (USD 8.18M). On July 19, lost another 4.00pp in third place, on the same trading session in which Ciro Gomes cited him nominally as a voting possibility. Narrative and price pointed in opposite directions, and there is no evidence that one explains the other.
- Conventions are the first structural event on the calendar. Deadline from July 20 to August 5. PL closes its ticket with Jair Bolsonaro under house arrest and Flávio suspended from visits until mid-October; Novo arrives without a running mate after Michelle declines the position on July 19. Watch the contracts for 2nd and 3rd place, not the winner's: it is in those that the definition of electoral alliance reprices.
8.Methodology
AFOS Tradeoff aggregates three signals without mediating them into a composite: Polymarket (denominated in USD, latency ~30min), surveys registered at TSE (declared intent, variable periodicity) and 400+ news sources (event flow). When the three diverge, the divergence is the signal, not consensus. USD volume is reported alongside implicit probability to separate conviction from artificial spike. Liquidity (order book depth) is not cited inline: low liquidity on Polymarket does not mean the price is wrong; the market is arbitraged continuously.
Weekly deltas in this edition come from snapshots persisted in our own database, not from retroactive reconstruction. Transparency note on data collection: three historical series returned empty on first query, and none of them was data absence, but name-matching failure in the query. They were recovered with positive control before closure. Absence of search result is not evidence of absence of fact, and treating it as such is the error that produced the correction published in Edition №8.
9.Additional reading · macro coverage
Key articles and sections from the week in major outlets. AFOS Tradeoff is the primary source (Polymarket + TSE + polls); the references below are supplementary reading for macro context alignment. Note: some content is behind paywall.
- Folha de S.Paulo · Ciro Gomes cites Caiado and Renan Santos as voting possibilities and ignores Flávio Bolsonaro (July 19)(paywall), www1.folha.uol.com.br/poder/2026/07/ciro-gomes-cita-caiado-e-renan-santos-como-possibilidades-de-voto-e-ignora-flavio-bolsonaro.shtml
- O Globo · Pivot of dispute between Flávio and Michelle, Ciro Gomes evaluates supporting candidacies of Caiado or Renan Santos (July 19)(paywall), oglobo.globo.com/politica/eleicoes-2026/noticia/2026/07/19/pivo-de-briga-entre-flavio-e-michelle-ciro-gomes-avalia-apoiar-candidaturas-de-caiado-ou-renan-santos-na-presidencia.ghtml
- Gazeta do Povo · Ciro signals support for Caiado or Renan and leaves Flávio Bolsonaro off the radar (July 19), www.gazetadopovo.com.br/eleicoes/2026/ciro-sinaliza-apoio-caiado-renan-deixa-flavio-bolsonaro-fora-do-radar/
- Estadão · Lula government circumvents fiscal framework and bets on R$ 145 billion credit package in election year (July 19)(paywall), www.estadao.com.br/economia/governo-lula-dribla-arcabouco-e-aposta-em-pacote-de-credito-de-r-145-bi-em-ano-eleitoral/
- Folha de S.Paulo · Trump may interfere in election to favor Flávio Bolsonaro and create chaos, says Brazil expert (July 19)(paywall), www1.folha.uol.com.br/poder/2026/07/trump-pode-interferir-na-eleicao-para-favorecer-flavio-bolsonaro-e-gerar-caos-diz-brasilianista.shtml
Macro context references. The primary signal of this edition (a 3.20pp rise by Flávio between July 13 and 17 against the direction measured by Quaest and PoderData in the same window, and Lula's breakthrough below the 60% floor on July 19) is direct observation of Polymarket pricing crossed with TSE records, not derived from the articles above.
Get it by email
Tomorrow's read, before you go looking for it.
AFOS Tradeoff is the weekly read on Brazilian political risk, for people who decide on data rather than headlines.
No spam. Unsubscribe anytime.
After signing up you choose whether to receive it in English, Portuguese or Spanish.