AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
BrazilIssue №11 left two scenarios written with numeric criteria, and the August 3-7 window split both down the middle. The base scenario projected that the August 5 Quaest would measure the first-round gap between 8pp and 10pp: it measured 9pp, and got it right. The same scenario projected the market gap between +39pp and +42pp: it never touched that band in any of the five sessions, opening Monday at +38.90pp and closing Friday at +37.50pp. The contrarian scenario's signature was written as compression below +38pp with Flávio above 25.50% in the winner contract, and it happened in the last two sessions, with the gap at +37.90pp and +37.50pp and Flávio at 26.60% and 27.00%. Except that scenario's premise was the Quaest coming in near PoderData's 6pp, and it came in at 9pp. The outcome arrived without the cause the previous issue wrote for it, and a scorecard checking only the number would mark it a hit. What replaced that cause sits in comparing each institute against itself, the only control that neutralizes house effect. Three houses repeated a round and none of them widened the gap: the BTG/Nexus of August 3 fell from 9pp to 4pp against its own July 27 round, Genial/Quaest fell from 12pp to 9pp against its own July 15 round, and Meio/Ideia went from 8.4pp to 8pp. On the macro side the Copom cut the Selic rate by 0.25pp to 14.00% on August 5, the fourth consecutive cut, and the inflation contract moved UP over the two following sessions: the 5.00% to 5.49% band rose 8.70pp on the week, to 45.40%. The presidential market has accumulated USD 121.43M. The next window brings six national polls between August 10 and 13, five of them with field dates inside the week in which price moved on its own.
1.Executive Summary
№11's base scenario got the polls right and the market wrong; the contrarian got the market right and the reason it would move wrong. The reading that survives was in neither text: of the three national polls in the window, two compress against their own previous round, one holds steady and none widens. Snapshot of August 9, 19:42 BRT, over the 17 books with volume in the presidential contract.
2.Why AFOS does not smooth
Three national polls in the window, and averaging them is the worst possible summary. The first-round gaps were 4pp, 8pp and 9pp, whose average is 7.0pp, a number none of the three measured. Worse: comparing that 7.0pp with №11's 8.35pp suggests a 1.35pp narrowing, when the two averages are composed of different institutes. The valid comparison exists in this window because all three houses had a previous round.
Why it matters: №11 treated PoderData's 6pp as the floor of the set and asked whether it was house effect or a leading indicator. Nexus came in at 4pp, below that floor, and Quaest held the ceiling at 9pp. The two extremes moved apart and the question remains unanswered, because that week's dissenter did not repeat a round in this one. Mandatory and repeated caveat: market price is probability of victory, not vote margin, and the two quantities do not subtract from each other. Quaest still records approval at 48% against 47%, identical to its own July 15 round: an indicator standing still in the same house did not announce the first-round compression.
What weakens this issue's thesis is the chronology. The gap kept falling on August 6 and 7, by 0.60pp and 0.40pp, with no national poll coming out at all, because the last ones were published on August 5. And on August 5 itself, a day with two national polls, it moved 0.10pp, the smallest variation of the window. The largest single-session drop was on August 3, 2.00pp against the Sunday close, the day of the 4pp Nexus and of Flávio's biggest jump, and that is a coincidence of two verified dates, not a demonstrated cause. After the window the gap went on to +36.50pp, which reinforces the direction and not the explanation.
The same framework applied to Brazil, measuring the distance between prediction market and polls and checking the signal against the actual election results, has already been tested against eleven elections across four continents, among them India, France, the United Kingdom, Mexico and the United States (2024), South Korea, Chile, Germany and Canada (2025) and Peru and Colombia (2026). India is the only case in which market and polls were wrong together in the same direction. The method reports convergence and divergence with no retroactive adjustment. The dataset is open under Apache 2.0 and has a permanent DOI at the Harvard Dataverse.
3.Weighted scenarios for the week
Scenarios for the August 10-14 window, the densest of the cycle in ballot-box measurement: six national polls registered between August 10 and 13, against three in the covered window, and five of them with field dates inside August 3 to 7. They measure precisely the week in which price moved with no new poll. Candidacy registration closes on August 15.
The six national polls confirm the direction the three houses recorded against themselves, with most measuring the first-round gap between 5pp and 9pp and none widening against its own previous round. Lula's winner contract stays between 62% and 65% and the gap settles between +34pp and +38pp, without returning to +40pp. The August 11 MDA either backs it or removes it. Net-neutral for BRL.
The national polls come in wide, with most above 9pp in the first round and at least two widening against their own previous round. In that case the price compression of the last two weeks has no ballot-box backing, and the Nexus 4pp goes back to being an isolated reading from one house, as PoderData's 6pp was in №11. The signature would be the gap back above +38pp with Flávio below 26% in the winner contract. The cleanest test is the August 10 Palver, n=5,000 and field dates from August 3 to 9, the only one covering the entire window, with the caveat that it is online and non-probabilistic. This scenario inverts this cover.
Candidacy registration closes on August 15, and a ticket change on that stretch moves the placement contracts, which add up to USD 4.50M and USD 0.52M against USD 121.43M for the presidential market. The fragility of those books is measured in this very window: Caiado's third place rose 10.00pp on the week and gave back 8.00pp in the two following sessions, in a book of USD 45 thousand. On the macro side, the inflation contract has bands with books of USD 5 thousand to USD 14 thousand; the full IPCA and the Copom minutes test the tone of the statement.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| Lula × Flávio gap | +37.50pp | ↓1.40pp week | 121.43M | Same magnitude as №11 with the sign flipped. Below the projected +39pp to +42pp band across the five sessions, it crossed +38pp on August 6 and went on to +36.50pp on August 9 |
| Lula · winner | 64.50% | ↓1.00pp week | 8.18M | Fell 1.00pp on August 4 and stood still for the four following sessions, without threatening the top of the set, 66.50% on August 1. It broke the plateau downward on August 9, outside the window |
| Flávio · winner | 27.00% | ↑0.40pp week | 8.10M | The weekly gain hides the shape: it rose 2.00pp on August 3, day of the 4pp Nexus, gave back 0.70pp and rose again on August 6 and 7, with no new poll on those two days |
| Flávio · third place first round | 1.20% | ↓3.70pp week | 24k | №10's thesis required going back above 8%, and the book went the other way, with 3.30pp lost on August 6 alone. It returned to 3.80% on August 9. Book of USD 24 thousand |
| Ronaldo Caiado · third place first round | 33.50% | ↑10.00pp week | 45k | The largest delta of the issue and the least reliable. Book of USD 45 thousand, it gave back 8.00pp on August 8 and 9, and in the polls it measures 4% and 5.7%. Unconfirmed movement |
| Renan Santos · winner | 7.80% | ↑0.80pp week | 9.28M | First week of gains after the fall recorded in №11, and the biggest volume gainer of the window, USD 335 thousand, ahead of Caiado's USD 278 thousand. In the polls, 4% and 4.7% |
| 2026 inflation · 5.00–5.49% band | 45.40% | ↑8.70pp week | 11k | It was modal across the five sessions and opened a lead, rising 6.60pp the day after the Copom. The 5.50% to 5.99% band rose 7.70pp: the whole distribution moved up |
| STF impeach < 2027 | 2.80% | ↓0.30pp week | 83k | Three orders of magnitude below the presidential market and with no reaction to the week. It went to 3.80% on August 8, outside the window |
5.Liquidity and market structure
The three largest accumulated volumes remain in contracts priced at 0.05%: Tarcísio (USD 13.88M), Eduardo Bolsonaro (USD 10.42M) and Carlos Massa (USD 10.42M). These are old positions never unwound, and accumulated volume measures historical interest, not current conviction. The latter two swapped positions separated by USD 3,926, with 10,421,284 against 10,417,358: the ranking order in that band is decided by less than four thousand dollars and means nothing. And the anomaly №11 recorded, with Eduardo Leite holding almost the same money as Lula at 0.05% against 65.50%, has come undone: Lula holds USD 8,183,935 against USD 7,944,637 for Leite, a distance of USD 239 thousand. Between the two snapshots, from August 2 to 9, Lula's book gained USD 275 thousand and Leite's USD 31 thousand.
The 5 largest volumes account for ~USD 53.69M (~44%) of the presidential market (total USD 121.43M across 32 markets, 17 with volume), which rose from USD 118.16M on August 2. The gap compression took place in the two largest books among live contenders, with USD 8.18M and USD 8.10M almost tied, so it is not shallow-book movement. In the placement contracts the reading is the opposite: the entire third place accumulates USD 0.52M, and that is where the largest delta of this issue sits, Caiado's, in a book of USD 45 thousand. Low-volume spikes (USD <500k) remain treated as noise until recurring flow confirms them.
6.Calendar of price-relevant prints
| Date | Sample | Why it matters | |
|---|---|---|---|
| Mon Aug 10 | Palver (national) | n=5,000 · BR-06596/2026 · conf. 0.70 | Largest sample of the set and the only one with field dates covering the entire window, August 3 to 9. Online and non-probabilistic, with weights anchored in IBGE and TSE data |
| Mon Aug 10 | Gerp and Nexus (national) | n=2,400 · BR-08045/2026 and n=2,000 · BR-08428/2026 · conf. 0.70 | Nexus is the one that matters most: the round following the 4pp one, same house and method, and it tells whether that reading was an outlier or the start of a series |
| Tue Aug 11 | MDA and 100 Cidades (national) | n=2,002 · BR-06935/2026 · conf. 0.90 and n=2,000 · BR-08109/2026 · conf. 0.70 | MDA has the highest reliability of the window and is the only one of the six declaring in-person interviews, with field dates August 5 to 9, after the Copom. 100 Cidades has field dates August 3 to 6 |
| Thu Aug 13 | PoderData (national) | n=2,400 · BR-06868/2026 · conf. 0.70 | The 6pp dissenter №11 pointed to measures again, field dates August 9 to 12. It answers directly the question that issue left open |
Source: TSE registry via AFOS API. With declared publication from August 10 to 13 there are 30 registrations, 23 with a sample ≥ 1,000; the table lists the six of national scope, and the other 24 stay out by scope, not for lack of registration. ⚠️ Correction to №11's calendar, which listed four national polls for the covered window: only three published. The fourth, Nexus's BR-05573/2026, scheduled for August 4 with n=1,200, comes classified as national in the registry, but the methodology declared to the TSE itself describes the surveyed population as voters of Tocantins. It did not enter the panel, and the defect is one of labeling at the source, not of collection. The six in this window were each put through the same test and none names a state.
7.Watch list, week triggers
- The August 10 Nexus is the direct test of this issue's finding. Same house, same method, the round following the 4pp one. Repeating near 4pp makes the compression a series; going back near 9pp makes the 4pp an outlier, and this cover loses its axis.
- Watch whether the gap loses +36pp. It stands at +36.50pp, with seven falls in eight sessions since the top of +41.80pp and a single day standing still. Below +36pp it stops being a giveback of the peak and becomes a new level.
- Separate who gives way when the gap narrows. In the window it fell with Flávio rising; on August 9 it fell with Lula giving way and Flávio standing still. Transfer between two names and price loss in the favorite give the same number and are not the same fact.
- The inflation contract tests the Copom's tone, not the cut. The cut was in the consensus and the book only moved after the statement. The full IPCA and the minutes say whether that rereading holds.
- Caiado in third place has already given back 80% of the week's move. It rose 10.00pp in the window and fell 8.00pp on August 8 and 9, in a book of USD 45 thousand, against 4% and 5.7% in the polls. A large delta in a small book does not become a thesis.
8.Methodology
AFOS Tradeoff aggregates three signals without averaging them into a composite: Polymarket (in USD, ~30min latency), polls registered with the TSE and 400+ press sources. When the three diverge, the divergence is the signal. USD volume goes alongside the implied probability to separate conviction from artificial spikes. Liquidity is not cited inline: low liquidity on Polymarket does not mean the price is wrong, the market is continuously arbitraged.
Weekly deltas come from snapshots persisted in our own database, not from retroactive reconstruction, and use the last point of each day. That produces differences of hundredths relative to the AFOS Daily, which reports the intraday capture: the August 9 gap appears here as +36.50pp and in the Daily as +36.55pp, because the day's close has Flávio at 27.00% and the 17:34 UTC capture has him at 26.95%. Both are correct within their own convention, and we record it because the alternative would be the reader running into the difference alone.
Comparison inside the institute: when this issue says a house 'fell from 9pp to 4pp', the confrontation is between two rounds of the same institute, with both dates declared. It is the only confrontation that neutralizes house effect, and it differs from comparing institutes against each other, which mixes method, sample and field dates. The two readings appear separately on purpose.
On superlatives: the public history API caps the query at 90 days, so any 'largest of the cycle' checked through it measures 90 days. The maxima and minima cited here are from the set of twenty sessions, July 21 to August 9, and no superlative in this issue goes beyond that window.
9.Additional reading · macro coverage
AFOS Tradeoff is a primary source (Polymarket + TSE + polls); what follows is complementary reading for macro context. Note: some of it operates behind a paywall.
- Nexus (institute) · BTG/Nexus presidential poll, August 3 round, released by the institute itself, www.nexus.fsb.com.br/estudos-divulgados/pesquisa-btg-nexus-de-intencao-de-votos-para-presidente-do-brasil-3-de-agosto-de-2026/
- CNN Brasil · Central Bank cuts the Selic rate to 14% a year and leaves the door open for the next meeting (August 5), www.cnnbrasil.com.br/economia/macroeconomia/bc-copom-agosto-2026/
- JOTA · Genial/Quaest: Lula leads with 39%, Flávio reaches 30% and narrows the distance (August 5), www.jota.info/eleicoes/eleicoes-2026/genial-quaest-lula-lidera-com-39-flavio-bolsonaro-chega-a-30-e-encurta-distancia
- Poder360 · Flávio announces Alfredo Gaspar as running mate, on the last day of the convention deadline (August 5), www.poder360.com.br/poder-eleicoes-2026/ao-lado-de-6-mulheres-flavio-bolsonaro-anuncia-alfredo-gaspar-como-vice/
- Poder360 · Lula and Flávio technically tied in the runoff in the Nexus round, 46% x 45% (August 3), www.poder360.com.br/poder-eleicoes-2026/lula-e-flavio-estao-tecnicamente-empatados-no-2o-turno-diz-nexus/
- Exame · Genial/Quaest: Lula has 44% and Flávio 39% in the runoff (August 5), exame.com/brasil/pesquisa-genial-quaest-lula-tem-44-e-flavio-bolsonaro-39-no-2o-turno/
This issue's primary signal (the gap falling outside the band projected by №11, the compression recorded by three houses against themselves, and the shift in the inflation distribution after the Copom statement) is direct observation of Polymarket pricing cross-referenced with the TSE registry, not derived from the articles above.
Get it by email
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.