AFOS Analytics
Real-time political pricing
Prediction markets × polls × news
no smoothed averages
BrazilThe Aug 10-14 window produced seven national polls and the widest dispersion this brief has ever recorded. The first-round gaps between the top two were 0pp, 4pp, 4.7pp, 5pp, 6pp, 7pp and 13.7pp. The average comes to 5.77pp, and none of the seven measured that. On Aug 11 three national polls came out on the same day, and the distance between the widest and the narrowest was 13.7pp: CNT/MDA printed 42.4% x 28.7% and Gerp printed 38% x 38%, a tie. In the runoff the dispersion is just as wide and more informative: Gerp records a statistical tie with Flávio numerically ahead, 43% x 45% inside a 2pp margin, and CNT/MDA puts Lula +9pp, 48% x 39%. Eleven points of amplitude on the same question, in the same week.
The market did not move. Lula's winner contract closed all five sessions of the window at 63.50%, without a single change in the close, and the gap oscillated inside a 1.30pp band, between 35.50pp and 36.80pp, ending at 35.60pp against 37.50pp at the Aug 7 close. Seven polling measurements with 13.7pp of amplitude and a motionless price is the crossing this brief exists to show, and it has two possible readings that AFOS does not choose between: either the book had already priced the whole range, or it does not read polling at that granularity at all.
⚠️ This issue also scores the scenarios from №12, which projected exactly this window. The base scenario got the three market conditions right and the two polling conditions wrong. It is the inverted mirror of №11, whose base got the polling right and the market wrong. Two consecutive issues in which the half of the scenario that speaks about price and the half that speaks about polls do not land together.
1.Executive Summary
The theme of the week is amplitude, and it shows up on two axes usually treated as one. Across houses, seven national polls measured the same race with 13.7pp between the widest and the narrowest gap. Across time, Caiado's third-place contract recorded a weekly change of 0.00pp while travelling 7.00pp from one end of the window to the other. Both cases carry the same signature: a summary number that erases the information that matters. The presidential market, by contrast, crossed all five sessions without moving the leader's closing price, which makes this the window with the largest distance between what the polls measured and what the price recorded since this brief began.
2.Why AFOS does not smooth
Seven nationals in the window, and their average destroys the finding. The first-round gaps were 0pp, 4pp, 4.7pp, 5pp, 6pp, 7pp and 13.7pp, whose arithmetic mean comes to 5.77pp. None of the seven measured 5.77pp, and none measured anything between 4.7pp and 5pp. The aggregate invents a centre the distribution does not have.
<strong>Why it matters:</strong> anyone trading against consensus needs to know whether that consensus is narrow or wide. An average of 5.77pp across seven houses suggests convergence; the range from 0pp to 13.7pp says there is no convergence at all on the size of the lead, and that disagreement between houses is currently larger than the weekly variation in price.
📐 The shape of the dispersion, which is more precise than its amplitude: the seven are not spread evenly. Five sit between 4pp and 7pp, a narrow band, and two run off in opposite directions, CNT/MDA at 13.7pp and Gerp at 0pp. The field has a centre and two extremes, not a continuum from 0pp to 13.7pp. ⚠️ The two extremes are the ones that carry the most methodological weight: CNT/MDA is the only one in the set that declares in-person interviewing and carries the highest registered reliability, and Gerp is the only one in which Flávio appears numerically ahead in the runoff. Neither is discarded for being an extreme, and it is precisely because they are not discarded that the published range runs from 0pp to 13.7pp instead of 4pp to 7pp.
The same framework applied to Brazil, measuring the distance between prediction market and polls and checking the signal against the actual ballot-box result, 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 without retroactive adjustment. The dataset is open under Apache 2.0 and carries a permanent DOI at the Harvard Dataverse.
3.Weighted scenarios for the week
Before the new scenarios, the scorecard for those in №12, which projected this window. The base got the three market conditions right and the two polling conditions wrong. It was right that Lula would sit between 62% and 65%, and he closed all five sessions at 63.50%; right that the gap would sit between +34pp and +38pp, and the band ran from 35.50pp to 36.80pp; right that it would not return to +40pp. It was wrong in projecting that the majority of the nationals would land between 5pp and 9pp, and only three of seven did; and wrong in projecting that none would widen against its own previous round, because BTG/Nexus went from 4pp on Aug 3 to 5pp on Aug 10. The contrarian scenario was wrong on both conditions it stated. The tail did not materialise as an event, but the fragility it identified was measured: Caiado's third place travelled 7.00pp inside the window in a book of USD 52 thousand. The scenarios below cover Aug 17-21, a week with only two registered nationals, against seven in the covered window.
With only two nationals scheduled, the week is one of polling scarcity after one of excess. Nexus on Aug 17 is the third consecutive round from the same house and says whether the 4pp of Aug 3 and the 5pp of Aug 10 are a series or noise. The price gap sits between +35pp and +38pp and Lula's contract between 64% and 67%, without the dispersion across houses resolving, because two measurements do not resolve an amplitude of 13.7pp. Net-neutral for BRL.
The market leaves the band it spent the entire window in. The signature would be the gap above +38pp with Lula's contract above 67%, which would require breaking the series top of 66.50%, touched on Aug 1 and again on Aug 16. ⚠️ This scenario already has partial support outside the covered window: on Aug 15 and 16 the contract rose to 65.50% and then 66.50%, and the move was recorded at 5:01 am BRT on Aug 16, before the campaign-opening rallies. If Nexus or Real Time Big Data come in wide, above 9pp, the price gains polling support it did not have in the covered window.
The chasing pack reorganises inside the placement contract. The move has already begun outside the window: Renan Santos's third place fell from 65.00% on Aug 9 to 54.00% and Caiado's rose to 37.50%, with the distance between the two going from more than 30pp to less than 17pp. In the winner contract Renan's fall is more violent: 7.20% at the Aug 14 close to 3.65% at 6:41 pm BRT on Aug 16. Books of USD 543 thousand in third place against USD 124.32M in the presidential one, so a large delta there costs little capital and should not be read as repricing of the same quality.
4.Indicator Grid
| Contract | Current | Δ week | Vol USD acc. | Implied reading |
|---|---|---|---|---|
| Gap Lula × Flávio | +35.60pp | ↓1.90pp week | 124.32M | Aug 14 close. The band inside the window was only 1.30pp, from 35.50pp to 36.80pp, against 13.7pp of amplitude across the seven nationals of the same period |
| Lula · winner | 63.50% | ↓1.00pp week | 8.40M | Closed all five sessions of the window at the same value, without a single change in the close. Outside the window it rose to 65.50% on Aug 15 and 66.50% on Aug 16, which matches the series top |
| Flávio · winner | 27.90% | ↑0.90pp week | 8.34M | All of the gap narrowing came from this side, because the leader did not move. On Aug 13 he pulled back to 26.70% and gave it back the following session |
| Renan Santos · winner | 7.20% | ↓0.60pp week | 9.77M | The weekly Δ is modest and what came after is not: 6.20% on Aug 15 and 3.65% at 6:41 pm BRT on Aug 16, that is, ↓3.55pp across two sessions outside the window |
| Ronaldo Caiado · first-round third place | 33.50% | 0.00pp week | 52k | A weekly Δ of zero covering 7.00pp of path, from 26.50% on Aug 10 to 33.50% on Aug 14. It is the textbook case of why end-to-end change describes a thin book badly |
| Renan Santos · first-round third place | 58.00% | ↓0.50pp week | 181k | Here too the weekly Δ misleads: the contract touched 65.00% on Aug 9 and had been falling ever since. On Aug 16 it stood at 54.00% |
| 2026 inflation · 5.00–5.49% band | 46.05% | ↑0.65pp since Aug 7 | 14k | Aug 16 reading. Still modal after the post-Copom jump that №12 recorded, and the neighbouring 5.50–5.99% band holds 20.50%. Book of USD 95 thousand in total |
| Supreme Court impeachment < 2027 | 3.40% | ↑0.60pp since Aug 7 | 84k | Thinnest contract among those tracked, three orders of magnitude below the presidential one. Movement here requires that caveat before any reading |
5.Liquidity and market structure
The three largest accumulated volumes remain in contracts priced at 0.05%: Tarcísio (USD 13.92M), Eduardo Bolsonaro (USD 10.44M) and Carlos Massa (USD 10.43M). These are old positions never unwound, and accumulated volume measures historical interest, not current conviction. All three have formally opened campaigns for other offices or are not presidential candidates, which pins the price at the floor without the volume moving. ⚠️ On Aug 16, Tarcísio opened a re-election campaign in São Paulo and Michelle registered a Senate candidacy for the Federal District, which turns into a dated public act what the book was already pricing.
The contrast with the placement book is the operational point of the week: first-round third place totals USD 543 thousand, and inside it Caiado's contract holds USD 52 thousand. That is where the 7.00pp of path in the window happened. A large delta in a small book is not repricing of the same quality as a small delta in a nine-figure book, and this brief does not treat them as equivalent.
6.Calendar of price-relevant prints
| Date | Sample | Why it matters | |
|---|---|---|---|
| Mon Aug 17 | Nexus (national) | n=2,000 · BR-03317/2026 · conf. 0.70 | Third consecutive round from the same house, after 4pp on Aug 3 and 5pp on Aug 10. It is the only test in the window comparing one house with itself three times |
| Wed Aug 19 | Real Time Big Data (national) | n=1,600 · BR-05423/2026 · conf. 0.80 | Second-highest reliability in the set and the only other national of the week. Smaller sample of the two |
| Sat Aug 15 | Close of candidate registration | 20,438 candidacies as of 7:30 pm · 13 for President | A settled fact, listed here because it is the trigger the tail of №12 pointed to. Poder360 notes that the count was still incomplete at the close |
| Sun Aug 16 | Start of the electoral campaign | street and internet propaganda released | First day of permitted campaigning. The day's price move was recorded at 5:01 am BRT, before any rally, so it is not attributed to the events |
Source: TSE registry via the AFOS API. In the Aug 17-23 window there are 17 records with declared publication; the table lists the two of national scope with sample ≥ 1,000, and the other 15 fall outside by scope, none by sample size. ⚠️ Two nationals against seven in the covered window, and that drop is itself a datum: next week has little polling with which to arbitrate the dispersion this issue describes.
7.Watch list, week triggers
- Nexus on Aug 17 is the only series test of the week. Third round from the same house in fifteen days, after 4pp and 5pp. Repeating in the 4pp to 5pp range turns the narrow reading into a series of three; jumping to near 9pp returns the house to the middle of the field and undoes the low end of the range measured in the window.
- Watch whether Lula's contract holds 66.50%. That value matches the top of a 174-point series running since May 19, touched before only on Aug 1, and no point in the series sits above it. Holding above it becomes unprecedented territory in the series; giving it back confirms the ceiling.
- Renan Santos's collapse is the fastest move in the presidential book. From 7.20% at the Aug 14 close to 3.65% at 6:41 pm BRT on Aug 16, accelerating on Sunday evening. His contract holds USD 9.77M, that is, it is not a thin book, and speed here weighs more than magnitude.
- Separate placement from victory inside the chasing pack. Renan falls on both questions at once, victory and third place, while Caiado falls on victory and rises on placement. The distance between the two in the placement contract narrowed from more than 30pp to less than 17pp across four sessions.
- Dispersion across houses, not the average, is the number to track. If Nexus and Real Time Big Data land inside the 4pp to 7pp band where five of the seven sit, the reading becomes consensus with two declared extremes. If either comes in above 9pp or at a tie, the 13.7pp amplitude stops being the effect of two houses.
8.Methodology
AFOS Tradeoff aggregates three signals without averaging them into a composite: Polymarket (USD-denominated, ~30min latency), polls registered with the TSE and 400+ press sources. When the three diverge, the divergence is the signal, and that is what this issue publishes instead of an average gap.
A price only enters after two independent readings taken eight minutes apart agree within 0.20pp; a reading that does not confirm is not published. Closing values are the last point recorded each day in BRT, and series are measured on the live window at every issue, never inherited from the previous one. Code and method are open under Apache 2.0 on GitHub, and the dataset of validated elections carries a permanent DOI at the Harvard Dataverse.
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.
- Poder360 · BTG/Nexus: Lula 47% x 44% for Flávio in the runoff (Aug 10), www.poder360.com.br/poder-eleicoes-2026/lula-tem-47-contra-44-de-flavio-no-2o-turno-diz-btg-nexus/
- Exame · Palver: first round 44% x 40%, and a tie in the runoff, with Renan Santos at 10% (Aug 10), exame.com/brasil/lula-tem-44-contra-40-de-flavio-em-primeiro-turno-e-renan-santos-aparece-com-10-diz-palver/
- JOTA · Palver: Lula leads the first round and ties the runoff (Aug 10). Online and non-probabilistic, which explains the high end of the set, www.jota.info/eleicoes/eleicoes-2026/lula-lidera-primeiro-turno-mas-empata-com-flavio-bolsonaro-no-segundo-aponta-palver
- Poder360 · CNT/MDA: Lula 48% x 39% for Flávio in the runoff (Aug 11). High end of the runoff amplitude, www.poder360.com.br/poder-eleicoes/lula-tem-48-contra-39-de-flavio-no-2o-turno-diz-pesquisa-cnt/
- Poder360 · Gerp: statistical tie in the runoff, 43% x 45%, with Flávio numerically ahead (Aug 11), www.poder360.com.br/poder-eleicoes-2026/flavio-e-lula-tem-empate-tecnico-no-2o-turno-diz-gerp/
- CNN Brasil · PoderData/Aya: Lula ties with Flávio, Zema and Caiado in the runoff (Aug 13), www.cnnbrasil.com.br/eleicoes/poderdata-aya-lula-empata-com-flavio-zema-e-caiado-no-2o-turno/
- G1 · Quaest: crosstab on how firm the vote is, with 77% of Lula's voters calling the decision final (Aug 15), g1.globo.com/politica/eleicoes/2026/pesquisa-eleitoral/noticia/2026/08/15/quaest-decisao-voto-14-agosto.ghtml
The primary signal of this issue (the 13.7pp amplitude across seven nationals, the statistical tie in Gerp's runoff with Flávio numerically ahead, the immobility of the winner contract across all five sessions and the weekly Δ of zero covering 7.00pp in the placement book) comes from our own measurement of Polymarket and the TSE registry, with a declared timestamp on every number. The readings above are coverage context, not the source of the finding.
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.