Name: AFOS Analytics Election Data
Creator: AFOS Analytics
License: https://creativecommons.org/licenses/by/4.0/

Three movements define the day. Vox Brasil published a national poll confirming the scenario outlined by AtlasIntel on May 19: Lula 46.8% × Flávio 38.1% in the runoff (gap +8.7pp); rejection of Flávio at 49.2%, Lula at 52.8%. Polymarket priced the collapse continuation: Flávio fell to 23.35% (↓4.80pp in 32h, absolute minimum of the cycle), and the Lula × Flávio gap reached +22.15pp — RECORD of the series. The PSDB initiated internal debate on launching Aécio Neves, while Caiado and Zema gave public speeches without citing Flávio but with clear reference.

1. Prediction market

The Polymarket markets of the Brazil 2026 cycle total USD 87.90M accumulated since opening (snapshot May 21 19:00 BRT), distributed among presidential USD 83.66M, 2nd place USD 3.55M, 3rd place USD 301k, STF impeachment USD 76k, Senate USD 254k and 2026 inflation bands USD 60k. The presidential market concentrates more than 95% of total volume — where the signal is most robust.

In the presidential Polymarket market, Lula closed the day at 45.50% (USD 5.70M), stable for more than 60 hours since AtlasIntel on May 19, while Flávio Bolsonaro closed at 24.75% (USD 5.88M) — partial recovery vs the absolute minimum of 23.35% registered in the morning (↑1.40pp in 8h). The Lula × Flávio gap closed at +20.75pp (↓1.45pp vs the morning, because Flávio recovered).

The third way reorganized asymmetrically and reversed the morning spike. Renan Santos closed at 13.05% (USD 5.57M) — corrected downward the peak of 15.60% registered in the morning, still consolidating the SECOND position in the presidential Poly above Zema (5.35%, USD 2.72M) and Haddad (5.05%, USD 4.84M). Michelle Bolsonaro at 3.55% (USD 6.47M) — the second largest position in volume among third-way candidates, signaling that the narrative of substitution on the PL slate continues to attract hedge capital. In the 3rd place market, Renan remained at the top with 34.00%, Zema at 24.50%, Caiado at 15.50%, and Flávio at 8.15% — the market continues pricing Flávio with material risk of finishing in third.

In the 2nd place market, Flávio maintained 55.00% at the top of the runoff. Renan rose to 14.50% (USD 1.0M) — only candidate with volume above USD 1M in 2nd place, conferring real weight to the signal that the market sees Renan as an ascending hypothesis for 2nd place. Lula at 12.50%, Haddad at 4.10%, Michelle at 3.35%, Zema at 3.25%.

The most relevant movement of the day came after close: the contract for impeachment of STF justice before 2027 fell to 4.35% (USD 75k accumulated) — decline of ↓3.20pp in 90 minutes against 7.55% at 17:30 close. Movement in sub-market of low aggregated volume — monitor over the next 24h before consolidating reading on the signal.

In the Senate — sub-market with very reduced volumes —, PL closed at 75.00% (USD 243k) confirming first decline of the cycle. PSB at 8.05%, PSD at 6.30%, União Brasil at 4.90%, MDB at 4.80%. Republicanos partially recovered to 4.35% after this week's collapse. In the 2026 inflation bands, the central band 4.50-4.99% rose to 30.20% (↑2.85pp in 8h signaling priced fiscal moderation); band 5.00-5.49% at 23.70%; high tail (≥6.50%) fell to 7.75%.

2. What the institutes recorded

Vox Brasil released a national poll on May 20 with Lula 46.8% and Flávio Bolsonaro 38.1% in the runoff (gap +8.7pp). Rejection of Lula was 52.8% and of Flávio 49.2%. Parallel coverage of the same survey in Exame, Folha de Alphaville and Revista Fórum confirmed the same numbers.

Vox Brasil is the second national poll to register Flávio's deterioration post-Vorcaro audio, after AtlasIntel on May 19 (1st round Lula 47% × Flávio 34.3% gap +12.7pp; runoff 48.9% × 41.8% gap +7.1pp). Taken together, the two institutes show consistent gap around +7-9pp Lula in the runoff — significant methodological convergence, given that AtlasIntel and Vox use distinct sampling designs. The Datafolha of May 16 remains as the most conservative reference point (1st round 38×35 gap +3pp; runoff 45×45 empate técnico), and the 5.7pp discrepancy between Vox 1st round and Datafolha 1st round remains the largest divergence among institutes in the cycle.

State polls also published: Real Time Big Data in Ceará shows Lula with wide advantage in first and second rounds. The next expected national release is Datafolha on May 22, n=2.004, including scenario with Michelle Bolsonaro as substitute on the PL slate.

📅 Survey calendar — next 7 days

Surveys registered with TSE with publication scheduled between May 22 and May 28. Inclusion in the table does not mean publication is confirmed — institutes may delay or cancel release. Filter applied: sample ≥ 1,000. Each protocol linked to TSE public consultation.

Date	Institute	Sample	Scope	TSE Protocol	Conf.
May 22	Datafolha	2,004	national	BR-074892026	0.9
May 22	Gerp	2,000	state	BR-079712026	0.7
May 22	PNH	1,500	state	BR-038502026	0.6
May 23	Data Census	2,000	state	BR-023422026	0.7
May 23	Econometrica	1,607	state	BR-082622026	0.6
May 23	C. Gomes Marketing	1,600	state	BR-012552026	0.6
May 23	ECM	1,200	state	BR-020842026	0.6
May 24	Inst. Vox Brasil	2,100	state	BR-079612026	0.7
May 24	Econometrica	1,607	state	BR-052272026	0.6
May 24	Cavalcante Falabella	1,070	state	BR-004322026	0.6
May 25	Data Census	2,000	state	BR-023422026	0.7
May 25	Nexus	2,000	state	BR-—/2026	0.7

Source: TSE public registry via disclosure portal and AFOS Analytics API. Status "registered ≠ published" — confirmation of effective disclosure requires verification from two primary sources before citation of numbers.

3. What the press covered

In the third way field, O Globo reported that PSDB initiated internal debate about launching Aécio Neves for President — first explicit signal of movement in the non-Bolsonaro right after Flávio's wear and tear. In parallel, O Globo registered that Caiado raised his tone stating that a politician "contaminated" by Vorcaro is not in a position to reach the Presidency, without citing Flávio by name. Folha de S.Paulo registered a symmetric movement: Zema said to mayors that credibility is needed to lead the country, also without citing Flávio directly.

In the Bolsonaro field, Folha reported that Michelle Bolsonaro gave a speech ignoring Flávio and criticized "alliance with evil" (reference to Ciro Gomes). Ricardo Salles stated to BBC that "a ticket with Michelle in Flávio's place would gain much support". Estadão reported that Flávio Bolsonaro changed his marketer after the Vorcaro case. Estado de Minas registered that Carlos Bolsonaro suggested that Zema is riding the wave of Flávio's crisis — signal of fracture within the clan itself.

In the government field, O Globo reported that Lula participated in a ceremony at TCU alongside Davi Alcolumbre — signal of institutional rapprochement between Executive and Senate after weeks of tension over the nomination of Jorge Messias to STF. In parallel, O Globo informed that PF investigates an amendment by Flávio Bolsonaro for an NGO suspected of being part of a deviation scheme — additional investigation that adds to the Vorcaro case.

4. Day's divergences

Market × research: Polymarket continues more aggressive than any published national poll. Gap of +22.15pp Lula × Flávio in the market vs +12.7pp in AtlasIntel (May 19), +8.7pp in Vox (May 20) and +3pp in Datafolha (May 16). The market × average-of-polls divergence is approximately 14pp — the market prices not only what polls have already captured but also the cumulative effect of subsequent events (PSDB-Aécio, Caiado/Zema positioning, Flávio changing strategist) that no poll in the field the previous week could measure.

Research × research: AtlasIntel 1º turno (47% × 34.3% gap +12.7pp) vs Vox 2º turno (46.8% × 38.1% gap +8.7pp) vs Datafolha 1º turno (38% × 35% gap +3pp). The AtlasIntel-Vox convergence in the 2º turno is significant (gap between the two of only 1.6pp in Lula's advantage over Flávio). The larger discrepancy remains between AtlasIntel/Vox and Datafolha — explainable by the time window (Datafolha in the field before Vorcaro repercussion reached the sample) and methodology (AtlasIntel included mention of the audio in the questionnaire, confirmed by the company itself on May 19).

Market × institutional field: Polymarket priced Flávio's continued collapse in 32 hours (drop of ↓4.80pp) without any single explosive institutional event. The movement appears to reflect accumulation: PSDB-Aécio + Caiado/Zema positioning + Michelle/Salles + Flávio changing strategist + PF investigating amendment. The market integrates cumulative signals that would not make headlines individually.

In summary

Vox Brasil confirmed Flávio's deterioration captured by AtlasIntel, with convergence between the two institutes around +7-9pp Lula in the runoff. The residual market × research divergence of ~14pp points to pricing of post-poll events that Datafolha on May 22 may validate or refute.
The non-Bolsonaro right entered a phase of explicit movement: PSDB discusses Aécio, Caiado delivers speech about being "contaminated by Vorcaro", Zema speaks of "credibility", Michelle criticizes "alliance with evil". Polymarket priced first material traction of the replacement narrative (Michelle 2L at 3.25%, ↑1.70pp) and Renan consolidating second place in the presidential race.
PL yielded for the first time in the cycle: Senate at 76.00% (↓2.50pp), 2L Flávio at 56.00% (↓6.50pp, PL base yields), Republicanos collapses to 2.50%. The technical reading: the market began to reprice PL hegemony in the Senate in parallel with Flávio's individual collapse — sign that the crisis may have consequences beyond the presidential ticket.—

Sources Consulted

Articles with direct links to the news:

Secondary articles (Google News redirect URL — click resolves to article):

Sources cited in this text: Vox Brasil via Poder360, Exame, Folha de Alphaville, Revista Fórum; AtlasIntel May 19; Datafolha May 16 via TSE; O Globo; Folha de S.Paulo; Estadão; BBC; CNN Brasil; Estado de Minas; Polymarket; TSE disclosure portal.

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

Integration: to see the live data and candidate analyses in detail, access the full Dashboard. To understand the method in depth, read The Method.

May 21, 2026