Es Posible
The U.S. midterms and Brazil's presidential race: Lula or Bolsonaro
Who called Trump, Keiko Fujimori and Abelardo de la Espriella when the polls said otherwise?
I did. In public, dated, months ahead.
Today Brazil is tied and Washington is up for grabs. I have three numbers that break from the consensus, and I am signing again before anyone votes.
Three elections the polls did not see coming, or saw too late. The same signature ahead of all of them.
I called it in May 2024 · 184 days ahead
I said it before Joe Biden dropped off the ticket, and I held it when Kamala Harris surged and the national averages put her ahead for much of the summer and fall.
Result: November 5, 2024. He wins all seven swing states and the popular vote.I called it on July 16, 2025 · 340 days ahead
The same day he announced his run, with no party and no polls in his favor. A week before the first round, the four main pollsters had Iván Cepeda ahead, and Invamer gave Cepeda the runoff 52.4 to 45.3.
Result: June 21, 2026. He wins 49.66 to 48.70, the narrowest margin in Colombian history.I called it in February 2026 · 142 days ahead
She carried three lost runoffs: 2011, 2016 and 2021. She took barely 17 percent in the first round, and the last Ipsos poll before the vote had Roberto Sánchez ahead, 43.8 to 43.2.
Result: Peru's electoral office declares her the winner with 50.135 percent. Final margin: 49,641 votes.In all three I read the same thing: a voter fed up with crime and with the government of the day, who doesn't answer the pollster, or answers something else. In the United States I saw it before the Democratic ticket changed. In Colombia, the day a lawyer with no party raised his hand. In Peru, when Fujimori's movement looked finished.
When a pattern repeats three times, it stops being luck and becomes a method.
Every call is dated in the public record of Simón Said It First, at registro.simonlevy.mx. Today I apply the same method to three more elections.
Two points of vote share is four points of margin. That is almost exactly how far the national polls missed in 2020. What happens if that error repeats is further down, with names and numbers.
There is a voter the polls cannot see, and in all three elections with Trump on the ballot that voter decided the outcome.
This voter does not lie. This voter does not answer. Does not pick up a call from an unknown number, does not fill out the online panel, does not tell a stranger how they will vote. And when the pollster weights the sample, this voter gets replaced by someone who looks the same on paper but does not vote the same.
In 2016 the state polls missed this voter in Wisconsin, Michigan and Pennsylvania. In 2020 the American Association for Public Opinion Research documented the largest national polling error in forty years, all of it in one direction. In 2024 it happened again. Smaller, same direction.
Brazil has its own version. In the first round of 2022, the traditional pollsters gave Jair Bolsonaro several points less than he actually got. The closest was AtlasIntel, which today has his son tied with Lula.
And there is one vote I know well because it belongs to many of my readers: the Latino vote. In 2024 nearly half of Latino voters went Republican, and almost no poll saw it coming at that size.
My forecast does not contradict the polls. It corrects them for the size of the vote that doesn't answer.
It is worth stating plainly what the models say today, because my numbers break from them. Decision Desk gives Democrats a 55 percent chance of winning the Senate and a 76 percent chance of winning the House, with 234 seats. Nate Silver has the generic ballot at D+9.6. AtlasIntel has Lula and Flávio tied in the runoff. My three numbers depart from that consensus in a single direction, for a single reason.
“The polls aren't lying.Chapter 01
There is a voter who doesn't answer.
Before my numbers, everyone else's.
| Race | What the consensus says | Source |
|---|---|---|
| U.S. Senate | 55% Democratic odds · 51 projected seats | Decision Desk HQ |
| U.S. House | 76% Democratic · 234 seats | Decision Desk HQ |
| Generic ballot | D+9.6 | Silver Bulletin |
| Split Congress | D House and R Senate: about 30% | Kalshi |
| Brazil runoff | Flávio 47.7 · Lula 47.6 | AtlasIntel/Bloomberg |
That is the consensus, and it is reasonable. It is built from hundreds of polls and millions of simulations. But it is built from the same polls that missed the voter who decided 2016, 2020, 2024 and Brazil's 2022 first round.
So I did what a model does not do: I corrected for that voter. The result is three numbers. On two I agree with the direction of the consensus, but not its size. On one I go against everyone.
And that one decides who runs Washington for the next two years.
Inside: the Senate, the House and Brazil with exact figures, which of the three goes against the entire consensus, the command center with both error dials, the maps of the United States and Brazil, the seat charts that recalculate on their own, and the signed and dated scorecard.
Every figure sourced. Published before anyone votes.