gyis mengus yandugs guxie lees 

 the second and final round of presidential elections. Polls suggest former leftist leader Luiz Inacio Lula da Silva is ahead of his far-right rival - the current president Jair Bolsonaro. But after the first round was closer than expected, some think that doesn't tell the whole story.

"I don't believe in the polls," says Janaina Devaney, a party shop owner in one of Sao Paulo's smarter neighbourhoods.

As she pumps up balloons for a client, gospel tunes are blaring out of a speaker by the till. Janaina is an evangelical Christian and she cites Jair Bolsonaro's family values as one of the main reasons she wants him to win.

"The streets say the exact opposite of the polls," she says. "I go to the demonstrations and they are peaceful. It's beautiful, the green and yellow colours [of the Brazilian flag], families, children, babies in buggies. If you just looked at the demonstrations across Brazil, you'd see it's so different."

But if the polls are correct, then it's not Bolsonaro but Lula who will take the presidency on Sunday. The latest Datafolha poll showed Lula with 53% of valid votes and Bolsonaro with 47%.

e month, had predicted as much as a 14 percentage point gap between Lula and Bolsonaro. In the end, it was just five points, vastly underestimating the support for the far-right leader.

The big gap between predictions and reality has propelled Bolsonaro's narrative that the polls are a "lie". But, say experts, it's not an exact science.

"People really believe in what they want to believe," says Felipe Nunes, political scientist and CEO of Quaest pollsters. "They look for evidence that confirms what they think. They don't look for evidence that helps them understand the situation better. As researchers, we have to understand that."

But what is harder to understand is a push by conservative lawmakers who now want to criminalise polls that later don't match the results.

"We would prefer, of course, to have the support of the officials in the country," admits Mr Nunes, defending his work. "We are very rigorous, we use scientific methods and sometimes we go wrong. Sometimes, the predictions don't go as we expect. And that's the role of