Polls measure what voters say today. The Sondagio model starts from the same thing, electorate by electorate: the share of each first-round candidate's voters who would back each finalist, as measured by the pollsters' vote-transfer tables. It then moves it toward what those voters actually did in the 2024 legislative run-offs, estimated from the results of every polling station.
How far? As far as history suggests. 172 days before the run-off, the polls gave Emmanuel Macron 63.5% in 2017 (he won 66.1%) and 55.6% in 2022 (he won 58.6%): the Republican front against the RN rebuilt itself late in the campaign, by 2.6 points in 2017 and 3.0 points in 2022. The model applies that late swing to the match-up against the centre, the only one 2017 and 2022 can measure, which means moving each electorate 26% of the way to its 2024 behaviour; it does the same in the other match-ups.
So a match-up whose voters already behave as they did in 2024 barely moves; one where they have drifted away moves more. The tables below show where each electorate goes, match-up by match-up.
Checked against the two previous run-offs: applied to the actual first-round result, the transfer tables the pollsters published between the rounds gave Emmanuel Macron 59.7% in 2017 (he won 66.1%) and 53.7% in 2022 (he won 58.6%). They understated the Republican front by more than the model corrects for: if that happens again in 2027, the model's figures for Le Pen's opponent are on the cautious side.
Limits: the transfer tables come from a single pollster (Toluna Harris Interactive, the only one to publish them in its notices) and the late swing rests on two elections. The range (8 simulations out of 10) allows for that, for margins of error, for uncertainty about the first round, and for campaign swings, which narrow as the vote gets closer.