Poll tracker

Every reconciled first-round poll, grouped by the exact candidate field it tested. Polls testing different fields are never pooled: an average only means something within one configuration, so each is shown with its own.

Polls shown are those reconciled to a Commission des sondages notice. A sparse page is the standard working, not a gap.

The average is a weighted mean under the frozen specification's recency and sample-size rules (P8, C1 — the parameters travel with every published run). It is arithmetic on published polls. It is not the forecast: no error model, no transfer matrices, no probability.

Second-round duel polls appear once duel reconciliation to the notice register lands; none is shown until then, under the same inclusion rule.

Share by candidate, across the fields tested

Each row spans that candidate's lowest to highest share across every reconciled poll in the window — the width of the bar is the point: the number depends on who is on the ballot. Window: the 180 days to the most recent poll (from 2026-02-20), 13 polls.

No per-candidate average is shown here, deliberately: averaging across different candidate fields would pool polls that measured different things. The averages live below, one per configuration.

One candidate over time

Every reconciled poll for the selected candidate. Lines connect only repeated tests of the same field (a pollster re-running its own hypothesis); no line is ever drawn across different fields.

Dots are individual polls, not a series. Where the same field has only been tested once, there is nothing to connect — that sparseness is the state of the evidence, not a display fault.