Understanding our numbers
Every technical term in the simulator, explained in plain language — with what to remember to read our numbers correctly.
A projection answers "what would today's polls produce, applied to the electoral map?" A prediction would claim to guess the result on election day. We do the first, never the second: between now and the vote, opinion will shift — and our numbers will shift with it, at every update. Hence our motto: projections, not predictions.
A method that takes its name from the casino, and that's no accident: rather than calculating a single result, the election is replayed 10,000 times, varying each time what the polls might have missed — their margin of error, their regional blind spots, local surprises. Each replay is a scenario. Counting the scenarios gives the probabilities: if the PQ wins a majority in 4,100 scenarios out of 10,000, its chances of a majority are 41%.
The National Assembly has 127 seats: a party has a majority at 64 seats or more — it governs alone. "PQ minority 53%" reads: in 53% of our scenarios, the PQ finishes first but under 64 seats — it would govern without a majority.
The result in the middle: half the scenarios give more, half give less. "PQ, median 61" means: in half of our 10,000 scenarios, the PQ tops 61 seats; in the other half, it falls short. The median can differ from the map's count — see the point and the range below.
The coloured band in our charts runs from the 5th percentile (p05) to the 95th percentile (p95): only 5% of scenarios fall below the first, only 5% above the second. In other words: 9 scenarios out of 10 fall within the band. "PQ 39 to 75" reads: it would be surprising — not impossible — for the PQ to fall outside that zone.
The map and the banner display the point: each riding assigned to whoever leads there today — one readable scenario, the central one. The odds display the range: what seats become when uncertainty is allowed to play out. The two can differ without contradicting each other: a party running second in thirty ridings can be first nowhere (0 seats at the point) while still winning somewhere in nearly every scenario (a much higher median). This isn't a flaw: they're two different questions.
A blend of now and forecast: a projection valid at the moment it is published, under two stated conditions — if the election were held today, and if the polls are accurate. What the campaign will change between now and the vote is deliberately left out.
We don't follow a single poll, but their combination: each poll receives a weight based on the polling firm that ran it, its type, and its age — a recent poll weighs more than an old one, and its weight halves every 14 days. The list of polls used is public (the polls behind this update); the exact coefficients are our house recipe.
How province-wide polls become 127 riding-level results: each polling division's 2022 result is stretched or compressed according to the movement in its region. If the CAQ is at half its 2022 score in a region, every polling division in that region has its CAQ share reduced in the same proportion, and the whole is then rescaled to 100%.
The smallest building block of the election: your polling station, a few hundred voters (414 at the median). The 2026 map counts 16,756 of them. This is the scale our calculations work at — not the average of a riding.
Polls measure the population; but not everyone votes — in 2022, 74% of those 55 and older voted, against 54% of 18-34 year-olds (Élections Québec). Our projections adjust voting intentions toward the likely electorate: the people who actually turn out, a bit older than the surveyed population.
Every riding has a life of its own: a star candidate, a local issue, an active ground campaign. By measuring past elections, we know how much a riding typically deviates from its region's movement — from 4 to 7 points depending on the party. Our simulations shake each riding by this measured amount: that's why the ranges are wide, and that's honest.
The risk that the margins of error in every poll lean the same way at the same time — it has happened: in the last three elections, the final gap favoured the PQ every time, and the CAQ was underestimated by 4 to 5 points in 2018 and 2022. Our simulations include this risk, calibrated on these measured gaps. It's what opens the tails of the ranges — the collapse or surprise scenarios that an overconfident model would declare impossible.
Ridings where the leader's edge is under 3 points — the zone where anything can tip. From 1 to 3 points, it's very close; under 1 point, it's a toss-up.
Our 10,000 scenarios start from a recorded random seed — the seed. With it, the simulation can be replayed identically, scenario for scenario, years later. A number that can't be replayed isn't a defensible number; all of ours are.
The full chain — data, tests, calibration — is described in our method.