Switzerland v Canada · the model leans Canada

This is a Pick. Switzerland v Canada in FIFA World Cup 2026 on Wednesday, and the model and the market disagree on the Canada side by +15.2 pts. That's the gap the verdict rests on.

Elo numbers going in: Switzerland 1830, Canada 1860. That's a 30-point lead for Canada — closer than a casual reader of the market price might assume. Both Elo numbers are pulled live from the public Elo source. No host, home-ground, or altitude adjustments apply here; the prior reads straight from Elo.

Three-way split from the model: Switzerland to win at 33%, the draw at 28%, and Canada to win at 39%. The Desk publishes a three-way read against the same W/D/L resolution Polymarket and Kalshi settle on — not a scoreline guess. Canada sits at 39% of the mass.

On the same three-way, Polymarket: Switzerland to win at 47%, the draw at 30%, and Canada to win at 23%. The Canada side trades at 23% on Polymarket, which prices out at American odds of +335. That's where the engine differs: our number reads 39%, +15.2 pts wider.

Pick gate is a lower-bound check, not the point estimate. A 100-sample bootstrap perturbs the model estimate (±20), the host bonus (±15), and the altitude bonus (±10), then reads the 5th-percentile probability back out. On the Canada side that lower bound is 32% — +8.7 pts clear of the market's 23%, comfortably past our +3 pts threshold. The claim is narrow: on this fixture, on this side, the price is below what the underlying read supports. We don't argue the market is broadly wrong — across the priced field the model and the line agree more often than they disagree. We surface the ones where they don't.

The read above is the most recent one, not a frozen call. With kickoff 24 days out, refresh is weekly; daily cadence picks up inside the final five days. Injury news, the actual XI, and conditions at kickoff all change what the model sees. We re-publish on every refresh. The page shows what the engine reads; what to do with that information is a reader's call.