Agent Lab — a live forecast exhibit · Applied AI
World Cup 2026 · a live forecast agent

World Cup forecasts, with the receipts.

Every match starts with a statistical baseline — the same method for every team, fit on decades of results. Then an AI agent checks what the model can’t see: injuries, line-ups, the week’s news, and the match-day conditions that swing a game.

For each fixture, the agent says whether the evidence is strong enough to move the forecast, and by how much — with the sources behind it. Once the match is played, we score it against the model. The agent gets an opinion; the scoreboard gets the last word.

A recent call
France v Iraq
FRA 70.2/20.5/9.0 82.3/13.3/4.3+12.1 pp
Likely score FRA 2–0 IRQ 2.1–0.3 xg
Strengthened the lean toward France.

France WWWLW (4 wins in last 5); Iraq LWWDL (2 wins, 1 draw, 2 losses in last 5).

Grades at kickoff 6 sources the full call →

Calibrated on 47,000+ international matches · scored against real results as they come in · the agent reads each group-stage fixture as its kickoff nears.

What changed

Champion odds over the last day

The agent, live

Updated All calls →

For each fixture the agent explains what sits behind the model’s number and adjusts it where the match-day context warrants — usually by a few points, the size of the evidence. Most calls stay close to the baseline; its sharpest move so far is France +12.1 pp.

How calls are graded
Once a match is played, we show whether the agent’s adjustment helped or hurt versus the model. Method →
United States v Paraguay
played
USA 42.6/27.9/29.5 50.7/26.7/22.7+8.1 pp
Agent called USA 1–0 PAR 1.3–0.6 xg final 4–1
Strengthened the lean toward United States.

USA recent form WLLWL, -0.03 pts/match vs expectation vs avg opp rank #11; Paraguay LWWLW, +0.35 pts/match vs expectation vs avg opp rank #51. Paraguay overperforming against weaker opposition, USA slightly underperforming against strong opposition.

Better than the model call right 9 sources the full call →
Qatar v Switzerland
QAT 12.4/18.6/68.4 13.0/20.0/67.0-1.4 pp
Likely score QAT 0–1 SUI 0.3–1.7 xg
Softened the lean toward Switzerland.

Qatar has LDLLD in last 5, -1.02 pts/match vs expectation vs avg opp rank #85; Switzerland has DLDWD, -0.42 pts/match vs expectation vs avg opp rank #37. Both underperform, but Qatar's underperformance is more severe and against weaker opposition.

Grades at kickoff 3 sources the full call →
Brazil v Morocco
BRA 46.0/32.0/22.0 53.3/27.7/19.0+7.3 pp
Likely score BRA 1–0 MAR 1.2–0.4 xg
Strengthened the lean toward Brazil.

Brazil's recent form is DLWWW (+0.18 pts/match vs expectation) against opponents averaging strength-rank #35; Morocco's is DWWWD (-0.00 pts/match) vs opponents averaging #61. Both are in line with expectations, but Brazil faced stronger opposition.

Grades at kickoff 9 sources the full call →

The forecast, in brief

Full 48-team table →

The top of the field — the baseline the agent reasons from. Each market’s de-vigged champion price sits beside the model’s, never blended. Top 8 ≈ 70% of the title mass — an unusually flat field.

# Team Champion Final Semi Polymarket Kalshi

How the forecast agent works

1
The model sets the prior

Team strength estimated from 47,000+ historical international results, with recent matches weighted most. Simulated tournaments are an output of that strength, not a teacher. The same disciplined method for every match, no opinions in it.

2
The agent reads the live context

For each fixture it searches the web and weighs what the model structurally can’t see — injuries, line-ups, suspensions, the week’s news — and the match-day conditions that bite: altitude, heat, a roof that closes. It cites what it found and says which way it points.

3
And it gets graded

Every adjustment is scored against the model once the match is played. Soccer is hard to call — low-scoring, and national sides barely play together — so we keep expectations honest and show how the agent does, win or lose. The method →