Forecasting Agent

Run the forecasting agent yourself.

The same pipeline behind our 87 published forecasts: an Elo prior, 100,000 Monte-Carlo simulations, and Bayesian updates from key evidence — never a single market price.

01

Pick a question

Any match, group winner, or title odds. You ask; the forecasting agent does the rest.

Input
02

Layered evidence

Public data gathered by priority: recent national-team results, club form, injuries and lineups, environment (venue, altitude, weather). Every fact dated and sourced; odds pages discarded on sight.

Data layers · market-blind
03

Statistical baseline

Live Elo feeds a Davidson three-way model for matches; qualification and title questions run 100,000 Monte-Carlo tournament simulations on the official bracket.

Elo · Monte Carlo
04

Bayesian update

Key evidence becomes a bounded shift on the prior: at most ±8 points per match, nothing moves without evidence; every output carries a confidence tier.

±8pp · confidence tiers
05

Report & public scoring

Multilingual reports with full sources and methods; every question Brier-scored in public after settlement — wrong calls stay on record.

Traceable

Baseline formulas

π = 10R/400P(A) = πAπA + πB + ν·√(πAπB)P(draw) = ν·√(πAπB)πA + πB + ν·√(πAπB)pfinal = norm(pstat + Δevidence)

ν = 0.7 (draw parameter) · R is the live Elo rating · Δ is the evidence-driven bounded shift (|Δ| ≤ 8pp per match, renormalized)

Workbench

Ready
Pick a question→Layered evidence→Statistical baseline→Bayesian update→Report & public scoring

The full workbench is coming soon — try the sample run below.

Sample run

Pick a real question to see the agent's actual output

2026 FIFA World Cup opening match (Group A): Mexico vs South Africa, Estadio Azteca, Mexico City — 90-minute (regulation incl. stoppage time) result: Mexico win / Draw / South Africa win

Mexico73%
Draw20%
South Africa6.7%

Mexico are the clear stronger side — home crowd, altitude, form, and the underlying strength gap all point one way, so we put their win probability at ~73%; but World Cup openers are historically cagey and South Africa's low-block defending makes the draw (~20%) the alternative outcome most worth taking seriously, while a South Africa win is a thin tail event (~7%).

  • Hosts are 16W-6D-1L in World Cup openers (the only defeat: Qatar 2022), and Mexico add ~2,240m altitude at the Azteca, ~80,000 home fans, and an unbeaten-in-8 2026 form edge over a winless South Africa.
  • Host openers end drawn ~26% of the time (6 of 23, well above the Elo model's 14.9%), and South Africa spent ~9-10 days at altitude in Pachuca (~2,440m, higher than Mexico City), so the altitude edge must be discounted.
Full reasoning report →42 sources · confidence medium