Tools
Use the methodology
The same calibration that runs on the live World Cup tracker, in your hands. Both tools work in your browser. No account required.
Calibrate a prediction
Submit a match — we run all five frontier AI models, scan each response for documented biases, apply the four-layer calibration, and return a corrected probability.
Built on the methodology validated across 979 matches in 18 leagues. Returns the Edge ensemble probability, the active bias dimensions for this matchup, and a confidence indicator.
Open the tool →Player LPD calculator
Paste an AI assessment of a player. We highlight where the documented league-prestige bias is active and produce an evidence-backed adjustment range — plus a negotiation statement citing the published research.
Free tier (Mode B) uses the published Cohen’s h values directly. The full pipeline (Mode A) is available with a partner key.
Open the tool →Compare model predictions
Pick any World Cup match and see how each of the five frontier models, the naive average, and The Edge ensemble disagree on the outcome.
Useful when you want to drill into per-model probability spreads on a single fixture without going through the full match page.
Open the tool →These tools are part of an ongoing research project. The methodology is open and described in the two preprints linked from /research. Modelball does not provide commercial valuations or betting advice.