The honest answer, with numbers
Short answer: good probabilities, not promises. As of 2026-11-08 the model has settled 227 official predictions in public — hit rates, misses and calibration scores all on the record.
Read the numbers honestly: a 60% favourite still loses about two matches in five, and a season is a small sample — what matters is calibration over many matches, which is why every pick is locked before kick-off and settled against the real score, misses included. Brier and RPS grade the probabilities themselves, not just the picks; lower is better, and both beat random guessing only if the model is genuinely calibrated.
Strength ratings
Recent performances and home advantage set each side's expected goals.
Expected goals become a probability for every scoreline, corrected for low-score draws.
The call locks 1 h before kick-off and is settled against the real final score in public.
Accuracy by league, today's calls, the methodology and yesterday's graded picks — every number on this page links to its source.