Premier League

1X2 fit in the 1X2 column evaluates the saved home / draw / away probabilities against the actual outcome using Brier score: Σ(probability − actual outcome)², with 1 for the actual outcome and 0 for the other two. It does not grade the total goals or demand an exact score: 1–0 and 5–1 receive the same 1X2 fit for the same forecast. xG margin fit on forecast xG compares who was expected to dominate, and by how much, with the actual match xG: |(λ home − λ away) − (xG home − xG away)|, where a margin is home minus away. It is direction-aware: a favourite that really dominated is not marked down just for creating more or less than forecast in total, while the wrong direction gives a large gap. It does not depend on the final score or on total xG (see TOTAL xG fit). A low 1X2 fit and close xG margin fit can occur together; finishing variation may be one explanation.

TOTAL xG fit compares the displayed forecast total with the final goal count: |TOTAL xG − (home goals + away goals)|. This describes distance from the expected mean, not the probability of the exact score. The expected total is not a guaranteed result; a large difference in one match does not establish a model error. Separate frozen reference forecast_total_fit_v1: 250 retrospective forecasts from the same dates. Close: difference ≤ 0.65 goals; Poor: difference ≥ 2.00 goals; Fair: between them. This fit does not require actual match xG.

BTTS and Over 2.5 colours evaluate their own Yes/No outcomes using binary Brier score. Green/teal → amber → red means closer → less close. Good / Close is the lowest-error quarter of the reference; Low / Poor is the highest-error quarter. Fair is between them. Hover over a coloured metric for its values. Without a valid pre-match snapshot or both actual xG values, the corresponding indicator stays neutral.

xG margin fit uses its own frozen reference forecast_xg_margin_fit_v1: 137 retrospective LOAD90 forecasts with actual match xG, 15 Aug–20 Sep 2026. Close: gap ≤ 0.38; Poor: gap ≥ 1.41; Fair: between them. It supersedes the older per-team xG scale of forecast_fit_v1 for the xG (forecast) column.

Fixed reference forecast_fit_v1: 250 retrospective LOAD90 forecasts, 15 Aug–20 Sep 2026; actual xG available for 137 matches. This is a retrospective colour scale, not evidence of live forecast accuracy. One match cannot establish whether a probability model is well calibrated.