THE RECORD

Every match, against the forecast that was already published

The model called 17 of the 34 matches it had already forecast, and 8 of them were draws. Always backing the market favourite would have called 17.

Matches from 2026-09-16 to 2026-09-20. Page generated on 2026-09-24.

A hit rate on a 1X2 market means nothing on its own: always backing the favourite lands somewhere between 45 % and 55 % depending on the league. That is why the comparison sits next to the number and not in a footnote. The model and the favourite pointed at the same outcome in 30 of the 34. They tied: both called 17 right. That is what the backtest over 12,610 matches predicts — the model's weight in the blend with the market came out at zero. It does not beat the price.

A draw is about one in four results, and it is almost never the single most likely of the three — 7 times in 15,631 matches. A well-calibrated model misses those often.

Each match is compared against the model fit that was published before it was played. Nothing is recomputed with the result already known: that would give a prettier number and no value at all.