Today carries this line under The record:
223 European league predictions scored: 94 right, 129 wrong, log loss 1.0721, worse than the league constant (1.0626).
What is being counted
Every European league prediction BetDNA published before kick-off, and only the earliest one for each fixture. A forecast written after the result, or rewritten with the result in hand, never counts - that is what as-of means. 223 scored: 94 right, 129 wrong, a hit rate of 42%.
What log loss means
Log loss is minus the natural logarithm of the probability the model gave the outcome that happened, averaged over every scored forecast. Lower is better. It punishes confidence more than it rewards it: a confident forecast on a result that did not happen costs far more than a hesitant one on a result that did.
Two lines to read it against. Knowing nothing - a third each way on a football match - scores 1.0986. The league constant, the long-run rate of home wins, draws and away wins in each division over the 155,079-match backtest, scores 1.0626. The model stands at 1.0721. Better than the constant means the model knows something the base rate does not; worse means it does not yet, and the page says so in those words.
Why a bad week is not a broken model
Football results are noisy. A model that is exactly right about its probabilities is wrong about the result most weekends, because most of its picks are less likely than not. The score moves in the direction of the truth over hundreds of forecasts, not tens. What would show a broken model is a log loss that stays worse than the league constant as the count grows - and that is precisely why the count is printed with it.
Where it lives
On Today, under The record, with By division and the latest scored beneath it. It is rebuilt with the board at 00:05 UTC each day.