The model earns a modest edge while the draw keeps its weight

A useful forecast needs to earn its confidence, and the advantage here is modest. The model scorecard is current as of 5 October 2026. Predictions the model called correctly: 209 of 455 reconciled predictions since 13 July 2026. The model's accuracy: 45.9% of 455 reconciled predictions since 13 July 2026. The preferred result deserves attention without becoming a promise. The gap between selecting a favourite and knowing what will happen remains central to reading the forecast well. Home sides won 197 of the 455 matches the model has been graded on. The comparison matters because a model should improve on a simple habit before its complexity earns much credit. The model's accuracy is 2.6 percentage points above the always-home baseline. Home backing offers a useful starting point. The model called 12 more of the 455 matches correctly than backing the home side every time would have. Beating that starting point is encouraging, while the modest separation argues for restrained language before the next fixture. The model's Brier score across 455 reconciled predictions: 0.6246. A Brier score measures the squared distance between the probabilities and the outcome; lower is better. A confident forecast that fails deserves a different judgement from a narrow preference that fails. Accuracy answers whether the leading choice was right. Probability quality also asks how much room the forecast left for the alternatives. The model made a draw its top pick in 0 of the 455 reconciled predictions. The draw is easy to lose sight of when the display puts a team at the top. 119 of the 455 matches the model has been graded on ended in a draw (26.2%). A result can remain plausible without being the leading choice. The model's average probability for a draw across the 455 reconciled predictions was 24.2%. The focus on the preferred name obscures that distinction and makes the forecast appear more certain than the probabilities themselves. The model made the home side its top pick in 321 of the 455 reconciled predictions. The preference for hosts is clear. When the model's top pick was the home side it was right 155 times in 321. The useful question is whether each preference earns its confidence against the alternatives. The model made the away side its top pick in 134 of the 455 reconciled predictions. The venue can help establish an expectation without settling an individual contest. When the model's top pick was the away side it was right 54 times in 134. A forecast becomes more informative when the opposing result remains visible. The model's average probability for a home win across the 455 reconciled predictions was 45.6%. The travelling sides remain relevant even when the model prefers the hosts. The model's average probability for an away win across the 455 reconciled predictions was 30.2%. The probabilities describe possibilities, rather than instructions to ignore the less likely result. Away sides won 139 of the 455 matches the model has been graded on. The whole distribution matters when judging an outcome, particularly if the favourite fails. Bayern beat Union Berlin 7-0 on 18 September 2026 in the Bundesliga. The margins differ sharply, but the probabilities concern the result category. The model gave Bayern an 87.2% chance of beating Union Berlin on 18 September 2026 (Bundesliga). Backing the right side does not mean predicting the size of the victory. The model gave Arsenal an 83.1% chance of beating Chelsea on 6 September 2026 (Premier League). Arsenal beat Chelsea 2-1 on 6 September 2026 in the Premier League. The model's accuracy for matches played in July 2026: 37.5% (12 of 32). The model's Brier score for matches played in July 2026: 0.6814 across 32 reconciled predictions. The opening month is too thin a foundation for a sweeping verdict. A weak early return deserves attention, but a smaller workload makes it sensitive to the next results. The sensible response is to keep watching whether the probabilities improve as the grading covers more football. The model's accuracy for matches played in August 2026: 45.5% (87 of 191). The larger workloads carry more of the argument than a flattering return from a handful of fixtures. The model's Brier score for matches played in August 2026: 0.6251 across 191 reconciled predictions. Improvement is encouraging when it appears in probability quality as well as in the leading choice. The model's accuracy for matches played in September 2026: 47.2% (108 of 229). A favourite still needs cautious language because even a better month leaves plenty of room for the other outcomes. The model's Brier score for matches played in September 2026: 0.618 across 229 reconciled predictions. The model's accuracy for matches played in October 2026: 66.7% (2 of 3). The model's Brier score for matches played in October 2026: 0.4945 across 3 reconciled predictions. That denominator should stop any celebration. A handful of matches cannot establish a lasting improvement, howev