Team Strength

How to read the ratings

The ratings are log-rate adjustments. An offense of +0.10 means the team's expected runs-per-game is e^0.10 ≈ 1.10× the league baseline; a defense of +0.10 cuts opponent scoring by the same factor. Overall strength is the sum of the two.

The bars show the 80% credible interval of the posterior. Wider bars mean we have less certainty about that team — usually because few games have been played, or results have been inconsistent.

Current form is each team's run differential over their most recent 10 completed games — a snapshot of what they've actually done lately, separate from the long-run model rating above. A team can be a strong rated club (high overall) yet be in a cold patch (negative current form), or vice versa. The model uses this short-window number as an explicit term in the log-rate equation so a current slump or streak modulates the per-game prediction rather than waiting for the slow recency weighting to catch up.

Model rank vs 2026-form rank. The two ranks on each card answer different questions. 2026 form ranks teams purely on this season's run differential — what they've done so far. The model rank blends 2026 with recency-weighted prior seasons, so a team that was poor for years carries a sceptical prior even after a strong half-season (and a recent powerhouse keeps some shine after a slow start). That can look harsh — a club outplaying its run differential, or rebounding from a bad multi-year stretch, will sit lower on the model rank than its 2026 form suggests. It is deliberate: a walk-forward backtest on the held-out 2025 season shows the blended rating predicts future games more accurately than a current-season-only rating at every recency setting tested (lower log-loss and lower standings-recovery error). Half a season of results is genuinely noisy, and the prior is the model declining to overreact to it. The 2026-form rank is here so you can see exactly how much skepticism the model is applying, team by team.