In late July 2026 the Icelandic and Nordic leagues on this site were posting extraordinary goal rates, and our ratings were dutifully chasing them upward. The question worth asking about any hot start is simple: how much of this survives?
The model blends each team's rate with a league base rate — the division's own goals average — which anchors teams with thin evidence. The flaw we found was that this anchor was computed from the current season alone. Six rounds into a campaign, a league that had happened to produce a run of 4-3s was handing every fixture in that division an inflated starting point. Worse, the effect compounded: high anchor lifts every team's rating, which lifts the next set of estimates.
Across our full dataset, a league's goals average over its first six rounds is a poor predictor of its own full-season figure — the early number regresses heavily toward the previous season's mark. Leagues do drift in character, but slowly, over campaigns, not over fortnights. A division that averaged 2.7 goals last year and opens at 3.4 will almost always end up nearer 2.8 than 3.4.
The league anchor now blends the current season with the previous one, weighted by how far the season has actually progressed: early on it leans heavily on last season's settled figure, and by season's end the current campaign carries roughly half. Team-level ratings still move fast — last season's identity fades out over a team's first 14 games — so a genuinely transformed side is picked up quickly. What can no longer happen is an entire division being swept upward by a fortnight of high-scoring luck.
This is the general shape of most model improvements we make: not a cleverer formula, but a stricter answer to the question how much evidence is this actually?
Method: walk-forward testing — every prediction is made using only information available before kick-off, then graded against what happened. Studies run June-July 2026 on the site's own dataset. Questions: contact us.