Sunderland vs Leeds United

The model projects 9.31 total corners for this fixture, against a Premier League baseline of 10.4. That gap is material. Both clubs carry recent match averages that track below league norms, with Sunderland averaging 8.3 per game and Leeds at 8.0 across their respective recent samples, and those compressed averages are the primary input driving the model's output well below the baseline figure.

The variance is the real obstacle. Leeds' last three match totals read 6, 13, 5, a sequence that resists any clean central estimate; Sunderland's sequence of 12, 6, 7 introduces its own competing signal. The 13 and the 12 are each well above either side's own average, and the model has to weigh whether those figures reflect structural shifts or isolated high-tempo contexts. It can't resolve that cleanly.

That ambiguity drives the medium-confidence classification. The averages pull the projection below the league baseline, but the outlier results create a wider uncertainty band than the data would need to support high confidence. Noise complicates it.

Medium confidence warrants proportionate sizing. Treating this identically to a high-confidence tier call would misread what the model is communicating, and the 9.31 projection should not carry full weight given the variance in recent match totals for both sides. Size accordingly.

Across 150 settled medium-confidence calls in the Premier League, the model records a 47.3% win rate.