For every match we publish one combination — the side we think won’t lose, that side’s shots, and that side’s corners. This puts them together across matches. Pick how many games you want and see what the model gives the lot.
Every combination we publish is settled afterwards from the real shot, corner and goal counts — nothing is graded by hand. Across 344 finished matches in the same range we pick from, one game’s combination was given 57.1% on average and landed 62.8% of the time.
| Games | Finished sets | We said | It landed |
|---|---|---|---|
| 1 | 344 | 57.1% | 62.8% |
| 2 | 159 | 32.6% | 37.8% |
| 3 | 100 | 18.7% | 23.1% |
| 4 | 71 | 10.6% | 14.7% |
The multi-game rows are measured on sets that share no match, so no match is counted twice. These have landed a little more often than we said, by more than normal variation would explain on this many matches. The model is running slightly under-confident in this range — the safe direction to be wrong in, but still wrong, and we would rather show you that than round it away. Percentages, not prices: this is what we think will happen, not advice to stake money on it.
One caveat we would rather state than bury: these matches are filtered to the same chance range the page picks from, but we only started recording how much of each club’s own history sat behind a shots or corners line after this record began — so a few of them are matches today’s page would leave out for thin data. That makes this a floor for what we publish now, not an exact like-for-like.