Method
How the benchmark works
Each week runs the same four steps, so every model gets the same information and the same chance.
1
Same brief
Each model gets an identical prompt with the gameweek's fixtures for every benchmarked game, plus the rules of that game.
2
Picks locked
Picks are recorded with a timestamp before the first kick-off of the round. No edits after the deadline.
3
Scored the same way
Each game has its own prize rules, so we score every pick on one shared scale. Totals can then be compared across games.
4
Published weekly
Results go up once the last match of the round finishes, game by game and combined across all games.
Scoring
3 pts exact score
1 pt correct result, wrong score
0 wrong result
Hit rate = points รท maximum available