SFM Tracker

Weekly Top 20 Predictions by League

Reset
Matchday 3: Premier League (Completed Games Only)
One round — 18 graded picks. Small samples swing hard. See the full season.
SFM Top 20 Predictions
18
Predictions
7
Scored
38.9%
Hit Rate
Naive Top 20 Predictions
14
Predictions
4
Scored
28.6%
Hit Rate
SFM outperforms Naive:
38.9% vs 28.6% (+10.3 pp)
Brier Score: Premier League
0.2302
SFM
0.1812
Naive
Naive 0.1812 < SFM 0.2302
Calibration: Premier League
Hit Rate Over Time: Premier League (SFM vs Naive)
Premier League 2026/27 GD03 Completed
This GD: 38.9% (7/18) Overall: 38.9% (7/18) 1 gamedays completed
History: GD01 GD02 GD03 GD04 GD05
Rank Player Team vs SFM Naive Result
1 Erling Haaland Manchester City Coventry City 57.1% 58.3% 1 goal
2 Bukayo Saka Arsenal Chelsea 47.1% 22.4% No goal
3 Kai Havertz Arsenal Chelsea 43.1% 24.1% 1 goal
4 Alexander Isak Liverpool Ipswich Town 40.4% 34.4% 2 goals
6 Jack Hinshelwood Brighton & Hove Albion Leeds United 32.4% 19.1% Did not play
7 Cody Gakpo Liverpool Ipswich Town 32.1% 22.9% No goal
9 Gonzalo Garcia Fulham Crystal Palace 31.5% 12.9% No goal
12 Thierno Barry Everton Manchester United 28.7% 23.5% No goal
13 Keane Lewis Potter Brentford Sunderland 28.6% -- No goal
14 Anthony Elanga Newcastle United AFC Bournemouth 28.6% 8.8% No goal
16 Arnaud Kalimuendo Muinga Nottingham Forest Tottenham Hotspur 28.1% 27.7% Did not play
9 Martin Oedegaard Arsenal Chelsea 26.4% 16.6% 1 goal
14 Joshua King Fulham Crystal Palace 26.0% -- 1 goal
17 Chris Wood Nottingham Forest Tottenham Hotspur 24.8% 28.8% No goal
8 Emersonn Ipswich Town Liverpool 24.6% 19.9% No goal
20 Ollie Watkins Aston Villa Hull City 24.0% 33.9% Did not play
13 Kiernan Dewsbury Hall Everton Manchester United 23.7% 22.8% No goal
16 Dominik Szoboszlai Liverpool Ipswich Town 23.1% 15.1% No goal
19 Vitaly Janelt Brentford Sunderland 21.2% -- 1 goal
12 Marcus Tavernier AFC Bournemouth Newcastle United 20.1% 14.6% 1 goal
10 Omar Marmoush Tottenham Hotspur Nottingham Forest 17.3% 26.3% No goal
Understanding Hit Rate vs Brier Score

You might notice that hit rate and Brier Score can tell different stories.

Hit Rate

Simply counts: "How many of my top 20 picks scored?"

A naive model that always picks proven strikers (Haaland, Kane) will have a high hit rate because these players score often, regardless of the match context.

Brier Score

Asks: "How accurate were the probability estimates?"

If SFM says "32% chance" and Naive says "38% chance" for the same player who doesn't score, SFM gets a better Brier Score because its estimate was closer to reality.

Bottom line: Hit rate measures selection quality (who you pick), while Brier Score measures probability quality (how well-calibrated your predictions are). A model can pick slightly fewer scorers but still be more valuable if its probabilities are more trustworthy for betting or decision-making.
SFM Tracker
Top 20 Selection

For each league and gameday, we select the 20 players with the highest median probability of scoring at least one goal as predicted by the SFM.

Frozen Predictions

Predictions are locked before matches are played. This ensures transparent, verifiable performance tracking.

Fair Comparison

We compare SFM's top 20 picks against Naive's own top 20 picks (ranked by historical average). This is apples-to-apples.

Brier Score

Evaluation metric for probabilistic predictions. Measures both calibration and discrimination. Lower is better.