Our track record
Every prediction locks at kick-off and is never edited. We mark it on the result, the exact score and log loss, which rewards confident correct calls and punishes confident wrong ones (lower is better). The bookmakers' closing odds are the benchmark.
2026-27 so far
| Predictor | Type | Games | Right result | Exact score | Log loss |
|---|---|---|---|---|---|
| Stats model | Pre-launch | 50 | 44% | 8% | 1.043 |
| Jev (AI) | Pre-launch | 50 | 50% | 6% | 1.049 |
| Consensus | Pre-launch | 50 | 44% | – | 1.042 |
| Bookmakers | Closing odds | 50 | 46% | – | 1.053 |
"Pre-launch" covers games played before the site went live: those predictions were generated afterwards, using only data from before each game. That makes them an honest test of the method, but they weren't published in advance. For log loss, every predictor's probabilities are floored at 1% per outcome, so one "0% chance" call can't make the average meaningless.
Long-run test since 2005-06
| Season | Our accuracy | Bookmakers | Our log loss | Bookmakers |
|---|---|---|---|---|
| 2026-27 | 44% | 46% | 1.043 | 1.053 |
| 2025-26 | 48% | 49% | 1.025 | 1.015 |
| 2024-25 | 55% | 55% | 0.970 | 0.966 |
| 2023-24 | 58% | 60% | 0.926 | 0.900 |
| 2022-23 | 54% | 55% | 0.979 | 0.962 |
| 2021-22 | 54% | 58% | 0.957 | 0.937 |
| 2020-21 | 51% | 52% | 1.011 | 0.997 |
| 2019-20 | 53% | 54% | 0.973 | 0.972 |
| 2018-19 | 61% | 58% | 0.903 | 0.891 |
| 2017-18 | 56% | 54% | 0.955 | 0.941 |
| 2016-17 | 58% | 61% | 0.939 | 0.906 |
| 2015-16 | 47% | 47% | 1.036 | 1.033 |
| 2014-15 | 53% | 53% | 0.980 | 0.970 |
| 2013-14 | 58% | 61% | 0.951 | 0.928 |
| 2012-13 | 52% | 53% | 0.966 | 0.957 |
| 2011-12 | 53% | 52% | 0.987 | 0.984 |
| 2010-11 | 49% | 50% | 1.021 | 1.011 |
| 2009-10 | 55% | 56% | 0.939 | 0.929 |
| 2008-09 | 57% | 54% | 0.960 | 0.960 |
| 2007-08 | 56% | 58% | 0.934 | 0.924 |
| 2006-07 | 52% | 52% | 0.985 | 0.973 |
| 2005-06 | 58% | 57% | 0.948 | 0.937 |
Walk-forward test: each week the model was refitted using only earlier matches, then predicted that week. Settings were tuned on 2005–2017 and checked, untouched, on 2018 onwards.
Every result this season
| Date | Match | Result | Predicted | H · D · A | |
|---|---|---|---|---|---|
| Sun 20 Sep | 1–1 | 1–2 | 34 · 27 · 40 | Missed | |
| Sun 20 Sep | 5–3 | 2–0 | 62 · 23 · 15 | Right result | |
| Sun 20 Sep | 0–0 | 2–1 | 54 · 24 · 22 | Missed | |
| Sun 20 Sep | 0–1 | 2–1 | 44 · 24 · 32 | Missed | |
| Sat 19 Sep | 0–1 | 1–0 | 55 · 25 · 19 | Missed | |
| Sat 19 Sep | 2–1 | 2–1 | 47 · 26 · 27 | Exact score | |
| Sat 19 Sep | 1–0 | 2–0 | 59 · 24 · 17 | Right result | |
| Sat 19 Sep | 3–0 | 1–2 | 24 · 26 · 50 | Missed | |
| Sat 19 Sep | 2–3 | 2–1 | 42 · 26 · 32 | Missed | |
| Fri 18 Sep | 3–0 | 2–1 | 43 · 24 · 33 | Right result | |
| Mon 14 Sep | 4–1 | 2–1 | 46 · 25 · 29 | Right result | |
| Sun 13 Sep | 0–1 | 1–2 | 31 · 25 · 44 | Right result | |
| Sun 13 Sep | 0–5 | 1–2 | 31 · 27 · 41 | Right result | |
| Sat 12 Sep | 0–2 | 0–1 | 19 · 27 · 54 | Right result | |
| Sat 12 Sep | 0–0 | 2–1 | 39 · 28 · 33 | Missed | |
| Sat 12 Sep | 0–0 | 2–1 | 58 · 22 · 19 | Missed | |
| Sat 12 Sep | 2–3 | 2–1 | 62 · 21 · 17 | Missed | |
| Sat 12 Sep | 2–2 | 2–1 | 53 · 24 · 22 | Missed | |
| Sat 12 Sep | 2–2 | 2–1 | 49 · 24 · 27 | Missed | |
| Sat 12 Sep | 1–2 | 2–1 | 43 · 27 · 30 | Missed | |
| Sun 6 Sep | 2–1 | 2–0 | 66 · 20 · 14 | Right result | |
| Sun 6 Sep | 2–2 | 2–1 | 37 · 27 · 36 | Missed | |
| Sat 5 Sep | 0–0 | 2–1 | 37 · 27 · 35 | Missed | |
| Sat 5 Sep | 0–0 | 2–1 | 46 · 26 · 28 | Missed | |
| Sat 5 Sep | 1–0 | 2–0 | 71 · 18 · 11 | Right result | |
| Sat 5 Sep | 2–3 | 2–1 | 42 · 27 · 31 | Missed | |
| Sat 5 Sep | 1–1 | 2–1 | 44 · 26 · 30 | Missed | |
| Sat 5 Sep | 1–1 | 2–1 | 49 · 26 · 25 | Missed | |
| Sat 5 Sep | 2–2 | 2–1 | 40 · 24 · 36 | Missed | |
| Fri 4 Sep | 0–2 | 1–2 | 17 · 20 · 63 | Right result | |
| Mon 31 Aug | 0–1 | 0–1 | 21 · 25 · 53 | Exact score | |
| Sun 30 Aug | 5–2 | 2–0 | 70 · 18 · 12 | Right result | |
| Sun 30 Aug | 1–0 | 1–0 | 40 · 29 · 31 | Exact score | |
| Sun 30 Aug | 1–1 | 2–1 | 47 · 25 · 28 | Missed | |
| Sun 30 Aug | 4–3 | 2–1 | 41 · 25 · 34 | Right result | |
| Sat 29 Aug | 0–2 | 2–1 | 39 · 24 · 37 | Missed | |
| Sat 29 Aug | 0–1 | 2–1 | 40 · 27 · 33 | Missed | |
| Sat 29 Aug | 1–1 | 2–1 | 51 · 25 · 24 | Missed | |
| Sat 29 Aug | 2–2 | 2–1 | 55 · 23 · 21 | Missed | |
| Fri 28 Aug | 1–4 | 1–2 | 25 · 24 · 51 | Right result | |
| Mon 24 Aug | 2–3 | 2–1 | 40 · 26 · 35 | Missed | |
| Sun 23 Aug | 2–2 | 2–1 | 38 · 24 · 38 | Missed | |
| Sun 23 Aug | 2–1 | 2–1 | 56 · 23 · 22 | Exact score | |
| Sun 23 Aug | 4–0 | 2–1 | 45 · 26 · 30 | Right result | |
| Sat 22 Aug | 3–0 | 2–1 | 46 · 24 · 29 | Right result | |
| Sat 22 Aug | 0–1 | 1–2 | 36 · 27 · 36 | Right result | |
| Sat 22 Aug | 2–1 | 0–1 | 26 · 27 · 46 | Missed | |
| Sat 22 Aug | 2–0 | 1–0 | 42 · 28 · 30 | Right result | |
| Sat 22 Aug | 2–0 | 1–2 | 27 · 25 · 48 | Missed | |
| Fri 21 Aug | 3–0 | 2–0 | 71 · 19 · 10 | Right result |
Questions
How accurate are these Premier League predictions?
Over 8,030 matches since 2005-06, the model picked the right result 54% of the time, against 55% for the bookmakers' closing favourite. This season so far: 44% from 50 games.
Do the predictions beat the bookmakers?
Not over the long run. Our log loss is 0.969 against the bookmakers' 0.957 (lower is better). Closing odds are the sharpest public forecast, so getting close is the realistic aim. This season so far it is 1.043 against 1.053.
Are predictions changed after kick-off?
No. Every prediction locks at kick-off and is never edited; the accuracy page scores only those locked predictions, with games from before the site launched shown separately.