Prediction Accuracy: August 2026
Every Best Pick and Value Analysis of the month, graded against the final result. Hits and misses count alike; void matches are left out.
Final figures for August 2026
Hit rate in August 2026
Best Picks
The single highest-conviction selection per fixture, across all markets.
+2.2 pts vs July 2026
Value Analysis
Selections where the model's probability diverges most from consensus market pricing.
-5.3 pts vs July 2026
Compared to guessing
An accuracy figure means nothing on its own. The only useful question is what a naive rule would have scored on exactly the same matches.
Measured across every graded fixture, not a selected subset. Home advantage is why the naive rule scores as well as it does; the gap above it is what the model contributes. n = 7,652.
Day by day in August 2026
Each line is a seven-day rolling average of settled selections, with a node marking the days behind it. Days with nothing settled break the line rather than being bridged across.
| Date | Best Picks | Value Analysis |
|---|---|---|
| 01.08.2026 | 73.4% | 61.6% |
| 02.08.2026 | 74.1% | 62.8% |
| 03.08.2026 | 72.2% | 61.8% |
| 04.08.2026 | 71.2% | 61.1% |
| 05.08.2026 | 69.9% | 59.8% |
| 06.08.2026 | 69.4% | 60.5% |
| 07.08.2026 | 68.2% | 61.8% |
| 08.08.2026 | 66.6% | 59.4% |
| 09.08.2026 | 62.6% | 58.5% |
| 10.08.2026 | 63.2% | 58.6% |
| 11.08.2026 | 63.4% | 58.5% |
| 12.08.2026 | 63.7% | 59.3% |
| 13.08.2026 | 64.7% | 58.2% |
| 14.08.2026 | 65.8% | 56.3% |
| 15.08.2026 | 63.5% | 55.5% |
| 16.08.2026 | 69.4% | 54.1% |
| 17.08.2026 | 70.2% | 54.5% |
| 18.08.2026 | 70.9% | 55.0% |
| 19.08.2026 | 71.9% | 54.0% |
| 20.08.2026 | 70.8% | 55.2% |
| 21.08.2026 | 69.5% | 55.5% |
| 22.08.2026 | 71.2% | 57.9% |
| 23.08.2026 | 72.3% | 57.8% |
| 24.08.2026 | 70.4% | 58.7% |
| 25.08.2026 | 71.6% | 57.0% |
| 26.08.2026 | 72.2% | 57.2% |
| 27.08.2026 | 72.3% | 56.6% |
| 28.08.2026 | 74.4% | 56.4% |
| 29.08.2026 | 75.9% | 53.2% |
| 30.08.2026 | 73.8% | 53.5% |
| 31.08.2026 | 74.7% | 51.6% |
Accuracy by competition in August 2026
The competitions with the most graded Best Picks this month, each with at least 20.
| Competition | Best Picks | Value Analysis |
|---|---|---|
 USA: Major League Soccer | 64.0%32 of 50 correct | 60.9%39 of 64 correct |
 England: League Cup | 86.4%38 of 44 correct | 54.2%13 of 24 correct |
 Argentina: Primera Nacional | 76.3%29 of 38 correct | 65.6%21 of 32 correct |
| Argentina: Torneo Federal A | 58.8%20 of 34 correct | 70.0%21 of 30 correct |
 World: UEFA Europa Conference League | 51.7%15 of 29 correct | 50.6%43 of 85 correct |
| Argentina: Liga Profesional Argentina | 71.4%20 of 28 correct | 62.7%42 of 67 correct |
 Chile: Primera División | 66.7%18 of 27 correct | 62.5%20 of 32 correct |
 Australia: New South Wales NPL | 76.0%19 of 25 correct | 65.0%13 of 20 correct |
| USA: USL Championship | 69.6%16 of 23 correct | 83.3%5 of 6 correct |
 Sweden: Allsvenskan | 59.1%13 of 22 correct | 46.7%14 of 30 correct |
Are the probabilities honest?
This is the test that matters. We group every match-result forecast by the probability we published, then plot what actually happened. A calibrated model sits on the diagonal.
What this chart coversMatch result across every graded fixture, not just selected picks - a selected subset could not test calibration honestly. Buckets holding too few fixtures to plot are merged into the end points, and point size scales with sample. Vertical bars are 95% intervals: a thin bucket is an uncertain reading, not a precise miss.
Brier score: 0.2007Mean squared error between published probability and outcome. Lower is better; 0.25 is what you would score by predicting 50% on everything.
Calibration by probability bin
| Probability band | Probability we published | How often it actually happened | 95% interval | Fixtures |
|---|---|---|---|---|
| 0.05-0.20 | 14.8% | 14.4% | 13.2% – 15.8% | 2815 |
| 0.20-0.25 | 22.8% | 21.6% | 20.3% – 23.0% | 3549 |
| 0.25-0.30 | 27.2% | 27.3% | 26.1% – 28.4% | 5663 |
| 0.30-0.35 | 32.3% | 32.6% | 30.8% – 34.4% | 2539 |
| 0.35-0.40 | 37.4% | 37.0% | 34.9% – 39.1% | 1997 |
| 0.40-0.45 | 42.4% | 41.4% | 39.3% – 43.6% | 1970 |
| 0.45-0.50 | 47.3% | 47.4% | 44.9% – 49.9% | 1507 |
| 0.50-0.55 | 52.4% | 53.9% | 50.9% – 57.0% | 1018 |
| 0.55-0.60 | 57.3% | 63.2% | 59.5% – 66.8% | 664 |
| 0.60-0.65 | 62.3% | 67.5% | 63.1% – 71.6% | 462 |
| 0.65-0.70 | 67.2% | 71.2% | 65.8% – 76.1% | 295 |
| 0.70-0.75 | 72.3% | 66.0% | 58.9% – 72.4% | 188 |
| 0.75-0.80 | 77.2% | 74.0% | 65.6% – 80.9% | 123 |
| 0.80-0.95 | 85.4% | 81.3% | 74.7% – 86.5% | 166 |
Interval is the 95% range the observed figure could take at that sample size. Narrower means more matches in the bin.
Which markets the picks came from
A hit rate across a mixed basket is not the same claim as a hit rate on match result alone. These are the six most-used markets in each basket, with the remainder grouped, so you can judge both headline figures for yourself.
Best Picks
- Home Win or Draw47.5% of picksn = 669
- Home Win14.3% of picksn = 202
- Draw or Away Win13.7% of picksn = 193
- Over 1.5 Goals12.4% of picksn = 175
- Away Win4.2% of picksn = 59
- Both Teams to Score4.2% of picksn = 59
- Other markets (3)3.6% of picksn = 52
Value Analysis
- Under 3.5 Goals19.5% of picksn = 401
- Under 2.5 Goals17.0% of picksn = 351
- Over 2.5 Goals9.1% of picksn = 188
- Over 1.5 Goals8.1% of picksn = 167
- Home Win or Draw6.6% of picksn = 136
- Home Win6.1% of picksn = 125
- Other markets (29)33.5% of picksn = 692
13 void matches were excluded from these figures.
Monthly reports
August 2026 accuracy report: the figures
In August 2026, 987 of 1409 Best Picks were correct: a hit rate of 70.0%. Value Analysis was right in 1177 of 2060 matches (57.1%). That is 2.2 points more than in July 2026.
On match result, the model was right in 50.3% of 7652 matches, against 43.5% for always picking the home team. Among competitions with at least 20 graded Best Picks, League Cup had the highest hit rate: 86.4% of 44.
Each Best Pick and Value Analysis is graded on the 90-minute result; nothing is removed once the outcome is known, and void matches are excluded. The all-time record and the method are on the Accuracy page.