Every month we publish how closely the TACTIX model’s probabilities matched what actually happened, bucket by bucket, and how they compared with the bookmakers’ closing prices. Good months and bad months are published the same way. This is August 2026: 1,090,874 settled predictions with a kick-off in the month, from the rule-based-v2 model.
Calibration
When the model gives an outcome a 60% chance, it should happen about 60% of the time. In August 2026 the model’s stated probabilities and the observed rates differed by 0.2 percentage points on average, weighted by the number of predictions in each bucket, across those 1,090,874 predictions.
| Probability bucket | Average predicted | Happened | Predictions |
|---|---|---|---|
| 0–10% | 4.4% | 4.7% | 238,571 |
| 10–20% | 14.8% | 14.8% | 102,305 |
| 20–30% | 24.9% | 25.1% | 96,166 |
| 30–40% | 35.1% | 35.3% | 88,819 |
| 40–50% | 45.0% | 44.9% | 93,203 |
| 50–60% | 55.0% | 55.4% | 87,322 |
| 60–70% | 64.9% | 64.9% | 78,078 |
| 70–80% | 75.0% | 74.6% | 82,393 |
| 80–90% | 85.2% | 85.5% | 78,057 |
| 90–100% | 95.4% | 95.0% | 145,960 |
Against the Bookmakers’ Closing Prices
For match-result, over/under 2.5 goals and both-teams-to-score predictions, we compare each probability with the bookmakers’ closing price for the same outcome, with the margin removed. A closing price counts only when it was captured within 6 hours of kick-off. In August 2026 a closing price was available for 11,375 of 32,623 of those predictions.
In August 2026 the bookmakers’ closing prices were more accurate than the model (log-loss difference +0.025, 95% interval +0.020 to +0.032 across 31 match days). Across those 11,375 paired predictions the model’s Brier score was 0.228 and the closing prices’ was 0.217 (lower is better for both).
| Market | With a closing price | Model Brier | Closing-price Brier | Result |
|---|---|---|---|---|
| Match result (1X2) | 5,139 of 5,757 | 0.204 | 0.193 | closing prices more accurate |
| Over/under 2.5 goals | 3,288 of 23,028 | 0.247 | 0.232 | closing prices more accurate |
| Both teams to score | 2,948 of 3,838 | 0.247 | 0.241 | closing prices more accurate |
On the 1,713 matches with a complete closing 1X2 price, the ranked probability score was 0.213 for the model and 0.199 for the closing prices (lower is better).
How This Is Measured
- A prediction belongs to the month its match kicked off in, not the month it was settled.
- Every settled prediction counts, winners and losers alike. A month with fewer than 500 settled predictions is not published.
- Calibration buckets are 10 points wide; the average gap is weighted by the predictions in each bucket.
- The closing price is the last complete bookmaker price at or before kick-off (Pinnacle first, then Betfair, then the median of complete books), with the margin removed. Older prices are not used.
- The interval on the difference is a 95% bootstrap interval that resamples whole match days, so one busy Saturday cannot decide the result.
The live, all-time picture is on How accurate is TACTIX? and the Model Scoreboard.
18+. TACTIX scorecards are model-informed analysis — not betting advice or guarantees. Please gamble responsibly — support at BeGambleAware.org.
This post is generated from TACTIX’s match data by our software and checked automatically. How it works →
Get today’s free slate
Free, every day: the model’s fixed slate of picks, each with its confidence tier and that tier’s settled record — plus yesterday’s slate, graded. Pro adds the full board, the value edge and the written analysis. No payment card.
Create free account