Updated after racing on Mon 19 Oct 2026 · 31 forward races · 379 runners
When we rate a horse at 20%, does it win about 1 in 5? This page checks, and shows the market alongside us, good or bad.
How to read this
We sort runners into bands by our rated win %. If our percentages are honest, a band rated around 20% should win about 20% of the time, so the actual wins should be close to the expected wins (our percentages added up). The two scores boil it down to one number each: the lower, the better, and the market’s score is on the same races for comparison.
Forward · our locked ratings
31 races so far. These numbers will move a lot until we have a few hundred.
Every runner’s rating as locked 4 minutes before the off, from the 50/50 blend we use for Top Rated, scored against the result. The market here is the market’s own chance for each runner at the same moment (1 ÷ odds, adjusted so each race adds to 100%). We only started saving full fields on 11 Oct, so earlier meetings are in the “top pick only” table below.
| Our rated % | Runners | Expected wins | Actual wins | Rated (avg) | Won (actual) |
|---|---|---|---|---|---|
| 0–5%small sample | 165 | 5.0 | 4 | 3.0% | 2.4% |
| 5–10%small sample | 123 | 8.8 | 13 | 7.2% | 10.6% |
| 10–15%small sample | 35 | 4.4 | 3 | 12.5% | 8.6% |
| 15–20%small sample | 28 | 4.7 | 4 | 16.9% | 14.3% |
| 20–30%small sample | 19 | 4.6 | 4 | 24.0% | 21.1% |
| 30%+small sample | 9 | 3.5 | 3 | 38.8% | 33.3% |
| Market % | Runners | Expected wins | Actual wins | Market’s chance (avg) | Won (actual) |
|---|---|---|---|---|---|
| 0–5%small sample | 175 | 4.8 | 4 | 2.7% | 2.3% |
| 5–10%small sample | 101 | 7.2 | 7 | 7.2% | 6.9% |
| 10–15%small sample | 47 | 5.7 | 12 | 12.2% | 25.5% |
| 15–20%small sample | 25 | 4.1 | 0 | 16.3% | 0.0% |
| 20–30%small sample | 19 | 4.4 | 4 | 23.1% | 21.1% |
| 30%+small sample | 12 | 4.7 | 4 | 39.4% | 33.3% |
Earlier meetings · top pick only
Before 11 Oct we saved only each race’s top three, so for these races we can check the Top Rated pick and nothing else. Rated % is our rating for the pick; expected wins is those ratings added up.
| Rating used | Races | Average rated | Expected wins | Actual wins |
|---|---|---|---|---|
| Model only, before we switched to the 50/50 blend on 7 Oct | 22 | 26.9% | 5.9 | 2 |
| 50/50 blend of model and market (from 7 Oct) | 40 | 26.8% | 10.7 | 8 |
The model-only top picks were rated about 27% on average and won 2 of 22 races, where about 5.9 would have been expected. That small sample is one reason we moved Top Rated to the 50/50 blend on 7 Oct. Small samples mislead in both directions, so we are not drawing more from it than that.
Test · 854 races we never trained on
854 races from 7 Sep 2025 to 15 Jul 2026, 10,687 runners. Our rating is the same 50/50 blend; the market is its final prices. That is not the price we can see 4 minutes before the off, so treat the market score as the market at its sharpest.
| Our rated % | Runners | Expected wins | Actual wins | Rated (avg) | Won (actual) |
|---|---|---|---|---|---|
| 0–5% | 4,694 | 126.8 | 103 | 2.7% | 2.2% |
| 5–10% | 3,104 | 222.9 | 215 | 7.2% | 6.9% |
| 10–15% | 1,450 | 177.6 | 192 | 12.3% | 13.2% |
| 15–20% | 674 | 116.4 | 124 | 17.3% | 18.4% |
| 20–30% | 556 | 133.7 | 138 | 24.0% | 24.8% |
| 30%+ | 209 | 76.5 | 82 | 36.6% | 39.2% |
| Market % | Runners | Expected wins | Actual wins | Market’s chance (avg) | Won (actual) |
|---|---|---|---|---|---|
| 0–5% | 5,104 | 113.9 | 102 | 2.2% | 2.0% |
| 5–10% | 2,712 | 196.1 | 179 | 7.2% | 6.6% |
| 10–15% | 1,244 | 153.2 | 169 | 12.3% | 13.6% |
| 15–20% | 672 | 116.0 | 129 | 17.3% | 19.2% |
| 20–30% | 648 | 155.2 | 160 | 23.9% | 24.7% |
| 30%+ | 307 | 119.5 | 115 | 38.9% | 37.5% |
Read this before drawing conclusions
Ratings, not tips. Updated automatically after each meeting’s results are in.