Reliability
A confidence number is only worth reading next to the record of what that confidence has been worth.
1,403,401 graded selectionswalk-forwardisotonic calibration
Graded selections
1,403,401
out of sample
Worst band gap
0.65pp
predicted vs actual
Bands
14
30% up to 100%
Does an 80% pick win 80% of the time?
Every graded selection, bucketed by what it claimed. Each band's interval is Wilson rather than the normal approximation, because the bands that matter most sit near 95% where a normal interval runs past 1.0 and stops meaning anything.
| Band | Bets | Said | Did | Gap | 95% interval |
|---|---|---|---|---|---|
| 30%-40% | 207,755 | 34.7% | 34.7% | +0.04pp | 34.5% – 35.0% |
| 40%-45% | 105,243 | 42.5% | 42.6% | +0.01pp | 42.3% – 42.9% |
| 45%-50% | 112,145 | 47.5% | 47.7% | +0.18pp | 47.4% – 48.0% |
| 50%-60% | 216,624 | 54.9% | 54.8% | -0.09pp | 54.6% – 55.0% |
| 60%-70% | 208,769 | 65.3% | 65.2% | -0.04pp | 65.0% – 65.4% |
| 70%-75% | 126,044 | 72.6% | 72.5% | -0.02pp | 72.3% – 72.8% |
| 75%-80% | 109,940 | 77.1% | 77.1% | +0.01pp | 76.9% – 77.4% |
| 80%-85% | 104,926 | 82.4% | 82.2% | -0.15pp | 82.0% – 82.4% |
| 85%-90% | 86,663 | 87.6% | 87.6% | -0.08pp | 87.3% – 87.8% |
| 90%-95% | 97,557 | 92.5% | 92.4% | -0.07pp | 92.3% – 92.6% |
| 95%-97% | 15,948 | 96.0% | 95.8% | -0.18pp | 95.5% – 96.1% |
| 97%-98% | 6,794 | 97.4% | 97.2% | -0.22pp | 96.8% – 97.6% |
| 98%-99% | 2,881 | 98.5% | 97.8% | -0.65pp | 97.2% – 98.3% |
| 99%-100% | 2,112 | 99.5% | 99.1% | -0.35pp | 98.7% – 99.5% |
Where the numbers stop being checked
Each market is trusted only as far up as its own record supports, and the walk up its bands stops for one of two reasons: the band above overstated itself, or too few bets that high have been graded to say. Right now the first applies to Corners. Picks above the ceiling are dropped from the card.
| Market | Checked up to | Why it stops here |
|---|---|---|
| Both teams to score | 70% | too few bets above it to check |
| Corners | 85% | overstated itself in the band above |
| 1X2 | 90% | too few bets above it to check |
| Draw no bet | 97% | too few bets above it to check |
| Double chance | 98% | too few bets above it to check |
| Handicap | no ceiling | tested to the top |
| Total goals | no ceiling | tested to the top |
| Team goals | no ceiling | tested to the top |
By market
The pooled table can hide a market that fails on its own. Pick one.
| Band | Bets | Said | Did | Gap | 95% interval |
|---|