matrix holdout: full rule grid, train/test, all leagues
← lab · n survivors 6 · predictions 219,577 · cells scanned 167 · ran 7/9/2026
What this is: A one-off study — its numbers below are exactly what the run reported.
| note | 1x2, sim-engine predictions, opening price, train<2022-07-01/test>=. Survive = +EV in BOTH halves, >=100 train & >=50 test bets. Multiple testing: ~a few false positives expected among the survivors; the train+test double-filter is the guard. Re-run after new odds land. |
| n survivors | 6 |
| predictions | 219,577 |
| cells scanned | 167 |
survivors
| sel | band | league | test n | train n | edge min | test roi |
|---|
| home side win | middog | EPL | 62 | 191 | 0.05 | 41.1 |
| home side win | middog | EPL | 78 | 223 | 0 | 22.6 |
| away side win | mid | EPL | 54 | 282 | 0.08 | 10.6 |
| away side win | mid | PPL | 81 | 290 | 0 | 6.5 |
| home side win | priced 1.00–1.80 | PPL | 90 | 600 | 0 | 4.7 |
| away side win | priced 1.00–1.80 | PPL | 63 | 252 | 0 | 3.5 |
Reading the columnswhat each number actually means
| AUC | predictability: 0.50 = coin flip, ~0.70 = ceiling for sports |
| Importance | how much the model leans on this factor (permutation importance) |
| Direction | sign of the raw correlation with the outcome |
| Survives all eras | effect points the same way in every historical era |
Spec · the reproducible recipe
{
"kind": "matrix_holdout",
"bands": [
"odds_on",
"even",
"mid",
"middog",
"long",
"xlong"
],
"edges": [
0,
0.05,
0.08,
0.1
]
}