NilsBriggen
88afbfad74
The CV harness gained a calibration experiment: per-outcome isotonic maps (pool-adjacent-violators) fit walk-forward on the five folds and judged leave-one-fold-out against a pre-registered bar. It cleared it: LOFO ECE improves 16% (0.0146 → 0.0122) with RPS slightly better too (0.1724 → 0.1722). The final maps (fit on all folds) are committed as a research artifact, embedded into ratings.json by buildRatings when the shipped config matches, and applied in matchMatrix by scaling the score-matrix outcome regions — scorelines, Monte Carlo and displayed probabilities all stay mutually consistent. Without the field, output is bit-identical (golden tests). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
70 lines
1.3 KiB
JSON
70 lines
1.3 KiB
JSON
{
|
|
"generatedAt": "2026-06-11T23:06:22.047Z",
|
|
"protocol": {
|
|
"folds": [
|
|
{
|
|
"from": "2014-01-01",
|
|
"to": "2016-01-01"
|
|
},
|
|
{
|
|
"from": "2016-01-01",
|
|
"to": "2018-01-01"
|
|
},
|
|
{
|
|
"from": "2018-01-01",
|
|
"to": "2020-01-01"
|
|
},
|
|
{
|
|
"from": "2020-01-01",
|
|
"to": "2022-01-01"
|
|
},
|
|
{
|
|
"from": "2022-01-01",
|
|
"to": "2024-01-01"
|
|
}
|
|
],
|
|
"test": {
|
|
"from": "2024-01-01",
|
|
"to": "2026-06-01"
|
|
},
|
|
"paramYears": 12,
|
|
"minGain": 0.0005,
|
|
"selection": "mean fold RPS, staged greedy"
|
|
},
|
|
"shipped": {
|
|
"xi": 0.25,
|
|
"k": 3,
|
|
"w": 0.45,
|
|
"m": 3,
|
|
"beta": 0.3,
|
|
"gamma": 0,
|
|
"homeAdv": 70,
|
|
"kScale": 1,
|
|
"meanFoldRps": 0.1724,
|
|
"testRps": 0.1649
|
|
},
|
|
"champion": {
|
|
"xi": 0.25,
|
|
"k": 3,
|
|
"w": 0.4,
|
|
"m": 3,
|
|
"beta": 0.4,
|
|
"gamma": 0,
|
|
"homeAdv": 90,
|
|
"kScale": 1.25,
|
|
"meanFoldRps": 0.1721,
|
|
"testRps": 0.1647
|
|
},
|
|
"shipChange": false,
|
|
"isotonic": {
|
|
"bar": "LOFO RPS worsens ≤0.0002 AND LOFO ECE improves ≥10%",
|
|
"lofo": {
|
|
"rawRps": 0.1724,
|
|
"calRps": 0.1722,
|
|
"rawEce": 0.0146,
|
|
"calEce": 0.0122
|
|
},
|
|
"ship": true
|
|
},
|
|
"runtimeSec": 73
|
|
} |