v5 model protocol: rolling-origin CV search harness

Five two-year walk-forward folds (2014-2024), each with its own pre-fold parameter fit; selection by mean fold RPS; staged greedy over Team-DC decay/shrinkage/weight, fast-Elo blend, form multiplier, and new Elo-walk axes (home advantage, K scale); untouched 2024-26 final test. Result: the shipped v4 config holds — the staged champion gained 0.0003, under the 0.0005 ship bar, and the final test agrees. v4 is no longer a one-window result. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-06-12 00:41:37 +02:00
parent 2c945dfd54
commit acd8c3e75d
2 changed files with 441 additions and 0 deletions
@@ -0,0 +1,60 @@
+{
+  "generatedAt": "2026-06-11T22:41:08.860Z",
+  "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,
+  "runtimeSec": 72
+}