启用概率校准和 WebSocket 报价配置

.env.example

@@ -177,6 +177,11 @@ POLYMARKET_CHAIN_ID=137
 POLYMARKET_HTTP_TIMEOUT_SEC=8
 POLYMARKET_MARKET_CACHE_TTL_SEC=180
 POLYMARKET_PRICE_CACHE_TTL_SEC=10
+POLYMARKET_WS_PRICE_ENABLED=false
+POLYMARKET_WS_MARKET_URL=wss://ws-subscriptions-clob.polymarket.com/ws/market
+POLYMARKET_WS_QUOTE_TTL_SEC=8
+POLYMARKET_WS_MAX_ASSETS=256
+POLYMARKET_WS_RECONNECT_DELAY_SEC=3
 POLYMARKET_DISCOVERY_PAGES=6
 POLYMARKET_DISCOVERY_LIMIT=200
 POLYMARKET_SIGNAL_MIN_LIQUIDITY=500

artifacts/local_runtime/probability_calibration/auto_retrain_report.json

@@ -1,14 +1,14 @@
 {
   "ok": true,
-  "version": "emos-auto-20260418212046",
+  "version": "emos-auto-20260421122743",
-  "generated_at": "2026-04-18T21:21:04.160748+00:00",
+  "generated_at": "2026-04-21T12:28:08.021388+00:00",
-  "candidate_dir": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260418212046",
+  "candidate_dir": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743",
-  "candidate_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260418212046\\default.json",
+  "candidate_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json",
-  "evaluation_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260418212046\\evaluation_report.json",
+  "evaluation_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\evaluation_report.json",
   "target_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\default.json",
-  "promote_requested": false,
+  "promote_requested": true,
-  "promoted": false,
+  "promoted": true,
-  "backup_path": null,
+  "backup_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\default.backup-20260421122808.json",
   "decision": {
     "decision": "promote",
     "ready_for_promotion": true,
@@ -20,10 +20,10 @@
       "min_delta_bucket_hit_rate": -0.05
     },
     "metrics": {
-      "sample_count": 847,
+      "sample_count": 978,
-      "delta_crps": -0.03617,
+      "delta_crps": -0.05721,
-      "delta_mae": -0.007896,
+      "delta_mae": -0.034568,
-      "delta_bucket_hit_rate": -0.009445
+      "delta_bucket_hit_rate": -0.00409
     }
   },
   "fit": {
@@ -31,9 +31,9 @@
       "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
       "scripts/fit_probability_calibration.py",
       "--output",
-      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260418212046\\default.json",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json",
       "--version",
-      "emos-auto-20260418212046",
+      "emos-auto-20260421122743",
       "--verbose",
       "--snapshot-limit",
       "50000"
@@ -47,9 +47,9 @@
       "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
       "scripts/evaluate_probability_calibration.py",
       "--calibration-file",
-      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260418212046\\default.json",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json",
       "--output",
-      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260418212046\\evaluation_report.json",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\evaluation_report.json",
       "--verbose",
       "--snapshot-limit",
       "50000"

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421115828/decision_report.json

@@ -0,0 +1,62 @@
+{
+  "ok": true,
+  "version": "emos-auto-20260421115828",
+  "generated_at": "2026-04-21T11:58:32.490537+00:00",
+  "candidate_dir": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421115828",
+  "candidate_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421115828\\default.json",
+  "evaluation_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421115828\\evaluation_report.json",
+  "target_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\default.json",
+  "promote_requested": false,
+  "promoted": false,
+  "backup_path": null,
+  "decision": {
+    "decision": "promote",
+    "ready_for_promotion": true,
+    "blocking_reasons": [],
+    "thresholds": {
+      "min_samples": 50,
+      "max_delta_crps": 0.0,
+      "max_delta_mae": 0.05,
+      "min_delta_bucket_hit_rate": -0.05
+    },
+    "metrics": {
+      "sample_count": 74,
+      "delta_crps": -0.142868,
+      "delta_mae": -0.056741,
+      "delta_bucket_hit_rate": 0.0
+    }
+  },
+  "fit": {
+    "command": [
+      "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
+      "scripts/fit_probability_calibration.py",
+      "--output",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421115828\\default.json",
+      "--version",
+      "emos-auto-20260421115828",
+      "--verbose",
+      "--snapshot-limit",
+      "50000"
+    ],
+    "returncode": 0,
+    "stdout": "",
+    "stderr": ""
+  },
+  "evaluate": {
+    "command": [
+      "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
+      "scripts/evaluate_probability_calibration.py",
+      "--calibration-file",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421115828\\default.json",
+      "--output",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421115828\\evaluation_report.json",
+      "--verbose",
+      "--snapshot-limit",
+      "50000"
+    ],
+    "returncode": 0,
+    "stdout": "",
+    "stderr": ""
+  },
+  "tests": null
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421115828/default.json

@@ -0,0 +1,115 @@
+{
+  "version": "emos-auto-20260421115828",
+  "trained_at": "2026-04-21T11:58:31.424860+00:00",
+  "global": {
+    "mu": {
+      "intercept": 2.60074353,
+      "raw_mu_coef": 1.97765258,
+      "deb_coef": -1.37457626,
+      "ens_median_coef": 0.10693472,
+      "max_so_far_gap_coef": -0.58256593
+    },
+    "sigma": {
+      "intercept": 0.6373609,
+      "raw_sigma_coef": -1.41537225,
+      "spread_coef": 2.12997577,
+      "peak_flag_coef": -0.08633203,
+      "max_so_far_gap_coef": -0.00716383
+    }
+  },
+  "sigma_constraints": {
+    "min_ratio": 0.85,
+    "max_ratio": 1.35,
+    "absolute_min": 0.25,
+    "absolute_max": 3.0
+  },
+  "selection_guardrails": {
+    "max_mae_increase": 0.02,
+    "max_bucket_hit_drop": 0.01,
+    "max_bucket_brier_increase": 0.05
+  },
+  "blending": {
+    "alpha_mu": 0.05,
+    "alpha_sigma": 0.7
+  },
+  "cities": {
+    "tokyo": {
+      "samples": 5,
+      "mu_bias": 2.101101,
+      "sigma_scale": 2.0,
+      "confidence": 0.625
+    },
+    "ankara": {
+      "samples": 6,
+      "mu_bias": -0.523145,
+      "sigma_scale": 1.139849,
+      "confidence": 0.75
+    },
+    "istanbul": {
+      "samples": 3,
+      "mu_bias": -0.285183,
+      "sigma_scale": 0.550171,
+      "confidence": 0.375
+    },
+    "hong kong": {
+      "samples": 7,
+      "mu_bias": 3.234676,
+      "sigma_scale": 2.0,
+      "confidence": 0.875
+    },
+    "taipei": {
+      "samples": 5,
+      "mu_bias": 4.731246,
+      "sigma_scale": 2.0,
+      "confidence": 0.625
+    },
+    "busan": {
+      "samples": 3,
+      "mu_bias": 0.177772,
+      "sigma_scale": 2.0,
+      "confidence": 0.375
+    },
+    "seoul": {
+      "samples": 3,
+      "mu_bias": -3.070325,
+      "sigma_scale": 1.462196,
+      "confidence": 0.375
+    },
+    "milan": {
+      "samples": 3,
+      "mu_bias": -4.249866,
+      "sigma_scale": 2.0,
+      "confidence": 0.375
+    },
+    "shanghai": {
+      "samples": 3,
+      "mu_bias": 1.080039,
+      "sigma_scale": 1.638119,
+      "confidence": 0.375
+    },
+    "warsaw": {
+      "samples": 3,
+      "mu_bias": -1.68611,
+      "sigma_scale": 1.38114,
+      "confidence": 0.375
+    }
+  },
+  "metrics": {
+    "sample_count": 74,
+    "mean_crps": 3.181501,
+    "legacy_mean_crps": 3.474108,
+    "legacy_mean_mae": 3.679324,
+    "legacy_bucket_hit_rate": 0.5,
+    "legacy_bucket_brier": 0.855395,
+    "selected_mean_crps": 3.322469,
+    "selected_mean_mae": 3.618254,
+    "selected_bucket_hit_rate": 0.5,
+    "selected_bucket_brier": 0.863048,
+    "selected_score": 5.08158,
+    "legacy_score": 5.233586,
+    "filled_actual_from_history": 0,
+    "settlement_history_city_count": 30,
+    "legacy_archive_samples": 54
+  },
+  "source": "artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421115828\\default.json"
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421115828/evaluation_report.json

@@ -0,0 +1,293 @@
+{
+  "summary": {
+    "sample_count": 74,
+    "filled_actual_from_history": 0,
+    "legacy": {
+      "mean_crps": 3.474108,
+      "mean_mae": 3.679324,
+      "bucket_hit_rate": 0.5
+    },
+    "emos": {
+      "mean_crps": 3.33124,
+      "mean_mae": 3.622584,
+      "bucket_hit_rate": 0.5
+    },
+    "delta": {
+      "crps": -0.142868,
+      "mae": -0.056741,
+      "bucket_hit_rate": 0.0
+    }
+  },
+  "by_city": {
+    "ankara": {
+      "samples": 6,
+      "legacy_mean_crps": 1.365135,
+      "emos_mean_crps": 1.392829,
+      "legacy_mean_mae": 1.466667,
+      "emos_mean_mae": 1.444819,
+      "legacy_bucket_hit_rate": 0.666667,
+      "emos_bucket_hit_rate": 0.666667
+    },
+    "atlanta": {
+      "samples": 2,
+      "legacy_mean_crps": 30.449382,
+      "emos_mean_crps": 29.886236,
+      "legacy_mean_mae": 32.015,
+      "emos_mean_mae": 31.322804,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "buenos aires": {
+      "samples": 2,
+      "legacy_mean_crps": 9.113412,
+      "emos_mean_crps": 8.463522,
+      "legacy_mean_mae": 10.27,
+      "emos_mean_mae": 10.024533,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "busan": {
+      "samples": 3,
+      "legacy_mean_crps": 0.276202,
+      "emos_mean_crps": 0.326485,
+      "legacy_mean_mae": 0.3,
+      "emos_mean_mae": 0.239636,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "chengdu": {
+      "samples": 1,
+      "legacy_mean_crps": 0.306469,
+      "emos_mean_crps": 0.378812,
+      "legacy_mean_mae": 0.3,
+      "emos_mean_mae": 0.153224,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "chicago": {
+      "samples": 1,
+      "legacy_mean_crps": 1.250268,
+      "emos_mean_crps": 0.714669,
+      "legacy_mean_mae": 0.0,
+      "emos_mean_mae": 0.319756,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "dallas": {
+      "samples": 1,
+      "legacy_mean_crps": 2.173363,
+      "emos_mean_crps": 0.840297,
+      "legacy_mean_mae": 0.0,
+      "emos_mean_mae": 1.028146,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "hong kong": {
+      "samples": 7,
+      "legacy_mean_crps": 0.706099,
+      "emos_mean_crps": 0.731926,
+      "legacy_mean_mae": 0.657143,
+      "emos_mean_mae": 0.669911,
+      "legacy_bucket_hit_rate": 0.714286,
+      "emos_bucket_hit_rate": 0.571429
+    },
+    "istanbul": {
+      "samples": 3,
+      "legacy_mean_crps": 1.264103,
+      "emos_mean_crps": 1.214967,
+      "legacy_mean_mae": 1.5,
+      "emos_mean_mae": 1.469837,
+      "legacy_bucket_hit_rate": 0.333333,
+      "emos_bucket_hit_rate": 0.666667
+    },
+    "london": {
+      "samples": 2,
+      "legacy_mean_crps": 3.885033,
+      "emos_mean_crps": 3.799841,
+      "legacy_mean_mae": 4.135,
+      "emos_mean_mae": 4.027503,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "lucknow": {
+      "samples": 2,
+      "legacy_mean_crps": 2.487193,
+      "emos_mean_crps": 2.107441,
+      "legacy_mean_mae": 3.205,
+      "emos_mean_mae": 2.876943,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "madrid": {
+      "samples": 2,
+      "legacy_mean_crps": 6.27726,
+      "emos_mean_crps": 5.810228,
+      "legacy_mean_mae": 7.33,
+      "emos_mean_mae": 7.195133,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "miami": {
+      "samples": 2,
+      "legacy_mean_crps": 28.637631,
+      "emos_mean_crps": 27.592664,
+      "legacy_mean_mae": 30.175,
+      "emos_mean_mae": 29.284979,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "milan": {
+      "samples": 3,
+      "legacy_mean_crps": 4.401392,
+      "emos_mean_crps": 3.782613,
+      "legacy_mean_mae": 4.06,
+      "emos_mean_mae": 3.989808,
+      "legacy_bucket_hit_rate": 0.666667,
+      "emos_bucket_hit_rate": 0.666667
+    },
+    "munich": {
+      "samples": 2,
+      "legacy_mean_crps": 3.145192,
+      "emos_mean_crps": 2.990712,
+      "legacy_mean_mae": 3.64,
+      "emos_mean_mae": 3.61502,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "new york": {
+      "samples": 1,
+      "legacy_mean_crps": 3.692845,
+      "emos_mean_crps": 3.021798,
+      "legacy_mean_mae": 4.94,
+      "emos_mean_mae": 4.54437,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "paris": {
+      "samples": 2,
+      "legacy_mean_crps": 4.013782,
+      "emos_mean_crps": 3.915303,
+      "legacy_mean_mae": 4.265,
+      "emos_mean_mae": 4.179694,
+      "legacy_bucket_hit_rate": 0.5,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "sao paulo": {
+      "samples": 2,
+      "legacy_mean_crps": 5.540967,
+      "emos_mean_crps": 5.018605,
+      "legacy_mean_mae": 6.57,
+      "emos_mean_mae": 6.314389,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "seattle": {
+      "samples": 1,
+      "legacy_mean_crps": 0.315488,
+      "emos_mean_crps": 0.524119,
+      "legacy_mean_mae": 0.0,
+      "emos_mean_mae": 0.673609,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "seoul": {
+      "samples": 3,
+      "legacy_mean_crps": 0.508331,
+      "emos_mean_crps": 0.520762,
+      "legacy_mean_mae": 0.2,
+      "emos_mean_mae": 0.247862,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "shanghai": {
+      "samples": 3,
+      "legacy_mean_crps": 0.299116,
+      "emos_mean_crps": 0.402453,
+      "legacy_mean_mae": 0.1,
+      "emos_mean_mae": 0.189437,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "shenzhen": {
+      "samples": 1,
+      "legacy_mean_crps": 1.198351,
+      "emos_mean_crps": 1.490963,
+      "legacy_mean_mae": 1.3,
+      "emos_mean_mae": 1.628189,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "singapore": {
+      "samples": 2,
+      "legacy_mean_crps": 0.281993,
+      "emos_mean_crps": 0.370374,
+      "legacy_mean_mae": 0.15,
+      "emos_mean_mae": 0.118005,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "taipei": {
+      "samples": 5,
+      "legacy_mean_crps": 0.950739,
+      "emos_mean_crps": 1.042332,
+      "legacy_mean_mae": 0.94,
+      "emos_mean_mae": 0.951241,
+      "legacy_bucket_hit_rate": 0.6,
+      "emos_bucket_hit_rate": 0.6
+    },
+    "tel aviv": {
+      "samples": 2,
+      "legacy_mean_crps": 0.446758,
+      "emos_mean_crps": 0.578691,
+      "legacy_mean_mae": 0.3,
+      "emos_mean_mae": 0.116966,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "tokyo": {
+      "samples": 5,
+      "legacy_mean_crps": 0.879366,
+      "emos_mean_crps": 0.876608,
+      "legacy_mean_mae": 1.022,
+      "emos_mean_mae": 1.008317,
+      "legacy_bucket_hit_rate": 0.2,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "toronto": {
+      "samples": 2,
+      "legacy_mean_crps": 5.497916,
+      "emos_mean_crps": 5.268783,
+      "legacy_mean_mae": 6.33,
+      "emos_mean_mae": 6.388522,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "warsaw": {
+      "samples": 3,
+      "legacy_mean_crps": 1.618875,
+      "emos_mean_crps": 1.524219,
+      "legacy_mean_mae": 2.056667,
+      "emos_mean_mae": 2.006524,
+      "legacy_bucket_hit_rate": 0.333333,
+      "emos_bucket_hit_rate": 0.333333
+    },
+    "wellington": {
+      "samples": 2,
+      "legacy_mean_crps": 0.364919,
+      "emos_mean_crps": 0.484124,
+      "legacy_mean_mae": 0.15,
+      "emos_mean_mae": 0.123335,
+      "legacy_bucket_hit_rate": 1.0,
+      "emos_bucket_hit_rate": 1.0
+    },
+    "wuhan": {
+      "samples": 1,
+      "legacy_mean_crps": 0.476225,
+      "emos_mean_crps": 0.578405,
+      "legacy_mean_mae": 0.8,
+      "emos_mean_mae": 0.962852,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    }
+  }
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421121159/decision_report.json

@@ -0,0 +1,21 @@
+{
+  "ok": false,
+  "version": "emos-auto-20260421121159",
+  "stage": "fit",
+  "fit": {
+    "command": [
+      "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
+      "scripts/fit_probability_calibration.py",
+      "--output",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121159\\default.json",
+      "--version",
+      "emos-auto-20260421121159",
+      "--verbose",
+      "--snapshot-limit",
+      "50000"
+    ],
+    "returncode": 1,
+    "stdout": "",
+    "stderr": ""
+  }
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421121920/decision_report.json

@@ -0,0 +1,62 @@
+{
+  "ok": true,
+  "version": "emos-auto-20260421121920",
+  "generated_at": "2026-04-21T12:19:41.596944+00:00",
+  "candidate_dir": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121920",
+  "candidate_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121920\\default.json",
+  "evaluation_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121920\\evaluation_report.json",
+  "target_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\default.json",
+  "promote_requested": false,
+  "promoted": false,
+  "backup_path": null,
+  "decision": {
+    "decision": "promote",
+    "ready_for_promotion": true,
+    "blocking_reasons": [],
+    "thresholds": {
+      "min_samples": 50,
+      "max_delta_crps": 0.0,
+      "max_delta_mae": 0.05,
+      "min_delta_bucket_hit_rate": -0.05
+    },
+    "metrics": {
+      "sample_count": 978,
+      "delta_crps": -0.05721,
+      "delta_mae": -0.034568,
+      "delta_bucket_hit_rate": -0.00409
+    }
+  },
+  "fit": {
+    "command": [
+      "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
+      "scripts/fit_probability_calibration.py",
+      "--output",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121920\\default.json",
+      "--version",
+      "emos-auto-20260421121920",
+      "--verbose",
+      "--snapshot-limit",
+      "50000"
+    ],
+    "returncode": 0,
+    "stdout": "",
+    "stderr": ""
+  },
+  "evaluate": {
+    "command": [
+      "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
+      "scripts/evaluate_probability_calibration.py",
+      "--calibration-file",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121920\\default.json",
+      "--output",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121920\\evaluation_report.json",
+      "--verbose",
+      "--snapshot-limit",
+      "50000"
+    ],
+    "returncode": 0,
+    "stdout": "",
+    "stderr": ""
+  },
+  "tests": null
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421121920/default.json

@@ -0,0 +1,361 @@
+{
+  "version": "emos-auto-20260421121920",
+  "trained_at": "2026-04-21T12:19:38.887631+00:00",
+  "global": {
+    "mu": {
+      "intercept": 0.34330438,
+      "raw_mu_coef": 0.48749629,
+      "deb_coef": 0.45733479,
+      "ens_median_coef": 0.05955297,
+      "max_so_far_gap_coef": -0.104935
+    },
+    "sigma": {
+      "intercept": -0.62765582,
+      "raw_sigma_coef": 0.38492261,
+      "spread_coef": 0.29095921,
+      "peak_flag_coef": 0.33966327,
+      "max_so_far_gap_coef": 0.08357376
+    }
+  },
+  "sigma_constraints": {
+    "min_ratio": 0.85,
+    "max_ratio": 1.35,
+    "absolute_min": 0.25,
+    "absolute_max": 3.0
+  },
+  "selection_guardrails": {
+    "max_mae_increase": 0.02,
+    "max_bucket_hit_drop": 0.01,
+    "max_bucket_brier_increase": 0.05
+  },
+  "blending": {
+    "alpha_mu": 0.25,
+    "alpha_sigma": 1.0
+  },
+  "cities": {
+    "istanbul": {
+      "samples": 20,
+      "mu_bias": -0.438949,
+      "sigma_scale": 0.553793,
+      "confidence": 1.0
+    },
+    "buenos aires": {
+      "samples": 22,
+      "mu_bias": -1.044971,
+      "sigma_scale": 1.389753,
+      "confidence": 1.0
+    },
+    "ankara": {
+      "samples": 23,
+      "mu_bias": 0.067406,
+      "sigma_scale": 1.270298,
+      "confidence": 1.0
+    },
+    "london": {
+      "samples": 22,
+      "mu_bias": -0.010833,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "paris": {
+      "samples": 22,
+      "mu_bias": -0.21677,
+      "sigma_scale": 1.095669,
+      "confidence": 1.0
+    },
+    "new york": {
+      "samples": 21,
+      "mu_bias": 1.072189,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "lagos": {
+      "samples": 10,
+      "mu_bias": -0.092732,
+      "sigma_scale": 1.558646,
+      "confidence": 1.0
+    },
+    "madrid": {
+      "samples": 22,
+      "mu_bias": -0.627862,
+      "sigma_scale": 1.761239,
+      "confidence": 1.0
+    },
+    "toronto": {
+      "samples": 22,
+      "mu_bias": -0.424101,
+      "sigma_scale": 1.326675,
+      "confidence": 1.0
+    },
+    "wellington": {
+      "samples": 22,
+      "mu_bias": -0.347796,
+      "sigma_scale": 0.893459,
+      "confidence": 1.0
+    },
+    "los angeles": {
+      "samples": 20,
+      "mu_bias": -0.659983,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "san francisco": {
+      "samples": 20,
+      "mu_bias": 0.941818,
+      "sigma_scale": 1.979403,
+      "confidence": 1.0
+    },
+    "aurora": {
+      "samples": 20,
+      "mu_bias": 2.040009,
+      "sigma_scale": 0.982492,
+      "confidence": 1.0
+    },
+    "austin": {
+      "samples": 20,
+      "mu_bias": 0.990023,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "houston": {
+      "samples": 20,
+      "mu_bias": 0.188787,
+      "sigma_scale": 1.819594,
+      "confidence": 1.0
+    },
+    "chicago": {
+      "samples": 21,
+      "mu_bias": 2.285098,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "dallas": {
+      "samples": 21,
+      "mu_bias": 0.258095,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "atlanta": {
+      "samples": 22,
+      "mu_bias": -2.093105,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "seattle": {
+      "samples": 21,
+      "mu_bias": -0.982188,
+      "sigma_scale": 1.695172,
+      "confidence": 1.0
+    },
+    "sao paulo": {
+      "samples": 22,
+      "mu_bias": -0.786242,
+      "sigma_scale": 1.427037,
+      "confidence": 1.0
+    },
+    "munich": {
+      "samples": 22,
+      "mu_bias": 0.096559,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "milan": {
+      "samples": 23,
+      "mu_bias": -0.045038,
+      "sigma_scale": 1.626331,
+      "confidence": 1.0
+    },
+    "warsaw": {
+      "samples": 23,
+      "mu_bias": -0.50853,
+      "sigma_scale": 1.239146,
+      "confidence": 1.0
+    },
+    "helsinki": {
+      "samples": 17,
+      "mu_bias": 0.442247,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "amsterdam": {
+      "samples": 17,
+      "mu_bias": 0.129889,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "panama city": {
+      "samples": 17,
+      "mu_bias": -0.040792,
+      "sigma_scale": 1.393259,
+      "confidence": 1.0
+    },
+    "cape town": {
+      "samples": 10,
+      "mu_bias": 0.507293,
+      "sigma_scale": 1.420004,
+      "confidence": 1.0
+    },
+    "miami": {
+      "samples": 22,
+      "mu_bias": -2.736316,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "manila": {
+      "samples": 5,
+      "mu_bias": 1.772229,
+      "sigma_scale": 2.0,
+      "confidence": 0.625
+    },
+    "singapore": {
+      "samples": 22,
+      "mu_bias": 0.289639,
+      "sigma_scale": 1.124006,
+      "confidence": 1.0
+    },
+    "tokyo": {
+      "samples": 22,
+      "mu_bias": -0.332129,
+      "sigma_scale": 0.729483,
+      "confidence": 1.0
+    },
+    "seoul": {
+      "samples": 22,
+      "mu_bias": 1.6533,
+      "sigma_scale": 0.82648,
+      "confidence": 1.0
+    },
+    "busan": {
+      "samples": 16,
+      "mu_bias": 0.343239,
+      "sigma_scale": 1.886868,
+      "confidence": 1.0
+    },
+    "mexico city": {
+      "samples": 20,
+      "mu_bias": 0.087122,
+      "sigma_scale": 1.108884,
+      "confidence": 1.0
+    },
+    "shanghai": {
+      "samples": 23,
+      "mu_bias": 0.348022,
+      "sigma_scale": 0.810527,
+      "confidence": 1.0
+    },
+    "beijing": {
+      "samples": 20,
+      "mu_bias": -0.288176,
+      "sigma_scale": 0.952972,
+      "confidence": 1.0
+    },
+    "shenzhen": {
+      "samples": 21,
+      "mu_bias": -0.006114,
+      "sigma_scale": 0.728494,
+      "confidence": 1.0
+    },
+    "wuhan": {
+      "samples": 20,
+      "mu_bias": -0.771154,
+      "sigma_scale": 1.30329,
+      "confidence": 1.0
+    },
+    "chengdu": {
+      "samples": 20,
+      "mu_bias": 0.445906,
+      "sigma_scale": 0.796989,
+      "confidence": 1.0
+    },
+    "chongqing": {
+      "samples": 20,
+      "mu_bias": -0.021301,
+      "sigma_scale": 1.106566,
+      "confidence": 1.0
+    },
+    "hong kong": {
+      "samples": 24,
+      "mu_bias": 0.37527,
+      "sigma_scale": 0.517756,
+      "confidence": 1.0
+    },
+    "taipei": {
+      "samples": 23,
+      "mu_bias": 0.600671,
+      "sigma_scale": 0.5,
+      "confidence": 1.0
+    },
+    "guangzhou": {
+      "samples": 6,
+      "mu_bias": 0.7082,
+      "sigma_scale": 1.975103,
+      "confidence": 0.75
+    },
+    "lau fau shan": {
+      "samples": 19,
+      "mu_bias": 0.41371,
+      "sigma_scale": 0.656578,
+      "confidence": 1.0
+    },
+    "kuala lumpur": {
+      "samples": 16,
+      "mu_bias": 0.499197,
+      "sigma_scale": 1.361171,
+      "confidence": 1.0
+    },
+    "jakarta": {
+      "samples": 16,
+      "mu_bias": 0.293537,
+      "sigma_scale": 0.751498,
+      "confidence": 1.0
+    },
+    "lucknow": {
+      "samples": 22,
+      "mu_bias": -1.593437,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "karachi": {
+      "samples": 5,
+      "mu_bias": -0.035595,
+      "sigma_scale": 0.588612,
+      "confidence": 0.625
+    },
+    "jeddah": {
+      "samples": 10,
+      "mu_bias": 0.516879,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "moscow": {
+      "samples": 20,
+      "mu_bias": -0.609602,
+      "sigma_scale": 0.783723,
+      "confidence": 1.0
+    },
+    "tel aviv": {
+      "samples": 22,
+      "mu_bias": 0.436837,
+      "sigma_scale": 1.46524,
+      "confidence": 1.0
+    }
+  },
+  "metrics": {
+    "sample_count": 978,
+    "mean_crps": 1.330733,
+    "legacy_mean_crps": 1.376936,
+    "legacy_mean_mae": 1.589438,
+    "legacy_bucket_hit_rate": 0.43865,
+    "legacy_bucket_brier": 0.824571,
+    "selected_mean_crps": 1.311763,
+    "selected_mean_mae": 1.553658,
+    "selected_bucket_hit_rate": 0.432515,
+    "selected_bucket_brier": 0.789817,
+    "selected_score": 2.910718,
+    "legacy_score": 2.996333,
+    "filled_actual_from_history": 0,
+    "settlement_history_city_count": 30,
+    "legacy_archive_samples": 54
+  },
+  "source": "artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421121920\\default.json"
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421121920/evaluation_report.json

@@ -0,0 +1,482 @@
+{
+  "summary": {
+    "sample_count": 978,
+    "filled_actual_from_history": 0,
+    "legacy": {
+      "mean_crps": 1.376936,
+      "mean_mae": 1.589438,
+      "bucket_hit_rate": 0.43865
+    },
+    "emos": {
+      "mean_crps": 1.319727,
+      "mean_mae": 1.554869,
+      "bucket_hit_rate": 0.43456
+    },
+    "delta": {
+      "crps": -0.05721,
+      "mae": -0.034568,
+      "bucket_hit_rate": -0.00409
+    }
+  },
+  "by_city": {
+    "amsterdam": {
+      "samples": 17,
+      "legacy_mean_crps": 0.868937,
+      "emos_mean_crps": 0.847633,
+      "legacy_mean_mae": 0.983529,
+      "emos_mean_mae": 0.979456,
+      "legacy_bucket_hit_rate": 0.529412,
+      "emos_bucket_hit_rate": 0.470588
+    },
+    "ankara": {
+      "samples": 23,
+      "legacy_mean_crps": 0.610138,
+      "emos_mean_crps": 0.569029,
+      "legacy_mean_mae": 0.726087,
+      "emos_mean_mae": 0.731312,
+      "legacy_bucket_hit_rate": 0.434783,
+      "emos_bucket_hit_rate": 0.391304
+    },
+    "atlanta": {
+      "samples": 22,
+      "legacy_mean_crps": 4.304444,
+      "emos_mean_crps": 4.435438,
+      "legacy_mean_mae": 4.79,
+      "emos_mean_mae": 4.992948,
+      "legacy_bucket_hit_rate": 0.181818,
+      "emos_bucket_hit_rate": 0.090909
+    },
+    "aurora": {
+      "samples": 20,
+      "legacy_mean_crps": 3.989577,
+      "emos_mean_crps": 3.688746,
+      "legacy_mean_mae": 4.5115,
+      "emos_mean_mae": 4.454993,
+      "legacy_bucket_hit_rate": 0.1,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "austin": {
+      "samples": 20,
+      "legacy_mean_crps": 2.261424,
+      "emos_mean_crps": 2.109386,
+      "legacy_mean_mae": 2.422,
+      "emos_mean_mae": 2.377675,
+      "legacy_bucket_hit_rate": 0.45,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "beijing": {
+      "samples": 20,
+      "legacy_mean_crps": 1.116518,
+      "emos_mean_crps": 1.123144,
+      "legacy_mean_mae": 1.4955,
+      "emos_mean_mae": 1.468371,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "buenos aires": {
+      "samples": 22,
+      "legacy_mean_crps": 1.632274,
+      "emos_mean_crps": 1.657172,
+      "legacy_mean_mae": 1.885,
+      "emos_mean_mae": 1.875542,
+      "legacy_bucket_hit_rate": 0.454545,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "busan": {
+      "samples": 16,
+      "legacy_mean_crps": 1.066146,
+      "emos_mean_crps": 0.99263,
+      "legacy_mean_mae": 1.304375,
+      "emos_mean_mae": 1.259453,
+      "legacy_bucket_hit_rate": 0.375,
+      "emos_bucket_hit_rate": 0.375
+    },
+    "cape town": {
+      "samples": 10,
+      "legacy_mean_crps": 0.7369,
+      "emos_mean_crps": 0.646728,
+      "legacy_mean_mae": 0.882,
+      "emos_mean_mae": 0.809407,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "chengdu": {
+      "samples": 20,
+      "legacy_mean_crps": 1.009046,
+      "emos_mean_crps": 0.964076,
+      "legacy_mean_mae": 1.1835,
+      "emos_mean_mae": 1.155294,
+      "legacy_bucket_hit_rate": 0.5,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "chicago": {
+      "samples": 21,
+      "legacy_mean_crps": 3.052956,
+      "emos_mean_crps": 2.842851,
+      "legacy_mean_mae": 3.398095,
+      "emos_mean_mae": 3.312975,
+      "legacy_bucket_hit_rate": 0.285714,
+      "emos_bucket_hit_rate": 0.095238
+    },
+    "chongqing": {
+      "samples": 20,
+      "legacy_mean_crps": 1.239078,
+      "emos_mean_crps": 1.231804,
+      "legacy_mean_mae": 1.4955,
+      "emos_mean_mae": 1.501486,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "dallas": {
+      "samples": 21,
+      "legacy_mean_crps": 1.795116,
+      "emos_mean_crps": 1.700865,
+      "legacy_mean_mae": 1.979048,
+      "emos_mean_mae": 2.017151,
+      "legacy_bucket_hit_rate": 0.285714,
+      "emos_bucket_hit_rate": 0.238095
+    },
+    "guangzhou": {
+      "samples": 6,
+      "legacy_mean_crps": 1.586654,
+      "emos_mean_crps": 1.565936,
+      "legacy_mean_mae": 1.991667,
+      "emos_mean_mae": 1.995792,
+      "legacy_bucket_hit_rate": 0.166667,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "helsinki": {
+      "samples": 17,
+      "legacy_mean_crps": 0.626777,
+      "emos_mean_crps": 0.489382,
+      "legacy_mean_mae": 0.785294,
+      "emos_mean_mae": 0.640606,
+      "legacy_bucket_hit_rate": 0.470588,
+      "emos_bucket_hit_rate": 0.529412
+    },
+    "hong kong": {
+      "samples": 24,
+      "legacy_mean_crps": 0.531004,
+      "emos_mean_crps": 0.473302,
+      "legacy_mean_mae": 0.552917,
+      "emos_mean_mae": 0.52881,
+      "legacy_bucket_hit_rate": 0.666667,
+      "emos_bucket_hit_rate": 0.75
+    },
+    "houston": {
+      "samples": 20,
+      "legacy_mean_crps": 1.488859,
+      "emos_mean_crps": 1.412564,
+      "legacy_mean_mae": 1.8085,
+      "emos_mean_mae": 1.80359,
+      "legacy_bucket_hit_rate": 0.25,
+      "emos_bucket_hit_rate": 0.2
+    },
+    "istanbul": {
+      "samples": 20,
+      "legacy_mean_crps": 0.586067,
+      "emos_mean_crps": 0.546494,
+      "legacy_mean_mae": 0.6225,
+      "emos_mean_mae": 0.557255,
+      "legacy_bucket_hit_rate": 0.7,
+      "emos_bucket_hit_rate": 0.7
+    },
+    "jakarta": {
+      "samples": 16,
+      "legacy_mean_crps": 0.620583,
+      "emos_mean_crps": 0.553084,
+      "legacy_mean_mae": 0.875625,
+      "emos_mean_mae": 0.793719,
+      "legacy_bucket_hit_rate": 0.4375,
+      "emos_bucket_hit_rate": 0.4375
+    },
+    "jeddah": {
+      "samples": 10,
+      "legacy_mean_crps": 1.822658,
+      "emos_mean_crps": 1.674198,
+      "legacy_mean_mae": 2.277,
+      "emos_mean_mae": 2.192556,
+      "legacy_bucket_hit_rate": 0.2,
+      "emos_bucket_hit_rate": 0.1
+    },
+    "karachi": {
+      "samples": 5,
+      "legacy_mean_crps": 0.361328,
+      "emos_mean_crps": 0.326866,
+      "legacy_mean_mae": 0.398,
+      "emos_mean_mae": 0.336276,
+      "legacy_bucket_hit_rate": 0.8,
+      "emos_bucket_hit_rate": 0.8
+    },
+    "kuala lumpur": {
+      "samples": 16,
+      "legacy_mean_crps": 0.927888,
+      "emos_mean_crps": 0.86633,
+      "legacy_mean_mae": 1.146875,
+      "emos_mean_mae": 1.083193,
+      "legacy_bucket_hit_rate": 0.375,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "lagos": {
+      "samples": 10,
+      "legacy_mean_crps": 1.174893,
+      "emos_mean_crps": 1.204827,
+      "legacy_mean_mae": 1.592,
+      "emos_mean_mae": 1.586221,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "lau fau shan": {
+      "samples": 19,
+      "legacy_mean_crps": 0.713018,
+      "emos_mean_crps": 0.63244,
+      "legacy_mean_mae": 0.706316,
+      "emos_mean_mae": 0.641667,
+      "legacy_bucket_hit_rate": 0.578947,
+      "emos_bucket_hit_rate": 0.578947
+    },
+    "london": {
+      "samples": 22,
+      "legacy_mean_crps": 1.298133,
+      "emos_mean_crps": 1.267504,
+      "legacy_mean_mae": 1.429091,
+      "emos_mean_mae": 1.422989,
+      "legacy_bucket_hit_rate": 0.409091,
+      "emos_bucket_hit_rate": 0.409091
+    },
+    "los angeles": {
+      "samples": 20,
+      "legacy_mean_crps": 2.051516,
+      "emos_mean_crps": 2.006381,
+      "legacy_mean_mae": 2.4145,
+      "emos_mean_mae": 2.366065,
+      "legacy_bucket_hit_rate": 0.3,
+      "emos_bucket_hit_rate": 0.3
+    },
+    "lucknow": {
+      "samples": 22,
+      "legacy_mean_crps": 1.134917,
+      "emos_mean_crps": 0.890494,
+      "legacy_mean_mae": 1.532727,
+      "emos_mean_mae": 1.201547,
+      "legacy_bucket_hit_rate": 0.136364,
+      "emos_bucket_hit_rate": 0.272727
+    },
+    "madrid": {
+      "samples": 22,
+      "legacy_mean_crps": 1.196523,
+      "emos_mean_crps": 1.197589,
+      "legacy_mean_mae": 1.380909,
+      "emos_mean_mae": 1.383717,
+      "legacy_bucket_hit_rate": 0.545455,
+      "emos_bucket_hit_rate": 0.636364
+    },
+    "manila": {
+      "samples": 5,
+      "legacy_mean_crps": 1.9342,
+      "emos_mean_crps": 1.510621,
+      "legacy_mean_mae": 2.408,
+      "emos_mean_mae": 1.914957,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "mexico city": {
+      "samples": 20,
+      "legacy_mean_crps": 0.560208,
+      "emos_mean_crps": 0.506071,
+      "legacy_mean_mae": 0.713,
+      "emos_mean_mae": 0.653865,
+      "legacy_bucket_hit_rate": 0.55,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "miami": {
+      "samples": 22,
+      "legacy_mean_crps": 3.359106,
+      "emos_mean_crps": 3.561458,
+      "legacy_mean_mae": 3.660455,
+      "emos_mean_mae": 3.936442,
+      "legacy_bucket_hit_rate": 0.454545,
+      "emos_bucket_hit_rate": 0.227273
+    },
+    "milan": {
+      "samples": 23,
+      "legacy_mean_crps": 1.184521,
+      "emos_mean_crps": 1.055058,
+      "legacy_mean_mae": 1.284348,
+      "emos_mean_mae": 1.220263,
+      "legacy_bucket_hit_rate": 0.304348,
+      "emos_bucket_hit_rate": 0.434783
+    },
+    "moscow": {
+      "samples": 20,
+      "legacy_mean_crps": 0.617727,
+      "emos_mean_crps": 0.578817,
+      "legacy_mean_mae": 0.6375,
+      "emos_mean_mae": 0.576256,
+      "legacy_bucket_hit_rate": 0.75,
+      "emos_bucket_hit_rate": 0.75
+    },
+    "munich": {
+      "samples": 22,
+      "legacy_mean_crps": 1.101444,
+      "emos_mean_crps": 1.049123,
+      "legacy_mean_mae": 1.298182,
+      "emos_mean_mae": 1.285139,
+      "legacy_bucket_hit_rate": 0.227273,
+      "emos_bucket_hit_rate": 0.227273
+    },
+    "new york": {
+      "samples": 21,
+      "legacy_mean_crps": 3.428596,
+      "emos_mean_crps": 3.466282,
+      "legacy_mean_mae": 3.988571,
+      "emos_mean_mae": 4.038895,
+      "legacy_bucket_hit_rate": 0.238095,
+      "emos_bucket_hit_rate": 0.238095
+    },
+    "panama city": {
+      "samples": 17,
+      "legacy_mean_crps": 0.347679,
+      "emos_mean_crps": 0.287754,
+      "legacy_mean_mae": 0.476471,
+      "emos_mean_mae": 0.391782,
+      "legacy_bucket_hit_rate": 0.588235,
+      "emos_bucket_hit_rate": 0.823529
+    },
+    "paris": {
+      "samples": 22,
+      "legacy_mean_crps": 0.718617,
+      "emos_mean_crps": 0.691394,
+      "legacy_mean_mae": 0.680455,
+      "emos_mean_mae": 0.696494,
+      "legacy_bucket_hit_rate": 0.772727,
+      "emos_bucket_hit_rate": 0.772727
+    },
+    "san francisco": {
+      "samples": 20,
+      "legacy_mean_crps": 2.623636,
+      "emos_mean_crps": 2.475182,
+      "legacy_mean_mae": 2.9075,
+      "emos_mean_mae": 2.837902,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.25
+    },
+    "sao paulo": {
+      "samples": 22,
+      "legacy_mean_crps": 0.930626,
+      "emos_mean_crps": 0.931492,
+      "legacy_mean_mae": 1.16,
+      "emos_mean_mae": 1.123038,
+      "legacy_bucket_hit_rate": 0.590909,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "seattle": {
+      "samples": 21,
+      "legacy_mean_crps": 1.378395,
+      "emos_mean_crps": 1.333721,
+      "legacy_mean_mae": 1.601905,
+      "emos_mean_mae": 1.565253,
+      "legacy_bucket_hit_rate": 0.428571,
+      "emos_bucket_hit_rate": 0.380952
+    },
+    "seoul": {
+      "samples": 22,
+      "legacy_mean_crps": 1.449956,
+      "emos_mean_crps": 1.264612,
+      "legacy_mean_mae": 1.817727,
+      "emos_mean_mae": 1.604882,
+      "legacy_bucket_hit_rate": 0.409091,
+      "emos_bucket_hit_rate": 0.454545
+    },
+    "shanghai": {
+      "samples": 23,
+      "legacy_mean_crps": 1.237937,
+      "emos_mean_crps": 1.196583,
+      "legacy_mean_mae": 1.458261,
+      "emos_mean_mae": 1.430445,
+      "legacy_bucket_hit_rate": 0.565217,
+      "emos_bucket_hit_rate": 0.608696
+    },
+    "shenzhen": {
+      "samples": 21,
+      "legacy_mean_crps": 0.481109,
+      "emos_mean_crps": 0.444093,
+      "legacy_mean_mae": 0.608095,
+      "emos_mean_mae": 0.581253,
+      "legacy_bucket_hit_rate": 0.619048,
+      "emos_bucket_hit_rate": 0.666667
+    },
+    "singapore": {
+      "samples": 22,
+      "legacy_mean_crps": 0.852942,
+      "emos_mean_crps": 0.758454,
+      "legacy_mean_mae": 1.020909,
+      "emos_mean_mae": 0.915982,
+      "legacy_bucket_hit_rate": 0.409091,
+      "emos_bucket_hit_rate": 0.454545
+    },
+    "taipei": {
+      "samples": 23,
+      "legacy_mean_crps": 0.801052,
+      "emos_mean_crps": 0.721468,
+      "legacy_mean_mae": 0.85087,
+      "emos_mean_mae": 0.814252,
+      "legacy_bucket_hit_rate": 0.478261,
+      "emos_bucket_hit_rate": 0.391304
+    },
+    "tel aviv": {
+      "samples": 22,
+      "legacy_mean_crps": 0.770763,
+      "emos_mean_crps": 0.672956,
+      "legacy_mean_mae": 0.904091,
+      "emos_mean_mae": 0.843671,
+      "legacy_bucket_hit_rate": 0.454545,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "tokyo": {
+      "samples": 22,
+      "legacy_mean_crps": 0.651385,
+      "emos_mean_crps": 0.641684,
+      "legacy_mean_mae": 0.798182,
+      "emos_mean_mae": 0.787542,
+      "legacy_bucket_hit_rate": 0.545455,
+      "emos_bucket_hit_rate": 0.590909
+    },
+    "toronto": {
+      "samples": 22,
+      "legacy_mean_crps": 2.733044,
+      "emos_mean_crps": 2.714624,
+      "legacy_mean_mae": 2.981818,
+      "emos_mean_mae": 2.969098,
+      "legacy_bucket_hit_rate": 0.318182,
+      "emos_bucket_hit_rate": 0.409091
+    },
+    "warsaw": {
+      "samples": 23,
+      "legacy_mean_crps": 0.668776,
+      "emos_mean_crps": 0.665164,
+      "legacy_mean_mae": 0.796087,
+      "emos_mean_mae": 0.796958,
+      "legacy_bucket_hit_rate": 0.652174,
+      "emos_bucket_hit_rate": 0.608696
+    },
+    "wellington": {
+      "samples": 22,
+      "legacy_mean_crps": 0.731346,
+      "emos_mean_crps": 0.726977,
+      "legacy_mean_mae": 0.895,
+      "emos_mean_mae": 0.892558,
+      "legacy_bucket_hit_rate": 0.5,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "wuhan": {
+      "samples": 20,
+      "legacy_mean_crps": 0.925443,
+      "emos_mean_crps": 0.897806,
+      "legacy_mean_mae": 1.0375,
+      "emos_mean_mae": 1.003133,
+      "legacy_bucket_hit_rate": 0.65,
+      "emos_bucket_hit_rate": 0.65
+    }
+  }
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421122743/decision_report.json

@@ -0,0 +1,62 @@
+{
+  "ok": true,
+  "version": "emos-auto-20260421122743",
+  "generated_at": "2026-04-21T12:28:08.021388+00:00",
+  "candidate_dir": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743",
+  "candidate_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json",
+  "evaluation_path": "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\evaluation_report.json",
+  "target_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\default.json",
+  "promote_requested": true,
+  "promoted": true,
+  "backup_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\default.backup-20260421122808.json",
+  "decision": {
+    "decision": "promote",
+    "ready_for_promotion": true,
+    "blocking_reasons": [],
+    "thresholds": {
+      "min_samples": 50,
+      "max_delta_crps": 0.0,
+      "max_delta_mae": 0.05,
+      "min_delta_bucket_hit_rate": -0.05
+    },
+    "metrics": {
+      "sample_count": 978,
+      "delta_crps": -0.05721,
+      "delta_mae": -0.034568,
+      "delta_bucket_hit_rate": -0.00409
+    }
+  },
+  "fit": {
+    "command": [
+      "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
+      "scripts/fit_probability_calibration.py",
+      "--output",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json",
+      "--version",
+      "emos-auto-20260421122743",
+      "--verbose",
+      "--snapshot-limit",
+      "50000"
+    ],
+    "returncode": 0,
+    "stdout": "",
+    "stderr": ""
+  },
+  "evaluate": {
+    "command": [
+      "C:\\Users\\Yang\\AppData\\Local\\Programs\\Python\\Python38\\python.exe",
+      "scripts/evaluate_probability_calibration.py",
+      "--calibration-file",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json",
+      "--output",
+      "E:\\web\\PolyWeather\\artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\evaluation_report.json",
+      "--verbose",
+      "--snapshot-limit",
+      "50000"
+    ],
+    "returncode": 0,
+    "stdout": "",
+    "stderr": ""
+  },
+  "tests": null
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421122743/default.json

@@ -0,0 +1,361 @@
+{
+  "version": "emos-auto-20260421122743",
+  "trained_at": "2026-04-21T12:28:05.031165+00:00",
+  "global": {
+    "mu": {
+      "intercept": 0.34330438,
+      "raw_mu_coef": 0.48749629,
+      "deb_coef": 0.45733479,
+      "ens_median_coef": 0.05955297,
+      "max_so_far_gap_coef": -0.104935
+    },
+    "sigma": {
+      "intercept": -0.62765582,
+      "raw_sigma_coef": 0.38492261,
+      "spread_coef": 0.29095921,
+      "peak_flag_coef": 0.33966327,
+      "max_so_far_gap_coef": 0.08357376
+    }
+  },
+  "sigma_constraints": {
+    "min_ratio": 0.85,
+    "max_ratio": 1.35,
+    "absolute_min": 0.25,
+    "absolute_max": 3.0
+  },
+  "selection_guardrails": {
+    "max_mae_increase": 0.02,
+    "max_bucket_hit_drop": 0.01,
+    "max_bucket_brier_increase": 0.05
+  },
+  "blending": {
+    "alpha_mu": 0.25,
+    "alpha_sigma": 1.0
+  },
+  "cities": {
+    "istanbul": {
+      "samples": 20,
+      "mu_bias": -0.438949,
+      "sigma_scale": 0.553793,
+      "confidence": 1.0
+    },
+    "buenos aires": {
+      "samples": 22,
+      "mu_bias": -1.044971,
+      "sigma_scale": 1.389753,
+      "confidence": 1.0
+    },
+    "ankara": {
+      "samples": 23,
+      "mu_bias": 0.067406,
+      "sigma_scale": 1.270298,
+      "confidence": 1.0
+    },
+    "london": {
+      "samples": 22,
+      "mu_bias": -0.010833,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "paris": {
+      "samples": 22,
+      "mu_bias": -0.21677,
+      "sigma_scale": 1.095669,
+      "confidence": 1.0
+    },
+    "new york": {
+      "samples": 21,
+      "mu_bias": 1.072189,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "lagos": {
+      "samples": 10,
+      "mu_bias": -0.092732,
+      "sigma_scale": 1.558646,
+      "confidence": 1.0
+    },
+    "madrid": {
+      "samples": 22,
+      "mu_bias": -0.627862,
+      "sigma_scale": 1.761239,
+      "confidence": 1.0
+    },
+    "toronto": {
+      "samples": 22,
+      "mu_bias": -0.424101,
+      "sigma_scale": 1.326675,
+      "confidence": 1.0
+    },
+    "wellington": {
+      "samples": 22,
+      "mu_bias": -0.347796,
+      "sigma_scale": 0.893459,
+      "confidence": 1.0
+    },
+    "los angeles": {
+      "samples": 20,
+      "mu_bias": -0.659983,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "san francisco": {
+      "samples": 20,
+      "mu_bias": 0.941818,
+      "sigma_scale": 1.979403,
+      "confidence": 1.0
+    },
+    "aurora": {
+      "samples": 20,
+      "mu_bias": 2.040009,
+      "sigma_scale": 0.982492,
+      "confidence": 1.0
+    },
+    "austin": {
+      "samples": 20,
+      "mu_bias": 0.990023,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "houston": {
+      "samples": 20,
+      "mu_bias": 0.188787,
+      "sigma_scale": 1.819594,
+      "confidence": 1.0
+    },
+    "chicago": {
+      "samples": 21,
+      "mu_bias": 2.285098,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "dallas": {
+      "samples": 21,
+      "mu_bias": 0.258095,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "atlanta": {
+      "samples": 22,
+      "mu_bias": -2.093105,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "seattle": {
+      "samples": 21,
+      "mu_bias": -0.982188,
+      "sigma_scale": 1.695172,
+      "confidence": 1.0
+    },
+    "sao paulo": {
+      "samples": 22,
+      "mu_bias": -0.786242,
+      "sigma_scale": 1.427037,
+      "confidence": 1.0
+    },
+    "munich": {
+      "samples": 22,
+      "mu_bias": 0.096559,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "milan": {
+      "samples": 23,
+      "mu_bias": -0.045038,
+      "sigma_scale": 1.626331,
+      "confidence": 1.0
+    },
+    "warsaw": {
+      "samples": 23,
+      "mu_bias": -0.50853,
+      "sigma_scale": 1.239146,
+      "confidence": 1.0
+    },
+    "helsinki": {
+      "samples": 17,
+      "mu_bias": 0.442247,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "amsterdam": {
+      "samples": 17,
+      "mu_bias": 0.129889,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "panama city": {
+      "samples": 17,
+      "mu_bias": -0.040792,
+      "sigma_scale": 1.393259,
+      "confidence": 1.0
+    },
+    "cape town": {
+      "samples": 10,
+      "mu_bias": 0.507293,
+      "sigma_scale": 1.420004,
+      "confidence": 1.0
+    },
+    "miami": {
+      "samples": 22,
+      "mu_bias": -2.736316,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "manila": {
+      "samples": 5,
+      "mu_bias": 1.772229,
+      "sigma_scale": 2.0,
+      "confidence": 0.625
+    },
+    "singapore": {
+      "samples": 22,
+      "mu_bias": 0.289639,
+      "sigma_scale": 1.124006,
+      "confidence": 1.0
+    },
+    "tokyo": {
+      "samples": 22,
+      "mu_bias": -0.332129,
+      "sigma_scale": 0.729483,
+      "confidence": 1.0
+    },
+    "seoul": {
+      "samples": 22,
+      "mu_bias": 1.6533,
+      "sigma_scale": 0.82648,
+      "confidence": 1.0
+    },
+    "busan": {
+      "samples": 16,
+      "mu_bias": 0.343239,
+      "sigma_scale": 1.886868,
+      "confidence": 1.0
+    },
+    "mexico city": {
+      "samples": 20,
+      "mu_bias": 0.087122,
+      "sigma_scale": 1.108884,
+      "confidence": 1.0
+    },
+    "shanghai": {
+      "samples": 23,
+      "mu_bias": 0.348022,
+      "sigma_scale": 0.810527,
+      "confidence": 1.0
+    },
+    "beijing": {
+      "samples": 20,
+      "mu_bias": -0.288176,
+      "sigma_scale": 0.952972,
+      "confidence": 1.0
+    },
+    "shenzhen": {
+      "samples": 21,
+      "mu_bias": -0.006114,
+      "sigma_scale": 0.728494,
+      "confidence": 1.0
+    },
+    "wuhan": {
+      "samples": 20,
+      "mu_bias": -0.771154,
+      "sigma_scale": 1.30329,
+      "confidence": 1.0
+    },
+    "chengdu": {
+      "samples": 20,
+      "mu_bias": 0.445906,
+      "sigma_scale": 0.796989,
+      "confidence": 1.0
+    },
+    "chongqing": {
+      "samples": 20,
+      "mu_bias": -0.021301,
+      "sigma_scale": 1.106566,
+      "confidence": 1.0
+    },
+    "hong kong": {
+      "samples": 24,
+      "mu_bias": 0.37527,
+      "sigma_scale": 0.517756,
+      "confidence": 1.0
+    },
+    "taipei": {
+      "samples": 23,
+      "mu_bias": 0.600671,
+      "sigma_scale": 0.5,
+      "confidence": 1.0
+    },
+    "guangzhou": {
+      "samples": 6,
+      "mu_bias": 0.7082,
+      "sigma_scale": 1.975103,
+      "confidence": 0.75
+    },
+    "lau fau shan": {
+      "samples": 19,
+      "mu_bias": 0.41371,
+      "sigma_scale": 0.656578,
+      "confidence": 1.0
+    },
+    "kuala lumpur": {
+      "samples": 16,
+      "mu_bias": 0.499197,
+      "sigma_scale": 1.361171,
+      "confidence": 1.0
+    },
+    "jakarta": {
+      "samples": 16,
+      "mu_bias": 0.293537,
+      "sigma_scale": 0.751498,
+      "confidence": 1.0
+    },
+    "lucknow": {
+      "samples": 22,
+      "mu_bias": -1.593437,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "karachi": {
+      "samples": 5,
+      "mu_bias": -0.035595,
+      "sigma_scale": 0.588612,
+      "confidence": 0.625
+    },
+    "jeddah": {
+      "samples": 10,
+      "mu_bias": 0.516879,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "moscow": {
+      "samples": 20,
+      "mu_bias": -0.609602,
+      "sigma_scale": 0.783723,
+      "confidence": 1.0
+    },
+    "tel aviv": {
+      "samples": 22,
+      "mu_bias": 0.436837,
+      "sigma_scale": 1.46524,
+      "confidence": 1.0
+    }
+  },
+  "metrics": {
+    "sample_count": 978,
+    "mean_crps": 1.330733,
+    "legacy_mean_crps": 1.376936,
+    "legacy_mean_mae": 1.589438,
+    "legacy_bucket_hit_rate": 0.43865,
+    "legacy_bucket_brier": 0.824571,
+    "selected_mean_crps": 1.311763,
+    "selected_mean_mae": 1.553658,
+    "selected_bucket_hit_rate": 0.432515,
+    "selected_bucket_brier": 0.789817,
+    "selected_score": 2.910718,
+    "legacy_score": 2.996333,
+    "filled_actual_from_history": 0,
+    "settlement_history_city_count": 30,
+    "legacy_archive_samples": 54
+  },
+  "source": "artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json"
+}

artifacts/local_runtime/probability_calibration/candidates/emos-auto-20260421122743/evaluation_report.json

@@ -0,0 +1,482 @@
+{
+  "summary": {
+    "sample_count": 978,
+    "filled_actual_from_history": 0,
+    "legacy": {
+      "mean_crps": 1.376936,
+      "mean_mae": 1.589438,
+      "bucket_hit_rate": 0.43865
+    },
+    "emos": {
+      "mean_crps": 1.319727,
+      "mean_mae": 1.554869,
+      "bucket_hit_rate": 0.43456
+    },
+    "delta": {
+      "crps": -0.05721,
+      "mae": -0.034568,
+      "bucket_hit_rate": -0.00409
+    }
+  },
+  "by_city": {
+    "amsterdam": {
+      "samples": 17,
+      "legacy_mean_crps": 0.868937,
+      "emos_mean_crps": 0.847633,
+      "legacy_mean_mae": 0.983529,
+      "emos_mean_mae": 0.979456,
+      "legacy_bucket_hit_rate": 0.529412,
+      "emos_bucket_hit_rate": 0.470588
+    },
+    "ankara": {
+      "samples": 23,
+      "legacy_mean_crps": 0.610138,
+      "emos_mean_crps": 0.569029,
+      "legacy_mean_mae": 0.726087,
+      "emos_mean_mae": 0.731312,
+      "legacy_bucket_hit_rate": 0.434783,
+      "emos_bucket_hit_rate": 0.391304
+    },
+    "atlanta": {
+      "samples": 22,
+      "legacy_mean_crps": 4.304444,
+      "emos_mean_crps": 4.435438,
+      "legacy_mean_mae": 4.79,
+      "emos_mean_mae": 4.992948,
+      "legacy_bucket_hit_rate": 0.181818,
+      "emos_bucket_hit_rate": 0.090909
+    },
+    "aurora": {
+      "samples": 20,
+      "legacy_mean_crps": 3.989577,
+      "emos_mean_crps": 3.688746,
+      "legacy_mean_mae": 4.5115,
+      "emos_mean_mae": 4.454993,
+      "legacy_bucket_hit_rate": 0.1,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "austin": {
+      "samples": 20,
+      "legacy_mean_crps": 2.261424,
+      "emos_mean_crps": 2.109386,
+      "legacy_mean_mae": 2.422,
+      "emos_mean_mae": 2.377675,
+      "legacy_bucket_hit_rate": 0.45,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "beijing": {
+      "samples": 20,
+      "legacy_mean_crps": 1.116518,
+      "emos_mean_crps": 1.123144,
+      "legacy_mean_mae": 1.4955,
+      "emos_mean_mae": 1.468371,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "buenos aires": {
+      "samples": 22,
+      "legacy_mean_crps": 1.632274,
+      "emos_mean_crps": 1.657172,
+      "legacy_mean_mae": 1.885,
+      "emos_mean_mae": 1.875542,
+      "legacy_bucket_hit_rate": 0.454545,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "busan": {
+      "samples": 16,
+      "legacy_mean_crps": 1.066146,
+      "emos_mean_crps": 0.99263,
+      "legacy_mean_mae": 1.304375,
+      "emos_mean_mae": 1.259453,
+      "legacy_bucket_hit_rate": 0.375,
+      "emos_bucket_hit_rate": 0.375
+    },
+    "cape town": {
+      "samples": 10,
+      "legacy_mean_crps": 0.7369,
+      "emos_mean_crps": 0.646728,
+      "legacy_mean_mae": 0.882,
+      "emos_mean_mae": 0.809407,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "chengdu": {
+      "samples": 20,
+      "legacy_mean_crps": 1.009046,
+      "emos_mean_crps": 0.964076,
+      "legacy_mean_mae": 1.1835,
+      "emos_mean_mae": 1.155294,
+      "legacy_bucket_hit_rate": 0.5,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "chicago": {
+      "samples": 21,
+      "legacy_mean_crps": 3.052956,
+      "emos_mean_crps": 2.842851,
+      "legacy_mean_mae": 3.398095,
+      "emos_mean_mae": 3.312975,
+      "legacy_bucket_hit_rate": 0.285714,
+      "emos_bucket_hit_rate": 0.095238
+    },
+    "chongqing": {
+      "samples": 20,
+      "legacy_mean_crps": 1.239078,
+      "emos_mean_crps": 1.231804,
+      "legacy_mean_mae": 1.4955,
+      "emos_mean_mae": 1.501486,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.4
+    },
+    "dallas": {
+      "samples": 21,
+      "legacy_mean_crps": 1.795116,
+      "emos_mean_crps": 1.700865,
+      "legacy_mean_mae": 1.979048,
+      "emos_mean_mae": 2.017151,
+      "legacy_bucket_hit_rate": 0.285714,
+      "emos_bucket_hit_rate": 0.238095
+    },
+    "guangzhou": {
+      "samples": 6,
+      "legacy_mean_crps": 1.586654,
+      "emos_mean_crps": 1.565936,
+      "legacy_mean_mae": 1.991667,
+      "emos_mean_mae": 1.995792,
+      "legacy_bucket_hit_rate": 0.166667,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "helsinki": {
+      "samples": 17,
+      "legacy_mean_crps": 0.626777,
+      "emos_mean_crps": 0.489382,
+      "legacy_mean_mae": 0.785294,
+      "emos_mean_mae": 0.640606,
+      "legacy_bucket_hit_rate": 0.470588,
+      "emos_bucket_hit_rate": 0.529412
+    },
+    "hong kong": {
+      "samples": 24,
+      "legacy_mean_crps": 0.531004,
+      "emos_mean_crps": 0.473302,
+      "legacy_mean_mae": 0.552917,
+      "emos_mean_mae": 0.52881,
+      "legacy_bucket_hit_rate": 0.666667,
+      "emos_bucket_hit_rate": 0.75
+    },
+    "houston": {
+      "samples": 20,
+      "legacy_mean_crps": 1.488859,
+      "emos_mean_crps": 1.412564,
+      "legacy_mean_mae": 1.8085,
+      "emos_mean_mae": 1.80359,
+      "legacy_bucket_hit_rate": 0.25,
+      "emos_bucket_hit_rate": 0.2
+    },
+    "istanbul": {
+      "samples": 20,
+      "legacy_mean_crps": 0.586067,
+      "emos_mean_crps": 0.546494,
+      "legacy_mean_mae": 0.6225,
+      "emos_mean_mae": 0.557255,
+      "legacy_bucket_hit_rate": 0.7,
+      "emos_bucket_hit_rate": 0.7
+    },
+    "jakarta": {
+      "samples": 16,
+      "legacy_mean_crps": 0.620583,
+      "emos_mean_crps": 0.553084,
+      "legacy_mean_mae": 0.875625,
+      "emos_mean_mae": 0.793719,
+      "legacy_bucket_hit_rate": 0.4375,
+      "emos_bucket_hit_rate": 0.4375
+    },
+    "jeddah": {
+      "samples": 10,
+      "legacy_mean_crps": 1.822658,
+      "emos_mean_crps": 1.674198,
+      "legacy_mean_mae": 2.277,
+      "emos_mean_mae": 2.192556,
+      "legacy_bucket_hit_rate": 0.2,
+      "emos_bucket_hit_rate": 0.1
+    },
+    "karachi": {
+      "samples": 5,
+      "legacy_mean_crps": 0.361328,
+      "emos_mean_crps": 0.326866,
+      "legacy_mean_mae": 0.398,
+      "emos_mean_mae": 0.336276,
+      "legacy_bucket_hit_rate": 0.8,
+      "emos_bucket_hit_rate": 0.8
+    },
+    "kuala lumpur": {
+      "samples": 16,
+      "legacy_mean_crps": 0.927888,
+      "emos_mean_crps": 0.86633,
+      "legacy_mean_mae": 1.146875,
+      "emos_mean_mae": 1.083193,
+      "legacy_bucket_hit_rate": 0.375,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "lagos": {
+      "samples": 10,
+      "legacy_mean_crps": 1.174893,
+      "emos_mean_crps": 1.204827,
+      "legacy_mean_mae": 1.592,
+      "emos_mean_mae": 1.586221,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "lau fau shan": {
+      "samples": 19,
+      "legacy_mean_crps": 0.713018,
+      "emos_mean_crps": 0.63244,
+      "legacy_mean_mae": 0.706316,
+      "emos_mean_mae": 0.641667,
+      "legacy_bucket_hit_rate": 0.578947,
+      "emos_bucket_hit_rate": 0.578947
+    },
+    "london": {
+      "samples": 22,
+      "legacy_mean_crps": 1.298133,
+      "emos_mean_crps": 1.267504,
+      "legacy_mean_mae": 1.429091,
+      "emos_mean_mae": 1.422989,
+      "legacy_bucket_hit_rate": 0.409091,
+      "emos_bucket_hit_rate": 0.409091
+    },
+    "los angeles": {
+      "samples": 20,
+      "legacy_mean_crps": 2.051516,
+      "emos_mean_crps": 2.006381,
+      "legacy_mean_mae": 2.4145,
+      "emos_mean_mae": 2.366065,
+      "legacy_bucket_hit_rate": 0.3,
+      "emos_bucket_hit_rate": 0.3
+    },
+    "lucknow": {
+      "samples": 22,
+      "legacy_mean_crps": 1.134917,
+      "emos_mean_crps": 0.890494,
+      "legacy_mean_mae": 1.532727,
+      "emos_mean_mae": 1.201547,
+      "legacy_bucket_hit_rate": 0.136364,
+      "emos_bucket_hit_rate": 0.272727
+    },
+    "madrid": {
+      "samples": 22,
+      "legacy_mean_crps": 1.196523,
+      "emos_mean_crps": 1.197589,
+      "legacy_mean_mae": 1.380909,
+      "emos_mean_mae": 1.383717,
+      "legacy_bucket_hit_rate": 0.545455,
+      "emos_bucket_hit_rate": 0.636364
+    },
+    "manila": {
+      "samples": 5,
+      "legacy_mean_crps": 1.9342,
+      "emos_mean_crps": 1.510621,
+      "legacy_mean_mae": 2.408,
+      "emos_mean_mae": 1.914957,
+      "legacy_bucket_hit_rate": 0.0,
+      "emos_bucket_hit_rate": 0.0
+    },
+    "mexico city": {
+      "samples": 20,
+      "legacy_mean_crps": 0.560208,
+      "emos_mean_crps": 0.506071,
+      "legacy_mean_mae": 0.713,
+      "emos_mean_mae": 0.653865,
+      "legacy_bucket_hit_rate": 0.55,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "miami": {
+      "samples": 22,
+      "legacy_mean_crps": 3.359106,
+      "emos_mean_crps": 3.561458,
+      "legacy_mean_mae": 3.660455,
+      "emos_mean_mae": 3.936442,
+      "legacy_bucket_hit_rate": 0.454545,
+      "emos_bucket_hit_rate": 0.227273
+    },
+    "milan": {
+      "samples": 23,
+      "legacy_mean_crps": 1.184521,
+      "emos_mean_crps": 1.055058,
+      "legacy_mean_mae": 1.284348,
+      "emos_mean_mae": 1.220263,
+      "legacy_bucket_hit_rate": 0.304348,
+      "emos_bucket_hit_rate": 0.434783
+    },
+    "moscow": {
+      "samples": 20,
+      "legacy_mean_crps": 0.617727,
+      "emos_mean_crps": 0.578817,
+      "legacy_mean_mae": 0.6375,
+      "emos_mean_mae": 0.576256,
+      "legacy_bucket_hit_rate": 0.75,
+      "emos_bucket_hit_rate": 0.75
+    },
+    "munich": {
+      "samples": 22,
+      "legacy_mean_crps": 1.101444,
+      "emos_mean_crps": 1.049123,
+      "legacy_mean_mae": 1.298182,
+      "emos_mean_mae": 1.285139,
+      "legacy_bucket_hit_rate": 0.227273,
+      "emos_bucket_hit_rate": 0.227273
+    },
+    "new york": {
+      "samples": 21,
+      "legacy_mean_crps": 3.428596,
+      "emos_mean_crps": 3.466282,
+      "legacy_mean_mae": 3.988571,
+      "emos_mean_mae": 4.038895,
+      "legacy_bucket_hit_rate": 0.238095,
+      "emos_bucket_hit_rate": 0.238095
+    },
+    "panama city": {
+      "samples": 17,
+      "legacy_mean_crps": 0.347679,
+      "emos_mean_crps": 0.287754,
+      "legacy_mean_mae": 0.476471,
+      "emos_mean_mae": 0.391782,
+      "legacy_bucket_hit_rate": 0.588235,
+      "emos_bucket_hit_rate": 0.823529
+    },
+    "paris": {
+      "samples": 22,
+      "legacy_mean_crps": 0.718617,
+      "emos_mean_crps": 0.691394,
+      "legacy_mean_mae": 0.680455,
+      "emos_mean_mae": 0.696494,
+      "legacy_bucket_hit_rate": 0.772727,
+      "emos_bucket_hit_rate": 0.772727
+    },
+    "san francisco": {
+      "samples": 20,
+      "legacy_mean_crps": 2.623636,
+      "emos_mean_crps": 2.475182,
+      "legacy_mean_mae": 2.9075,
+      "emos_mean_mae": 2.837902,
+      "legacy_bucket_hit_rate": 0.4,
+      "emos_bucket_hit_rate": 0.25
+    },
+    "sao paulo": {
+      "samples": 22,
+      "legacy_mean_crps": 0.930626,
+      "emos_mean_crps": 0.931492,
+      "legacy_mean_mae": 1.16,
+      "emos_mean_mae": 1.123038,
+      "legacy_bucket_hit_rate": 0.590909,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "seattle": {
+      "samples": 21,
+      "legacy_mean_crps": 1.378395,
+      "emos_mean_crps": 1.333721,
+      "legacy_mean_mae": 1.601905,
+      "emos_mean_mae": 1.565253,
+      "legacy_bucket_hit_rate": 0.428571,
+      "emos_bucket_hit_rate": 0.380952
+    },
+    "seoul": {
+      "samples": 22,
+      "legacy_mean_crps": 1.449956,
+      "emos_mean_crps": 1.264612,
+      "legacy_mean_mae": 1.817727,
+      "emos_mean_mae": 1.604882,
+      "legacy_bucket_hit_rate": 0.409091,
+      "emos_bucket_hit_rate": 0.454545
+    },
+    "shanghai": {
+      "samples": 23,
+      "legacy_mean_crps": 1.237937,
+      "emos_mean_crps": 1.196583,
+      "legacy_mean_mae": 1.458261,
+      "emos_mean_mae": 1.430445,
+      "legacy_bucket_hit_rate": 0.565217,
+      "emos_bucket_hit_rate": 0.608696
+    },
+    "shenzhen": {
+      "samples": 21,
+      "legacy_mean_crps": 0.481109,
+      "emos_mean_crps": 0.444093,
+      "legacy_mean_mae": 0.608095,
+      "emos_mean_mae": 0.581253,
+      "legacy_bucket_hit_rate": 0.619048,
+      "emos_bucket_hit_rate": 0.666667
+    },
+    "singapore": {
+      "samples": 22,
+      "legacy_mean_crps": 0.852942,
+      "emos_mean_crps": 0.758454,
+      "legacy_mean_mae": 1.020909,
+      "emos_mean_mae": 0.915982,
+      "legacy_bucket_hit_rate": 0.409091,
+      "emos_bucket_hit_rate": 0.454545
+    },
+    "taipei": {
+      "samples": 23,
+      "legacy_mean_crps": 0.801052,
+      "emos_mean_crps": 0.721468,
+      "legacy_mean_mae": 0.85087,
+      "emos_mean_mae": 0.814252,
+      "legacy_bucket_hit_rate": 0.478261,
+      "emos_bucket_hit_rate": 0.391304
+    },
+    "tel aviv": {
+      "samples": 22,
+      "legacy_mean_crps": 0.770763,
+      "emos_mean_crps": 0.672956,
+      "legacy_mean_mae": 0.904091,
+      "emos_mean_mae": 0.843671,
+      "legacy_bucket_hit_rate": 0.454545,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "tokyo": {
+      "samples": 22,
+      "legacy_mean_crps": 0.651385,
+      "emos_mean_crps": 0.641684,
+      "legacy_mean_mae": 0.798182,
+      "emos_mean_mae": 0.787542,
+      "legacy_bucket_hit_rate": 0.545455,
+      "emos_bucket_hit_rate": 0.590909
+    },
+    "toronto": {
+      "samples": 22,
+      "legacy_mean_crps": 2.733044,
+      "emos_mean_crps": 2.714624,
+      "legacy_mean_mae": 2.981818,
+      "emos_mean_mae": 2.969098,
+      "legacy_bucket_hit_rate": 0.318182,
+      "emos_bucket_hit_rate": 0.409091
+    },
+    "warsaw": {
+      "samples": 23,
+      "legacy_mean_crps": 0.668776,
+      "emos_mean_crps": 0.665164,
+      "legacy_mean_mae": 0.796087,
+      "emos_mean_mae": 0.796958,
+      "legacy_bucket_hit_rate": 0.652174,
+      "emos_bucket_hit_rate": 0.608696
+    },
+    "wellington": {
+      "samples": 22,
+      "legacy_mean_crps": 0.731346,
+      "emos_mean_crps": 0.726977,
+      "legacy_mean_mae": 0.895,
+      "emos_mean_mae": 0.892558,
+      "legacy_bucket_hit_rate": 0.5,
+      "emos_bucket_hit_rate": 0.5
+    },
+    "wuhan": {
+      "samples": 20,
+      "legacy_mean_crps": 0.925443,
+      "emos_mean_crps": 0.897806,
+      "legacy_mean_mae": 1.0375,
+      "emos_mean_mae": 1.003133,
+      "legacy_bucket_hit_rate": 0.65,
+      "emos_bucket_hit_rate": 0.65
+    }
+  }
+}

artifacts/probability_calibration/default.backup-20260421122808.json

@@ -0,0 +1,115 @@
+{
+  "version": "emos-20260418192717",
+  "trained_at": "2026-04-18T19:27:17.348931+00:00",
+  "global": {
+    "mu": {
+      "intercept": 2.60074353,
+      "raw_mu_coef": 1.97765258,
+      "deb_coef": -1.37457626,
+      "ens_median_coef": 0.10693472,
+      "max_so_far_gap_coef": -0.58256593
+    },
+    "sigma": {
+      "intercept": 0.6373609,
+      "raw_sigma_coef": -1.41537225,
+      "spread_coef": 2.12997577,
+      "peak_flag_coef": -0.08633203,
+      "max_so_far_gap_coef": -0.00716383
+    }
+  },
+  "sigma_constraints": {
+    "min_ratio": 0.85,
+    "max_ratio": 1.35,
+    "absolute_min": 0.25,
+    "absolute_max": 3.0
+  },
+  "selection_guardrails": {
+    "max_mae_increase": 0.02,
+    "max_bucket_hit_drop": 0.01,
+    "max_bucket_brier_increase": 0.05
+  },
+  "blending": {
+    "alpha_mu": 0.05,
+    "alpha_sigma": 0.7
+  },
+  "cities": {
+    "tokyo": {
+      "samples": 5,
+      "mu_bias": 2.101101,
+      "sigma_scale": 2.0,
+      "confidence": 0.625
+    },
+    "ankara": {
+      "samples": 6,
+      "mu_bias": -0.523145,
+      "sigma_scale": 1.139849,
+      "confidence": 0.75
+    },
+    "istanbul": {
+      "samples": 3,
+      "mu_bias": -0.285183,
+      "sigma_scale": 0.550171,
+      "confidence": 0.375
+    },
+    "hong kong": {
+      "samples": 7,
+      "mu_bias": 3.234676,
+      "sigma_scale": 2.0,
+      "confidence": 0.875
+    },
+    "taipei": {
+      "samples": 5,
+      "mu_bias": 4.731246,
+      "sigma_scale": 2.0,
+      "confidence": 0.625
+    },
+    "busan": {
+      "samples": 3,
+      "mu_bias": 0.177772,
+      "sigma_scale": 2.0,
+      "confidence": 0.375
+    },
+    "seoul": {
+      "samples": 3,
+      "mu_bias": -3.070325,
+      "sigma_scale": 1.462196,
+      "confidence": 0.375
+    },
+    "milan": {
+      "samples": 3,
+      "mu_bias": -4.249866,
+      "sigma_scale": 2.0,
+      "confidence": 0.375
+    },
+    "shanghai": {
+      "samples": 3,
+      "mu_bias": 1.080039,
+      "sigma_scale": 1.638119,
+      "confidence": 0.375
+    },
+    "warsaw": {
+      "samples": 3,
+      "mu_bias": -1.68611,
+      "sigma_scale": 1.38114,
+      "confidence": 0.375
+    }
+  },
+  "metrics": {
+    "sample_count": 74,
+    "mean_crps": 3.181501,
+    "legacy_mean_crps": 3.474108,
+    "legacy_mean_mae": 3.679324,
+    "legacy_bucket_hit_rate": 0.5,
+    "legacy_bucket_brier": 0.855395,
+    "selected_mean_crps": 3.322469,
+    "selected_mean_mae": 3.618254,
+    "selected_bucket_hit_rate": 0.5,
+    "selected_bucket_brier": 0.863048,
+    "selected_score": 5.08158,
+    "legacy_score": 5.233586,
+    "filled_actual_from_history": 0,
+    "settlement_history_city_count": 30,
+    "legacy_archive_samples": 54
+  },
+  "source": "artifacts\\probability_calibration\\default.json"
+}

artifacts/probability_calibration/default.json

@@ -1,20 +1,20 @@
 {
-  "version": "emos-20260418192717",
+  "version": "emos-auto-20260421122743",
-  "trained_at": "2026-04-18T19:27:17.348931+00:00",
+  "trained_at": "2026-04-21T12:28:05.031165+00:00",
   "global": {
     "mu": {
-      "intercept": 2.60074353,
+      "intercept": 0.34330438,
-      "raw_mu_coef": 1.97765258,
+      "raw_mu_coef": 0.48749629,
-      "deb_coef": -1.37457626,
+      "deb_coef": 0.45733479,
-      "ens_median_coef": 0.10693472,
+      "ens_median_coef": 0.05955297,
-      "max_so_far_gap_coef": -0.58256593
+      "max_so_far_gap_coef": -0.104935
     },
     "sigma": {
-      "intercept": 0.6373609,
+      "intercept": -0.62765582,
-      "raw_sigma_coef": -1.41537225,
+      "raw_sigma_coef": 0.38492261,
-      "spread_coef": 2.12997577,
+      "spread_coef": 0.29095921,
-      "peak_flag_coef": -0.08633203,
+      "peak_flag_coef": 0.33966327,
-      "max_so_far_gap_coef": -0.00716383
+      "max_so_far_gap_coef": 0.08357376
     }
   },
   "sigma_constraints": {
@@ -29,87 +29,333 @@
     "max_bucket_brier_increase": 0.05
   },
   "blending": {
-    "alpha_mu": 0.05,
+    "alpha_mu": 0.25,
-    "alpha_sigma": 0.7
+    "alpha_sigma": 1.0
   },
   "cities": {
-    "tokyo": {
+    "istanbul": {
-      "samples": 5,
+      "samples": 20,
-      "mu_bias": 2.101101,
+      "mu_bias": -0.438949,
-      "sigma_scale": 2.0,
+      "sigma_scale": 0.553793,
-      "confidence": 0.625
+      "confidence": 1.0
+    },
+    "buenos aires": {
+      "samples": 22,
+      "mu_bias": -1.044971,
+      "sigma_scale": 1.389753,
+      "confidence": 1.0
     },
     "ankara": {
-      "samples": 6,
+      "samples": 23,
-      "mu_bias": -0.523145,
+      "mu_bias": 0.067406,
-      "sigma_scale": 1.139849,
+      "sigma_scale": 1.270298,
-      "confidence": 0.75
+      "confidence": 1.0
     },
-    "istanbul": {
+    "london": {
-      "samples": 3,
+      "samples": 22,
-      "mu_bias": -0.285183,
+      "mu_bias": -0.010833,
-      "sigma_scale": 0.550171,
-      "confidence": 0.375
-    },
-    "hong kong": {
-      "samples": 7,
-      "mu_bias": 3.234676,
       "sigma_scale": 2.0,
-      "confidence": 0.875
+      "confidence": 1.0
     },
-    "taipei": {
+    "paris": {
+      "samples": 22,
+      "mu_bias": -0.21677,
+      "sigma_scale": 1.095669,
+      "confidence": 1.0
+    },
+    "new york": {
+      "samples": 21,
+      "mu_bias": 1.072189,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "lagos": {
+      "samples": 10,
+      "mu_bias": -0.092732,
+      "sigma_scale": 1.558646,
+      "confidence": 1.0
+    },
+    "madrid": {
+      "samples": 22,
+      "mu_bias": -0.627862,
+      "sigma_scale": 1.761239,
+      "confidence": 1.0
+    },
+    "toronto": {
+      "samples": 22,
+      "mu_bias": -0.424101,
+      "sigma_scale": 1.326675,
+      "confidence": 1.0
+    },
+    "wellington": {
+      "samples": 22,
+      "mu_bias": -0.347796,
+      "sigma_scale": 0.893459,
+      "confidence": 1.0
+    },
+    "los angeles": {
+      "samples": 20,
+      "mu_bias": -0.659983,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "san francisco": {
+      "samples": 20,
+      "mu_bias": 0.941818,
+      "sigma_scale": 1.979403,
+      "confidence": 1.0
+    },
+    "aurora": {
+      "samples": 20,
+      "mu_bias": 2.040009,
+      "sigma_scale": 0.982492,
+      "confidence": 1.0
+    },
+    "austin": {
+      "samples": 20,
+      "mu_bias": 0.990023,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "houston": {
+      "samples": 20,
+      "mu_bias": 0.188787,
+      "sigma_scale": 1.819594,
+      "confidence": 1.0
+    },
+    "chicago": {
+      "samples": 21,
+      "mu_bias": 2.285098,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "dallas": {
+      "samples": 21,
+      "mu_bias": 0.258095,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "atlanta": {
+      "samples": 22,
+      "mu_bias": -2.093105,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "seattle": {
+      "samples": 21,
+      "mu_bias": -0.982188,
+      "sigma_scale": 1.695172,
+      "confidence": 1.0
+    },
+    "sao paulo": {
+      "samples": 22,
+      "mu_bias": -0.786242,
+      "sigma_scale": 1.427037,
+      "confidence": 1.0
+    },
+    "munich": {
+      "samples": 22,
+      "mu_bias": 0.096559,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "milan": {
+      "samples": 23,
+      "mu_bias": -0.045038,
+      "sigma_scale": 1.626331,
+      "confidence": 1.0
+    },
+    "warsaw": {
+      "samples": 23,
+      "mu_bias": -0.50853,
+      "sigma_scale": 1.239146,
+      "confidence": 1.0
+    },
+    "helsinki": {
+      "samples": 17,
+      "mu_bias": 0.442247,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "amsterdam": {
+      "samples": 17,
+      "mu_bias": 0.129889,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "panama city": {
+      "samples": 17,
+      "mu_bias": -0.040792,
+      "sigma_scale": 1.393259,
+      "confidence": 1.0
+    },
+    "cape town": {
+      "samples": 10,
+      "mu_bias": 0.507293,
+      "sigma_scale": 1.420004,
+      "confidence": 1.0
+    },
+    "miami": {
+      "samples": 22,
+      "mu_bias": -2.736316,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "manila": {
       "samples": 5,
-      "mu_bias": 4.731246,
+      "mu_bias": 1.772229,
       "sigma_scale": 2.0,
       "confidence": 0.625
     },
-    "busan": {
+    "singapore": {
-      "samples": 3,
+      "samples": 22,
-      "mu_bias": 0.177772,
+      "mu_bias": 0.289639,
-      "sigma_scale": 2.0,
+      "sigma_scale": 1.124006,
-      "confidence": 0.375
+      "confidence": 1.0
+    },
+    "tokyo": {
+      "samples": 22,
+      "mu_bias": -0.332129,
+      "sigma_scale": 0.729483,
+      "confidence": 1.0
     },
     "seoul": {
-      "samples": 3,
+      "samples": 22,
-      "mu_bias": -3.070325,
+      "mu_bias": 1.6533,
-      "sigma_scale": 1.462196,
+      "sigma_scale": 0.82648,
-      "confidence": 0.375
+      "confidence": 1.0
     },
-    "milan": {
+    "busan": {
-      "samples": 3,
+      "samples": 16,
-      "mu_bias": -4.249866,
+      "mu_bias": 0.343239,
-      "sigma_scale": 2.0,
+      "sigma_scale": 1.886868,
-      "confidence": 0.375
+      "confidence": 1.0
+    },
+    "mexico city": {
+      "samples": 20,
+      "mu_bias": 0.087122,
+      "sigma_scale": 1.108884,
+      "confidence": 1.0
     },
     "shanghai": {
-      "samples": 3,
+      "samples": 23,
-      "mu_bias": 1.080039,
+      "mu_bias": 0.348022,
-      "sigma_scale": 1.638119,
+      "sigma_scale": 0.810527,
-      "confidence": 0.375
+      "confidence": 1.0
     },
-    "warsaw": {
+    "beijing": {
-      "samples": 3,
+      "samples": 20,
-      "mu_bias": -1.68611,
+      "mu_bias": -0.288176,
-      "sigma_scale": 1.38114,
+      "sigma_scale": 0.952972,
-      "confidence": 0.375
+      "confidence": 1.0
+    },
+    "shenzhen": {
+      "samples": 21,
+      "mu_bias": -0.006114,
+      "sigma_scale": 0.728494,
+      "confidence": 1.0
+    },
+    "wuhan": {
+      "samples": 20,
+      "mu_bias": -0.771154,
+      "sigma_scale": 1.30329,
+      "confidence": 1.0
+    },
+    "chengdu": {
+      "samples": 20,
+      "mu_bias": 0.445906,
+      "sigma_scale": 0.796989,
+      "confidence": 1.0
+    },
+    "chongqing": {
+      "samples": 20,
+      "mu_bias": -0.021301,
+      "sigma_scale": 1.106566,
+      "confidence": 1.0
+    },
+    "hong kong": {
+      "samples": 24,
+      "mu_bias": 0.37527,
+      "sigma_scale": 0.517756,
+      "confidence": 1.0
+    },
+    "taipei": {
+      "samples": 23,
+      "mu_bias": 0.600671,
+      "sigma_scale": 0.5,
+      "confidence": 1.0
+    },
+    "guangzhou": {
+      "samples": 6,
+      "mu_bias": 0.7082,
+      "sigma_scale": 1.975103,
+      "confidence": 0.75
+    },
+    "lau fau shan": {
+      "samples": 19,
+      "mu_bias": 0.41371,
+      "sigma_scale": 0.656578,
+      "confidence": 1.0
+    },
+    "kuala lumpur": {
+      "samples": 16,
+      "mu_bias": 0.499197,
+      "sigma_scale": 1.361171,
+      "confidence": 1.0
+    },
+    "jakarta": {
+      "samples": 16,
+      "mu_bias": 0.293537,
+      "sigma_scale": 0.751498,
+      "confidence": 1.0
+    },
+    "lucknow": {
+      "samples": 22,
+      "mu_bias": -1.593437,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "karachi": {
+      "samples": 5,
+      "mu_bias": -0.035595,
+      "sigma_scale": 0.588612,
+      "confidence": 0.625
+    },
+    "jeddah": {
+      "samples": 10,
+      "mu_bias": 0.516879,
+      "sigma_scale": 2.0,
+      "confidence": 1.0
+    },
+    "moscow": {
+      "samples": 20,
+      "mu_bias": -0.609602,
+      "sigma_scale": 0.783723,
+      "confidence": 1.0
+    },
+    "tel aviv": {
+      "samples": 22,
+      "mu_bias": 0.436837,
+      "sigma_scale": 1.46524,
+      "confidence": 1.0
     }
   },
   "metrics": {
-    "sample_count": 74,
+    "sample_count": 978,
-    "mean_crps": 3.181501,
+    "mean_crps": 1.330733,
-    "legacy_mean_crps": 3.474108,
+    "legacy_mean_crps": 1.376936,
-    "legacy_mean_mae": 3.679324,
+    "legacy_mean_mae": 1.589438,
-    "legacy_bucket_hit_rate": 0.5,
+    "legacy_bucket_hit_rate": 0.43865,
-    "legacy_bucket_brier": 0.855395,
+    "legacy_bucket_brier": 0.824571,
-    "selected_mean_crps": 3.322469,
+    "selected_mean_crps": 1.311763,
-    "selected_mean_mae": 3.618254,
+    "selected_mean_mae": 1.553658,
-    "selected_bucket_hit_rate": 0.5,
+    "selected_bucket_hit_rate": 0.432515,
-    "selected_bucket_brier": 0.863048,
+    "selected_bucket_brier": 0.789817,
-    "selected_score": 5.08158,
+    "selected_score": 2.910718,
-    "legacy_score": 5.233586,
+    "legacy_score": 2.996333,
     "filled_actual_from_history": 0,
     "settlement_history_city_count": 30,
     "legacy_archive_samples": 54
   },
-  "source": "artifacts\\probability_calibration\\default.json"
+  "source": "artifacts\\local_runtime\\probability_calibration\\candidates\\emos-auto-20260421122743\\default.json"
 }

frontend/components/dashboard/Dashboard.module.css

@@ -1075,6 +1075,72 @@
   border: 1px solid rgba(251, 113, 133, 0.26);
 }
 
+.root :global(.prob-price-card) {
+  display: grid;
+  gap: 8px;
+  padding: 10px;
+  border: 1px solid rgba(34, 211, 238, 0.16);
+  border-radius: 8px;
+  background: rgba(8, 20, 32, 0.52);
+}
+
+.root :global(.prob-price-head) {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 10px;
+}
+
+.root :global(.prob-price-head span) {
+  color: var(--cyan);
+  font-size: 11px;
+  font-weight: 800;
+  text-transform: uppercase;
+  letter-spacing: 0.04em;
+}
+
+.root :global(.prob-price-head strong) {
+  color: var(--text-primary);
+  font-size: 13px;
+  font-weight: 800;
+}
+
+.root :global(.prob-price-grid) {
+  display: grid;
+  grid-template-columns: repeat(4, minmax(0, 1fr));
+  gap: 8px;
+}
+
+.root :global(.prob-price-grid > div) {
+  display: grid;
+  gap: 3px;
+  min-width: 0;
+  padding: 8px;
+  border: 1px solid rgba(148, 163, 184, 0.12);
+  border-radius: 8px;
+  background: rgba(15, 23, 42, 0.45);
+}
+
+.root :global(.prob-price-grid span),
+.root :global(.prob-price-grid em),
+.root :global(.prob-price-card p) {
+  color: var(--text-muted);
+  font-size: 11px;
+  font-style: normal;
+  line-height: 1.35;
+}
+
+.root :global(.prob-price-grid strong) {
+  color: var(--text-primary);
+  font-size: 13px;
+  font-weight: 800;
+  font-variant-numeric: tabular-nums;
+}
+
+.root :global(.prob-price-card p) {
+  margin: 0;
+}
+
 .root :global(.prob-model-hint) {
   display: flex;
   align-items: center;
@@ -2466,6 +2532,15 @@
   .root :global(.marker-name) {
     font-size: 11px;
   }
+
+  .root :global(.prob-price-grid) {
+    grid-template-columns: repeat(2, minmax(0, 1fr));
+  }
+
+  .root :global(.prob-price-head) {
+    align-items: flex-start;
+    flex-direction: column;
+  }
 }
 
 /* ── History Modal ── */

frontend/components/dashboard/PanelSections.tsx

@@ -278,6 +278,29 @@ function normalizeMarketProbability(value?: number | null) {
   return Math.max(0, Math.min(1, numeric));
 }
 
+function normalizeSignedProbability(value?: number | null) {
+  if (value == null) return null;
+  const numeric = Number(value);
+  if (!Number.isFinite(numeric)) return null;
+  if (Math.abs(numeric) > 1) return numeric / 100;
+  return numeric;
+}
+
+function formatSignedPercent(value?: number | null, digits = 1) {
+  const normalized = normalizeSignedProbability(value);
+  if (normalized == null) return "--";
+  const percent = normalized * 100;
+  const sign = percent > 0 ? "+" : "";
+  return `${sign}${percent.toFixed(digits)}%`;
+}
+
+function formatKellyFraction(value?: number | null) {
+  const normalized = normalizeSignedProbability(value);
+  if (normalized == null) return "--";
+  if (normalized <= 0) return "0%";
+  return `${(normalized * 100).toFixed(1)}%`;
+}
+
 function getMarketTopBuckets(scan?: MarketScan | null) {
   const buckets = Array.isArray(scan?.top_buckets) ? scan.top_buckets : [];
   if (!buckets.length) return [];
@@ -657,6 +680,38 @@ export function ProbabilityDistribution({
   const topProbabilityLabel = topProbability
     ? topProbability.label || `${topProbability.value}${detail.temp_symbol}`
     : null;
+  const priceAnalysis = marketScan?.price_analysis;
+  const yesPriceView = priceAnalysis?.yes;
+  const noPriceView = priceAnalysis?.no;
+  const preferredPriceView =
+    priceAnalysis?.best_side === "no" ? noPriceView : yesPriceView;
+  const preferredSideLabel =
+    priceAnalysis?.best_side === "no"
+      ? locale === "en-US"
+        ? "NO"
+        : "NO"
+      : locale === "en-US"
+        ? "YES"
+        : "YES";
+  const hasPriceAnalysis =
+    Boolean(priceAnalysis?.available) &&
+    (yesPriceView?.ask != null || noPriceView?.ask != null);
+  const lockEdge = normalizeSignedProbability(priceAnalysis?.lock?.edge);
+  const lockAvailable = Boolean(priceAnalysis?.lock?.available && lockEdge != null);
+  const quoteSource =
+    marketScan?.yes_token?.quote_source ||
+    marketScan?.no_token?.quote_source ||
+    null;
+  const quoteAgeMs =
+    marketScan?.yes_token?.quote_age_ms ?? marketScan?.no_token?.quote_age_ms;
+  const quoteSourceLabel =
+    quoteSource === "polymarket_ws"
+      ? locale === "en-US"
+        ? `WS live${quoteAgeMs != null ? ` · ${Math.max(0, Math.round(Number(quoteAgeMs) / 1000))}s` : ""}`
+        : `WS 实时${quoteAgeMs != null ? ` · ${Math.max(0, Math.round(Number(quoteAgeMs) / 1000))}秒` : ""}`
+      : locale === "en-US"
+        ? "CLOB fallback"
+        : "CLOB 兜底";
 
   return (
     <section className="prob-section">
@@ -702,6 +757,63 @@ export function ProbabilityDistribution({
                 : `市场仅作参考：该温度桶 ${marketYesText}`}
           </div>
         )}
+        {hasPriceAnalysis && (
+          <div className="prob-price-card">
+            <div className="prob-price-head">
+              <span>
+                {locale === "en-US" ? "Price reference" : "价格参考"}
+              </span>
+              <strong>
+                {preferredPriceView?.edge != null
+                  ? locale === "en-US"
+                    ? `${preferredSideLabel} edge ${formatSignedPercent(preferredPriceView.edge)}`
+                    : `${preferredSideLabel} 边际 ${formatSignedPercent(preferredPriceView.edge)}`
+                  : locale === "en-US"
+                    ? "Waiting for executable quote"
+                    : "等待可执行报价"}
+              </strong>
+            </div>
+            <div className="prob-price-grid">
+              <div>
+                <span>YES ask</span>
+                <strong>{toPriceCents(yesPriceView?.ask) || "--"}</strong>
+                <em>{formatSignedPercent(yesPriceView?.edge)}</em>
+              </div>
+              <div>
+                <span>NO ask</span>
+                <strong>{toPriceCents(noPriceView?.ask) || "--"}</strong>
+                <em>{formatSignedPercent(noPriceView?.edge)}</em>
+              </div>
+              <div>
+                <span>{locale === "en-US" ? "1/4 Kelly" : "1/4 Kelly"}</span>
+                <strong>{formatKellyFraction(preferredPriceView?.quarter_kelly)}</strong>
+                <em>{locale === "en-US" ? "cap required" : "需限额"}</em>
+              </div>
+              <div>
+                <span>{locale === "en-US" ? "Lock" : "锁价"}</span>
+                <strong>
+                  {lockAvailable
+                    ? formatSignedPercent(lockEdge)
+                    : locale === "en-US"
+                      ? "None"
+                      : "无"}
+                </strong>
+                <em>
+                  {priceAnalysis?.lock?.ask_sum != null
+                    ? locale === "en-US"
+                      ? `sum ${toPriceCents(priceAnalysis.lock.ask_sum)}`
+                      : `合计 ${toPriceCents(priceAnalysis.lock.ask_sum)}`
+                    : "--"}
+                </em>
+              </div>
+            </div>
+            <p>
+              {locale === "en-US"
+                ? `Read-only sizing reference from executable quotes; it does not place orders. Source: ${quoteSourceLabel}.`
+                : `基于可执行盘口的只读仓位参考；系统不会下单。来源：${quoteSourceLabel}。`}
+            </p>
+          </div>
+        )}
         {modelVoteHint && (
           <div className="prob-model-hint">
             <span>

frontend/lib/dashboard-types.ts

@@ -289,6 +289,8 @@ export interface MarketToken {
   sell_price?: number | null;
   midpoint?: number | null;
   last_trade_price?: number | null;
+  quote_source?: string | null;
+  quote_age_ms?: number | null;
 }
 
 export interface MarketPrimary {
@@ -314,11 +316,42 @@ export interface MarketTopBucket {
   yes_sell?: number | null;
   no_buy?: number | null;
   no_sell?: number | null;
+  quote_source?: string | null;
+  quote_age_ms?: number | null;
   slug?: string | null;
   question?: string | null;
   is_primary?: boolean;
 }
 
+export interface MarketPriceSide {
+  side?: "yes" | "no" | string | null;
+  model_probability?: number | null;
+  ask?: number | null;
+  bid?: number | null;
+  edge?: number | null;
+  edge_percent?: number | null;
+  kelly_fraction?: number | null;
+  quarter_kelly?: number | null;
+}
+
+export interface MarketPriceAnalysis {
+  available?: boolean;
+  source?: string | null;
+  model_probability?: number | null;
+  yes?: MarketPriceSide | null;
+  no?: MarketPriceSide | null;
+  best_side?: "yes" | "no" | string | null;
+  lock?: {
+    available?: boolean;
+    ask_sum?: number | null;
+    edge?: number | null;
+  } | null;
+  sell_side?: {
+    bid_sum?: number | null;
+    edge?: number | null;
+  } | null;
+}
+
 export interface MarketScan {
   available?: boolean;
   reason?: string | null;
@@ -343,6 +376,7 @@ export interface MarketScan {
   last_trade_price?: number | null;
   liquidity?: number | null;
   volume?: number | null;
+  price_analysis?: MarketPriceAnalysis | null;
   sparkline?: number[];
   top_buckets?: MarketTopBucket[] | null;
   recent_trades?: unknown[];

requirements.txt

@@ -9,3 +9,4 @@ lightgbm
 web3
 fastapi
 uvicorn
+websockets

src/data_collection/polymarket_readonly.py

@@ -23,6 +23,7 @@ import httpx
 from loguru import logger
 
 from src.data_collection.city_registry import ALIASES, CITY_REGISTRY
+from src.data_collection.polymarket_ws_cache import PolymarketWsQuoteCache
 
 try:
     from py_clob_client.client import ClobClient  # type: ignore
@@ -173,6 +174,16 @@ def _extract_price(value: Any) -> Optional[float]:
     return None
 
 
+def _clamp_probability(value: Optional[float]) -> Optional[float]:
+    if value is None:
+        return None
+    if value < 0.0:
+        return 0.0
+    if value > 1.0:
+        return 1.0
+    return value
+
+
 def _extract_iso_date(value: Any) -> Optional[str]:
     if not value:
         return None
@@ -394,6 +405,7 @@ class PolymarketReadOnlyLayer:
         self._lock = threading.Lock()
         self._clob_client: Any = None
         self._clob_unavailable_reason: Optional[str] = None
+        self._ws_quote_cache = PolymarketWsQuoteCache.from_env()
 
     def build_market_scan(
         self,
@@ -430,6 +442,7 @@ class PolymarketReadOnlyLayer:
             "last_trade_price": None,
             "liquidity": None,
             "volume": None,
+            "price_analysis": None,
             "sparkline": fallback_sparkline or [],
             "top_buckets": [],
             "recent_trades": [],
@@ -570,6 +583,8 @@ class PolymarketReadOnlyLayer:
             "sell_price": _extract_price(yes_prices.get("sell")),
             "midpoint": _extract_price(yes_prices.get("midpoint")),
             "last_trade_price": _extract_price(yes_prices.get("last_trade_price")),
+            "quote_source": yes_prices.get("quote_source"),
+            "quote_age_ms": _safe_int(yes_prices.get("quote_age_ms"), 0),
             "book": yes_prices.get("book"),
         }
         no_payload = {
@@ -580,8 +595,17 @@ class PolymarketReadOnlyLayer:
             "sell_price": _extract_price(no_prices.get("sell")),
             "midpoint": _extract_price(no_prices.get("midpoint")),
             "last_trade_price": _extract_price(no_prices.get("last_trade_price")),
+            "quote_source": no_prices.get("quote_source"),
+            "quote_age_ms": _safe_int(no_prices.get("quote_age_ms"), 0),
             "book": no_prices.get("book"),
         }
+        price_analysis = self._build_price_analysis(
+            model_probability=model_probability,
+            yes_buy=_extract_price(yes_payload.get("buy_price")),
+            yes_sell=_extract_price(yes_payload.get("sell_price")),
+            no_buy=_extract_price(no_payload.get("buy_price")),
+            no_sell=_extract_price(no_payload.get("sell_price")),
+        )
 
         sparkline_values: List[float] = []
         for candidate in (
@@ -617,12 +641,15 @@ class PolymarketReadOnlyLayer:
                 "last_trade_price": last_trade_price,
                 "liquidity": liquidity,
                 "volume": volume,
+                "price_analysis": price_analysis,
                 "sparkline": sparkline_values,
                 "top_buckets": top_buckets,
                 "all_buckets": all_buckets,
-                "websocket": {
+            "websocket": {
-                    "market_url": market_url,
+                "enabled": self._ws_quote_cache.enabled,
-                    "asset_ids": [
+                "status": self._ws_quote_cache.status(),
+                "market_url": market_url,
+                "asset_ids": [
                         token
                         for token in [
                             yes_payload.get("token_id"),
@@ -636,6 +663,112 @@ class PolymarketReadOnlyLayer:
         )
         return scan
 
+    def _build_price_analysis(
+        self,
+        *,
+        model_probability: Optional[float],
+        yes_buy: Optional[float],
+        yes_sell: Optional[float],
+        no_buy: Optional[float],
+        no_sell: Optional[float],
+    ) -> Dict[str, Any]:
+        """Build read-only market price diagnostics.
+
+        Polymarket CLOB naming is from the user's perspective:
+        BUY is the executable ask to buy that outcome, SELL is the executable bid.
+        Kelly here is a sizing reference only; no order execution is performed.
+        """
+        p_yes = _clamp_probability(_safe_float(model_probability))
+        p_no = _clamp_probability(1.0 - p_yes if p_yes is not None else None)
+        yes_ask = _clamp_probability(_safe_float(yes_buy))
+        no_ask = _clamp_probability(_safe_float(no_buy))
+        yes_bid = _clamp_probability(_safe_float(yes_sell))
+        no_bid = _clamp_probability(_safe_float(no_sell))
+
+        yes = self._build_side_price_analysis("yes", p_yes, yes_ask, yes_bid)
+        no = self._build_side_price_analysis("no", p_no, no_ask, no_bid)
+
+        ask_sum = None
+        lock_edge = None
+        lock_available = False
+        if yes_ask is not None and no_ask is not None:
+            ask_sum = yes_ask + no_ask
+            lock_edge = 1.0 - ask_sum
+            lock_available = lock_edge > 0
+
+        bid_sum = None
+        sell_side_edge = None
+        if yes_bid is not None and no_bid is not None:
+            bid_sum = yes_bid + no_bid
+            sell_side_edge = bid_sum - 1.0
+
+        best_side = None
+        side_rows = [
+            row
+            for row in [yes, no]
+            if isinstance(row.get("edge"), (int, float))
+            and isinstance(row.get("kelly_fraction"), (int, float))
+            and row.get("kelly_fraction") > 0
+        ]
+        if side_rows:
+            best_side = max(
+                side_rows,
+                key=lambda row: (
+                    float(row.get("edge") or 0.0),
+                    float(row.get("kelly_fraction") or 0.0),
+                ),
+            ).get("side")
+
+        return {
+            "available": any(
+                value is not None
+                for value in (yes_ask, no_ask, yes_bid, no_bid, p_yes)
+            ),
+            "source": "polymarket_clob_orderbook",
+            "model_probability": p_yes,
+            "yes": yes,
+            "no": no,
+            "best_side": best_side,
+            "lock": {
+                "available": lock_available,
+                "ask_sum": ask_sum,
+                "edge": lock_edge,
+            },
+            "sell_side": {
+                "bid_sum": bid_sum,
+                "edge": sell_side_edge,
+            },
+        }
+
+    def _build_side_price_analysis(
+        self,
+        side: str,
+        probability: Optional[float],
+        ask: Optional[float],
+        bid: Optional[float],
+    ) -> Dict[str, Any]:
+        edge = None
+        kelly_fraction = None
+        if probability is not None and ask is not None:
+            edge = probability - ask
+            if 0.0 < ask < 1.0:
+                kelly_fraction = edge / (1.0 - ask)
+
+        return {
+            "side": side,
+            "model_probability": probability,
+            "ask": ask,
+            "bid": bid,
+            "edge": edge,
+            "edge_percent": edge * 100.0 if edge is not None else None,
+            "kelly_fraction": kelly_fraction,
+            "quarter_kelly": (
+                max(0.0, kelly_fraction) / 4.0
+                if kelly_fraction is not None
+                else None
+            ),
+        }
+
     def _market_trade_state(self, market: Dict[str, Any]) -> Dict[str, Any]:
         active = _safe_bool(market.get("active"))
         closed_raw = _safe_bool(market.get("closed"))
@@ -1193,6 +1326,11 @@ class PolymarketReadOnlyLayer:
         if not token_id:
             return {}
 
+        self._ws_quote_cache.subscribe([token_id])
+        ws_data = self._ws_quote_cache.get_market_data(token_id)
+        if ws_data is not None:
+            return ws_data
+
         now = time.time()
         with self._lock:
             cached = self._price_cache.get(token_id)
@@ -1473,6 +1611,8 @@ class PolymarketReadOnlyLayer:
                         "yes_sell": yes_sell,
                         "no_buy": no_buy,
                         "no_sell": no_sell,
+                        "quote_source": yes_prices.get("quote_source"),
+                        "quote_age_ms": _safe_int(yes_prices.get("quote_age_ms"), 0),
                         "slug": market_slug or None,
                         "question": market.get("question") or market.get("title"),
                         "is_primary": bool(

src/data_collection/polymarket_ws_cache.py

@@ -0,0 +1,404 @@
+"""
+Read-only Polymarket market WebSocket quote cache.
+
+The cache subscribes to public market-channel asset ids and stores executable
+best bid / ask updates. It is deliberately optional: callers should keep REST
+or CLOB polling as a fallback when the WebSocket client is unavailable.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import math
+import os
+import threading
+import time
+from typing import Any, Dict, Iterable, Optional, Set
+
+from loguru import logger
+
+
+def _safe_float(value: Any) -> Optional[float]:
+    if value is None:
+        return None
+    try:
+        if isinstance(value, str):
+            value = value.strip()
+            if not value:
+                return None
+        numeric = float(value)
+        if math.isnan(numeric) or math.isinf(numeric):
+            return None
+        return numeric
+    except Exception:
+        return None
+
+
+def _env_bool(name: str, default: bool = False) -> bool:
+    raw = os.getenv(name)
+    if raw is None:
+        return default
+    return raw.strip().lower() in {"1", "true", "yes", "on"}
+
+
+class PolymarketWsQuoteCache:
+    def __init__(
+        self,
+        *,
+        enabled: bool = False,
+        endpoint: Optional[str] = None,
+        quote_ttl_sec: int = 8,
+        max_assets: int = 256,
+        reconnect_delay_sec: float = 3.0,
+    ) -> None:
+        self.enabled = enabled
+        self.endpoint = (
+            endpoint
+            or os.getenv(
+                "POLYMARKET_WS_MARKET_URL",
+                "wss://ws-subscriptions-clob.polymarket.com/ws/market",
+            )
+            or ""
+        ).strip()
+        self.quote_ttl_sec = max(1, int(quote_ttl_sec or 8))
+        self.max_assets = max(1, int(max_assets or 256))
+        self.reconnect_delay_sec = max(0.5, float(reconnect_delay_sec or 3.0))
+
+        self._desired_assets: Set[str] = set()
+        self._quotes: Dict[str, Dict[str, Any]] = {}
+        self._lock = threading.Lock()
+        self._thread: Optional[threading.Thread] = None
+        self._stop_event = threading.Event()
+        self._started = False
+        self._last_error: Optional[str] = None
+        self._last_connected_at: Optional[float] = None
+        self._last_message_at: Optional[float] = None
+
+    @classmethod
+    def from_env(cls) -> "PolymarketWsQuoteCache":
+        return cls(
+            enabled=_env_bool("POLYMARKET_WS_PRICE_ENABLED", False),
+            endpoint=os.getenv("POLYMARKET_WS_MARKET_URL"),
+            quote_ttl_sec=int(os.getenv("POLYMARKET_WS_QUOTE_TTL_SEC", "8")),
+            max_assets=int(os.getenv("POLYMARKET_WS_MAX_ASSETS", "256")),
+            reconnect_delay_sec=float(
+                os.getenv("POLYMARKET_WS_RECONNECT_DELAY_SEC", "3")
+            ),
+        )
+
+    def start(self) -> None:
+        if not self.enabled or not self.endpoint:
+            return
+        with self._lock:
+            if self._started:
+                return
+            self._started = True
+        self._thread = threading.Thread(
+            target=self._thread_main,
+            name="polymarket-ws-quotes",
+            daemon=True,
+        )
+        self._thread.start()
+
+    def stop(self) -> None:
+        self._stop_event.set()
+
+    def subscribe(self, asset_ids: Iterable[Any]) -> None:
+        if not self.enabled:
+            return
+        normalized = []
+        for asset_id in asset_ids:
+            text = str(asset_id or "").strip()
+            if text:
+                normalized.append(text)
+        if not normalized:
+            return
+
+        with self._lock:
+            remaining = self.max_assets - len(self._desired_assets)
+            for asset_id in normalized:
+                if asset_id in self._desired_assets:
+                    continue
+                if remaining <= 0:
+                    break
+                self._desired_assets.add(asset_id)
+                remaining -= 1
+        self.start()
+
+    def get_market_data(self, asset_id: Any) -> Optional[Dict[str, Any]]:
+        quote = self.get_quote(asset_id)
+        if not quote:
+            return None
+
+        best_bid = _safe_float(quote.get("best_bid"))
+        best_ask = _safe_float(quote.get("best_ask"))
+        if best_bid is None and best_ask is None:
+            return None
+
+        midpoint = None
+        if best_bid is not None and best_ask is not None:
+            midpoint = (best_bid + best_ask) / 2.0
+
+        age_ms = int((time.time() - float(quote.get("t") or time.time())) * 1000)
+        return {
+            "buy": best_ask,
+            "sell": best_bid,
+            "midpoint": midpoint,
+            "last_trade_price": _safe_float(quote.get("last_trade_price")),
+            "book": {
+                "best_bid": best_bid,
+                "best_ask": best_ask,
+                "bid_levels": [[best_bid, 0.0]] if best_bid is not None else [],
+                "ask_levels": [[best_ask, 0.0]] if best_ask is not None else [],
+            },
+            "book_liquidity": None,
+            "quote_source": "polymarket_ws",
+            "quote_age_ms": age_ms,
+        }
+
+    def get_quote(self, asset_id: Any) -> Optional[Dict[str, Any]]:
+        text = str(asset_id or "").strip()
+        if not text:
+            return None
+        now = time.time()
+        with self._lock:
+            quote = self._quotes.get(text)
+            if not quote:
+                return None
+            if now - float(quote.get("t") or 0.0) > self.quote_ttl_sec:
+                return None
+            return dict(quote)
+
+    def status(self) -> Dict[str, Any]:
+        with self._lock:
+            return {
+                "enabled": self.enabled,
+                "started": self._started,
+                "endpoint": self.endpoint,
+                "asset_count": len(self._desired_assets),
+                "quote_count": len(self._quotes),
+                "last_error": self._last_error,
+                "last_connected_at": self._last_connected_at,
+                "last_message_at": self._last_message_at,
+            }
+
+    def _thread_main(self) -> None:
+        try:
+            asyncio.run(self._run_forever())
+        except Exception as exc:  # pragma: no cover - defensive thread guard
+            with self._lock:
+                self._last_error = str(exc)
+            logger.warning(f"Polymarket WS quote cache stopped: {exc}")
+
+    async def _run_forever(self) -> None:
+        try:
+            import websockets  # type: ignore
+        except Exception as exc:
+            with self._lock:
+                self._last_error = f"websockets import failed: {exc}"
+            logger.warning(self._last_error)
+            return
+
+        while not self._stop_event.is_set():
+            try:
+                async with websockets.connect(
+                    self.endpoint,
+                    ping_interval=None,
+                    close_timeout=2,
+                ) as ws:
+                    with self._lock:
+                        self._last_connected_at = time.time()
+                        self._last_error = None
+                    subscribed: Set[str] = set()
+                    last_ping = 0.0
+
+                    while not self._stop_event.is_set():
+                        desired = self._snapshot_assets()
+                        missing = desired - subscribed
+                        if missing:
+                            await self._send_subscription(
+                                ws,
+                                missing,
+                                initial=not subscribed,
+                            )
+                            subscribed.update(missing)
+
+                        now = time.time()
+                        if now - last_ping >= 10:
+                            await ws.send(json.dumps({}))
+                            last_ping = now
+
+                        try:
+                            raw = await asyncio.wait_for(ws.recv(), timeout=1.0)
+                        except asyncio.TimeoutError:
+                            continue
+                        self._handle_message(raw)
+            except Exception as exc:
+                with self._lock:
+                    self._last_error = str(exc)
+                logger.warning(f"Polymarket WS reconnecting after error: {exc}")
+                await asyncio.sleep(self.reconnect_delay_sec)
+
+    def _snapshot_assets(self) -> Set[str]:
+        with self._lock:
+            return set(self._desired_assets)
+
+    async def _send_subscription(
+        self,
+        ws: Any,
+        asset_ids: Iterable[str],
+        *,
+        initial: bool,
+    ) -> None:
+        batch = [asset_id for asset_id in asset_ids if asset_id]
+        if not batch:
+            return
+        payload = {
+            "assets_ids": batch,
+            "custom_feature_enabled": True,
+        }
+        if initial:
+            payload["type"] = "market"
+        else:
+            payload["operation"] = "subscribe"
+        await ws.send(json.dumps(payload))
+
+    def _handle_message(self, raw: Any) -> None:
+        if raw in (None, "", "PONG"):
+            return
+        try:
+            payload = json.loads(raw) if isinstance(raw, str) else raw
+        except Exception:
+            return
+
+        if isinstance(payload, list):
+            for item in payload:
+                self._handle_event(item)
+            return
+        self._handle_event(payload)
+
+    def _handle_event(self, event: Any) -> None:
+        if not isinstance(event, dict):
+            return
+        if "event_type" in event:
+            event_type = str(event.get("event_type") or "").strip().lower()
+        else:
+            event_type = str(event.get("type") or "").strip().lower()
+
+        if event_type in {
+            "best_bid_ask",
+            "price_change",
+            "book",
+            "last_trade_price",
+        }:
+            self._handle_quote_event(event_type, event)
+
+    def _handle_quote_event(self, event_type: str, event: Dict[str, Any]) -> None:
+        candidates = (
+            event.get("price_changes")
+            or event.get("changes")
+            or event.get("assets")
+            or event.get("data")
+        )
+        if isinstance(candidates, list):
+            for item in candidates:
+                if isinstance(item, dict):
+                    self._upsert_quote(event_type, item, parent=event)
+            return
+        self._upsert_quote(event_type, event, parent=event)
+
+    def _upsert_quote(
+        self,
+        event_type: str,
+        item: Dict[str, Any],
+        *,
+        parent: Dict[str, Any],
+    ) -> None:
+        asset_id = str(
+            item.get("asset_id")
+            or item.get("assetId")
+            or item.get("token_id")
+            or item.get("tokenId")
+            or parent.get("asset_id")
+            or parent.get("assetId")
+            or ""
+        ).strip()
+        if not asset_id:
+            return
+
+        best_bid = (
+            _safe_float(item.get("best_bid"))
+            or _safe_float(item.get("bid"))
+            or _safe_float(item.get("bestBid"))
+        )
+        best_ask = (
+            _safe_float(item.get("best_ask"))
+            or _safe_float(item.get("ask"))
+            or _safe_float(item.get("bestAsk"))
+        )
+        if event_type == "book":
+            parsed_bid, parsed_ask = self._extract_book_top(item)
+            best_bid = best_bid if best_bid is not None else parsed_bid
+            best_ask = best_ask if best_ask is not None else parsed_ask
+        price = _safe_float(item.get("price"))
+        side = str(item.get("side") or "").strip().upper()
+        if event_type == "price_change" and price is not None:
+            if side == "BUY":
+                best_ask = price
+            elif side == "SELL":
+                best_bid = price
+
+        last_trade = (
+            _safe_float(item.get("last_trade_price"))
+            or _safe_float(item.get("lastTradePrice"))
+            or (price if event_type == "last_trade_price" else None)
+        )
+
+        now = time.time()
+        with self._lock:
+            previous = dict(self._quotes.get(asset_id) or {})
+            if best_bid is not None:
+                previous["best_bid"] = best_bid
+            if best_ask is not None:
+                previous["best_ask"] = best_ask
+            if last_trade is not None:
+                previous["last_trade_price"] = last_trade
+            previous["asset_id"] = asset_id
+            previous["event_type"] = event_type
+            previous["t"] = now
+            self._quotes[asset_id] = previous
+            self._last_message_at = now
+
+    def _extract_book_top(
+        self,
+        payload: Dict[str, Any],
+    ) -> tuple[Optional[float], Optional[float]]:
+        best_bid = None
+        best_ask = None
+
+        bids = payload.get("bids")
+        if isinstance(bids, list):
+            for item in bids:
+                price = self._extract_level_price(item)
+                if price is None:
+                    continue
+                best_bid = price if best_bid is None else max(best_bid, price)
+
+        asks = payload.get("asks")
+        if isinstance(asks, list):
+            for item in asks:
+                price = self._extract_level_price(item)
+                if price is None:
+                    continue
+                best_ask = price if best_ask is None else min(best_ask, price)
+
+        return best_bid, best_ask
+
+    @staticmethod
+    def _extract_level_price(level: Any) -> Optional[float]:
+        if isinstance(level, dict):
+            return _safe_float(level.get("price"))
+        if isinstance(level, (list, tuple)) and level:
+            return _safe_float(level[0])
+        return None

tests/test_polymarket_readonly.py

@@ -61,6 +61,62 @@ def test_fetch_token_market_data_prefers_orderbook_executable_prices():
     assert data["last_trade_price"] == 0.49
 
 
+def test_get_token_market_data_prefers_fresh_ws_cache():
+    layer = PolymarketReadOnlyLayer()
+
+    class FakeWsCache:
+        enabled = True
+
+        def subscribe(self, asset_ids):
+            self.asset_ids = list(asset_ids)
+
+        @staticmethod
+        def get_market_data(_token_id):
+            return {
+                "buy": 0.33,
+                "sell": 0.31,
+                "midpoint": 0.32,
+                "quote_source": "polymarket_ws",
+                "quote_age_ms": 80,
+            }
+
+    layer._ws_quote_cache = FakeWsCache()
+    layer._fetch_token_market_data = lambda _token_id: {"buy": 0.99}
+
+    data = layer._get_token_market_data("token-1")
+
+    assert data["buy"] == 0.33
+    assert data["sell"] == 0.31
+    assert data["quote_source"] == "polymarket_ws"
+
+
+def test_price_analysis_computes_edge_kelly_and_lock():
+    layer = PolymarketReadOnlyLayer()
+
+    analysis = layer._build_price_analysis(
+        model_probability=0.62,
+        yes_buy=0.52,
+        yes_sell=0.50,
+        no_buy=0.45,
+        no_sell=0.43,
+    )
+
+    assert analysis["available"] is True
+    assert abs(analysis["yes"]["edge"] - 0.10) < 0.000001
+    assert round(analysis["yes"]["kelly_fraction"], 6) == round(
+        (0.62 - 0.52) / (1.0 - 0.52),
+        6,
+    )
+    assert round(analysis["yes"]["quarter_kelly"], 6) == round(
+        ((0.62 - 0.52) / (1.0 - 0.52)) / 4.0,
+        6,
+    )
+    assert abs(analysis["no"]["edge"] - -0.07) < 0.000001
+    assert analysis["lock"]["available"] is True
+    assert round(analysis["lock"]["edge"], 6) == 0.03
+    assert analysis["best_side"] == "yes"
+
+
 def test_build_top_temperature_buckets_dedupes_same_temperature():
     layer = PolymarketReadOnlyLayer()
 

tests/test_polymarket_ws_cache.py

@@ -0,0 +1,124 @@
+import asyncio
+import json
+import time
+
+from src.data_collection.polymarket_ws_cache import PolymarketWsQuoteCache
+
+
+def test_ws_cache_parses_best_bid_ask_event():
+    cache = PolymarketWsQuoteCache(enabled=True, quote_ttl_sec=30)
+
+    cache.subscribe(["asset-1"])
+    cache._handle_message(
+        {
+            "event_type": "best_bid_ask",
+            "asset_id": "asset-1",
+            "best_bid": "0.41",
+            "best_ask": "0.44",
+        }
+    )
+
+    data = cache.get_market_data("asset-1")
+
+    assert data is not None
+    assert data["sell"] == 0.41
+    assert data["buy"] == 0.44
+    assert data["midpoint"] == 0.425
+    assert data["quote_source"] == "polymarket_ws"
+
+
+def test_ws_cache_ignores_stale_quotes():
+    cache = PolymarketWsQuoteCache(enabled=True, quote_ttl_sec=1)
+    cache._quotes["asset-1"] = {
+        "asset_id": "asset-1",
+        "best_bid": 0.41,
+        "best_ask": 0.44,
+        "t": time.time() - 10,
+    }
+
+    assert cache.get_market_data("asset-1") is None
+
+
+def test_ws_cache_parses_price_change_side_updates():
+    cache = PolymarketWsQuoteCache(enabled=True, quote_ttl_sec=30)
+
+    cache._handle_message(
+        {
+            "event_type": "price_change",
+            "changes": [
+                {
+                    "asset_id": "asset-1",
+                    "side": "BUY",
+                    "price": "0.48",
+                },
+                {
+                    "asset_id": "asset-1",
+                    "side": "SELL",
+                    "price": "0.45",
+                },
+            ],
+        }
+    )
+
+    data = cache.get_market_data("asset-1")
+
+    assert data is not None
+    assert data["buy"] == 0.48
+    assert data["sell"] == 0.45
+
+
+def test_ws_cache_parses_price_changes_key_and_book_event():
+    cache = PolymarketWsQuoteCache(enabled=True, quote_ttl_sec=30)
+
+    cache._handle_message(
+        {
+            "event_type": "book",
+            "asset_id": "asset-1",
+            "bids": [{"price": "0.42"}, {"price": "0.44"}],
+            "asks": [{"price": "0.51"}, {"price": "0.49"}],
+        }
+    )
+    cache._handle_message(
+        {
+            "event_type": "price_change",
+            "price_changes": [
+                {
+                    "asset_id": "asset-1",
+                    "side": "BUY",
+                    "price": "0.48",
+                }
+            ],
+        }
+    )
+
+    data = cache.get_market_data("asset-1")
+
+    assert data is not None
+    assert data["sell"] == 0.44
+    assert data["buy"] == 0.48
+
+
+def test_ws_cache_subscription_payloads_match_market_channel_shape():
+    class FakeWs:
+        def __init__(self):
+            self.messages = []
+
+        async def send(self, payload):
+            self.messages.append(json.loads(payload))
+
+    cache = PolymarketWsQuoteCache(enabled=True)
+    ws = FakeWs()
+
+    asyncio.run(cache._send_subscription(ws, ["asset-1"], initial=True))
+    asyncio.run(cache._send_subscription(ws, ["asset-2"], initial=False))
+
+    assert ws.messages[0] == {
+        "type": "market",
+        "assets_ids": ["asset-1"],
+        "custom_feature_enabled": True,
+    }
+    assert ws.messages[1] == {
+        "operation": "subscribe",
+        "assets_ids": ["asset-2"],
+        "custom_feature_enabled": True,
+    }