{ "evaluation_report_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\evaluation_report.json", "shadow_report_path": "E:\\web\\PolyWeather\\artifacts\\probability_calibration\\shadow_report.json", "evaluation_report_exists": true, "shadow_report_exists": true, "decision": { "decision": "hold", "ready_for_primary": false, "summary": "当前指标不足以切换 emos_primary,应继续保持 shadow。", "thresholds": { "evaluation_min_samples": 80, "shadow_min_samples": 50, "max_delta_mae": 0.05, "min_delta_crps": -0.02, "min_delta_bucket_hit_rate": 0.0, "max_delta_bucket_brier_promote": 0.02, "max_delta_bucket_brier_observe": 0.15 }, "evaluation": { "sample_count": 54, "delta_crps": -0.086732, "delta_mae": 0.0, "delta_bucket_hit_rate": 0.0 }, "shadow": { "sample_count": 48, "delta_mae": 0.0, "delta_bucket_hit_rate": 0.041666, "delta_bucket_brier": 0.123252 }, "blocking_reasons": [ "离线评估样本不足:54 < 80", "shadow 样本不足:48 < 50", "shadow bucket brier 退化超限:delta=0.123252" ], "worst_shadow_regressions": [ { "city": "dallas", "samples": 1, "delta_mae": 0.0, "delta_bucket_hit_rate": 0.0, "delta_bucket_brier": 0.792585 }, { "city": "chicago", "samples": 1, "delta_mae": 0.0, "delta_bucket_hit_rate": 0.0, "delta_bucket_brier": 0.791878 }, { "city": "seattle", "samples": 1, "delta_mae": 0.0, "delta_bucket_hit_rate": 0.0, "delta_bucket_brier": 0.61609 }, { "city": "wellington", "samples": 2, "delta_mae": 0.0, "delta_bucket_hit_rate": 0.0, "delta_bucket_brier": 0.509203 }, { "city": "tel aviv", "samples": 2, "delta_mae": 0.0, "delta_bucket_hit_rate": 0.0, "delta_bucket_brier": 0.439879 } ] } }