gg2

ai-engine/services/orchestrator/market_board.py

@@ -56,7 +56,7 @@ from services.match_commentary import generate_match_commentary
 from utils.top_leagues import load_top_league_ids
 from utils.league_reliability import load_league_reliability
 from config.config_loader import build_threshold_dict, get_threshold_default
-from models.calibration import get_calibrator
+from models.calibration import get_calibrator, get_final_recalibrator
 
 # ── V30: Post-calibration trust factors ─────────────────────────────
 # Controls how much to trust isotonic calibrator vs raw model output.
@@ -1153,6 +1153,18 @@ class MarketBoardMixin:
             # overconfidence without destroying probability signal.
             # The tier system (V31b) is the real profitability gatekeeper.
             calibrated_conf = max(1.0, min(99.0, raw_conf * 0.92))
+
+        # ── FINAL-OUTPUT RECALIBRATION (V31e) ──────────────────────────
+        # Last-step per-market map: "system says X% -> reality is Y%". ONLY
+        # badly-miscalibrated markets carry a map (fit-ECE >= 5: OU15, OU35,
+        # HT_OU05, HT_OU15). MS and every already-good market pass through
+        # UNCHANGED -> guaranteed no regression. Out-of-sample proven (e.g.
+        # HT_OU15 ECE 29.2->0.8) and identity-safe for MS (1.1->1.3).
+        # This adjusts ONLY the displayed confidence so users see honest
+        # probabilities; all analysis below (probabilities, edges, vetoes,
+        # tiers, bands) is preserved, and the pre-recal value is kept for audit.
+        pre_recal_conf = calibrated_conf
+        calibrated_conf = get_final_recalibrator().recalibrate_conf(market, calibrated_conf)
         min_conf = self.market_min_conf.get(market, 55.0)
 
         implied_prob = (1.0 / odd) if odd > 1.0 else 0.0
@@ -1361,6 +1373,7 @@ class MarketBoardMixin:
             {
                 "raw_confidence": round(raw_conf, 1),
                 "calibrated_confidence": round(calibrated_conf, 1),
+                "calibrated_confidence_pre_recal": round(pre_recal_conf, 1),
                 "unified_score": round(bgs, 1),
                 "unified_score_label": bgs_label,
                 "min_required_confidence": round(min_conf, 1),