"""Prediction Mixin — V25 signal extraction and prediction building.

Auto-extracted mixin module — split from services/single_match_orchestrator.py.
All methods here are composed into SingleMatchOrchestrator via inheritance.
`self` attributes (self.dsn, self.enrichment, self.v25_predictor, etc.) are
initialised in the main __init__.
"""

from __future__ import annotations

import json
import re
import time
import math
import os
import pickle
from collections import defaultdict
from typing import Any, Dict, List, Optional, Set, Tuple, overload

import pandas as pd
import numpy as np

import psycopg2
from psycopg2.extras import RealDictCursor

from data.db import get_clean_dsn
from schemas.prediction import FullMatchPrediction
from schemas.match_data import MatchData
from models.v25_ensemble import V25Predictor, get_v25_predictor
try:
    from models.v27_predictor import V27Predictor, compute_divergence, compute_value_edge
except ImportError:
    class V27Predictor:  # type: ignore[no-redef]
        def __init__(self): self.models = {}
        def load_models(self): return False
        def predict_all(self, features): return {}
    def compute_divergence(*args, **kwargs):
        return {}
    def compute_value_edge(*args, **kwargs):
        return {}
from features.odds_band_analyzer import OddsBandAnalyzer
try:
    from models.basketball_v25 import (
        BasketballMatchPrediction,
        get_basketball_v25_predictor,
    )
except ImportError:
    BasketballMatchPrediction = Any  # type: ignore[misc]
    def get_basketball_v25_predictor() -> Any:
        raise ImportError("Basketball predictor is not available")
from core.engines.player_predictor import PlayerPrediction, get_player_predictor
from services.feature_enrichment import FeatureEnrichmentService
from services.betting_brain import BettingBrain
from services.v26_shadow_engine import V26ShadowEngine, get_v26_shadow_engine
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, get_config
from models.calibration import get_calibrator
from models.league_model import get_league_model_loader, FILE_TO_SIGNAL


class PredictionMixin:
    def _get_score_model(self) -> Optional[Dict]:
        """Load XGBoost score prediction model (non-fatal)."""
        if hasattr(self, "_score_model_cache"):
            return self._score_model_cache
        score_model_path = os.path.join(
            os.path.dirname(os.path.dirname(os.path.abspath(__file__))),
            "models", "xgb_score.pkl",
        )
        try:
            if os.path.exists(score_model_path):
                with open(score_model_path, "rb") as f:
                    model_data = pickle.load(f)
                if all(k in model_data for k in ("home_model", "away_model", "ht_home_model", "ht_away_model", "features")):
                    self._score_model_cache = model_data
                    print(f"[SCORE] ✅ Score model loaded ({len(model_data['features'])} features)")
                    return self._score_model_cache
        except Exception as e:
            print(f"[SCORE] ⚠ Load failed (non-fatal, using heuristic): {e}")
        self._score_model_cache = None
        return None

    def _predict_score_with_model(self, features: Dict[str, float]) -> Optional[Dict[str, float]]:
        """Predict FT/HT scores using XGBoost score model."""
        score_model = self._get_score_model()
        if score_model is None:
            return None
        try:
            import pandas as _pd
            model_features = score_model["features"]
            row = {f: float(features.get(f, 0)) for f in model_features}
            df = _pd.DataFrame([row])
            ft_home = max(0.0, float(score_model["home_model"].predict(df)[0]))
            ft_away = max(0.0, float(score_model["away_model"].predict(df)[0]))
            ht_home = max(0.0, float(score_model["ht_home_model"].predict(df)[0]))
            ht_away = max(0.0, float(score_model["ht_away_model"].predict(df)[0]))
            return {
                "ft_home": round(ft_home, 2),
                "ft_away": round(ft_away, 2),
                "ht_home": round(ht_home, 2),
                "ht_away": round(ht_away, 2),
            }
        except Exception as e:
            print(f"[SCORE] ⚠ Prediction error (fallback to heuristic): {e}")
            return None

    _V25_KEY_MAP = {
        "ms": "MS",
        "ou15": "OU15",
        "ou25": "OU25",
        "ou35": "OU35",
        "btts": "BTTS",
        "ht_result": "HT",
        "ht_ou05": "HT_OU05",
        "ht_ou15": "HT_OU15",
        "htft": "HTFT",
        "cards_ou45": "CARDS",
        "handicap_ms": "HCAP",
        "odd_even": "OE",
    }

    def _get_v25_signal(
        self,
        data: MatchData,
        features: Optional[Dict[str, float]] = None,
    ) -> Dict[str, Any]:
        """
        Get V25 ensemble predictions for all available markets.
        Returns a dict keyed by UPPERCASE market name (MS, OU25, BTTS, etc.)
        each with a 'probs' sub-dict that _prob_map can consume.

        CRITICAL: Keys MUST be uppercase to match _build_v25_prediction lookups.
        """
        v25 = self._get_v25_predictor()
        feature_row = features or self._build_v25_features(data)

        signal: Dict[str, Any] = {}

        # ── League-specific model override ─────────────────────────────────
        league_id = getattr(data, "league_id", None)
        league_model = None
        if league_id:
            try:
                league_model = get_league_model_loader().get(league_id)
            except Exception:
                league_model = None

        if league_model:
            # Predict all available markets with league-specific XGBoost
            for mkey, sig_key in FILE_TO_SIGNAL.items():
                probs = league_model.predict_market(mkey, feature_row)
                if probs:
                    best_label = max(probs, key=probs.__getitem__)
                    signal[sig_key] = {
                        "probs": probs,
                        "raw_probs": probs,
                        "pick": best_label,
                        "probability": float(probs[best_label]),
                        "confidence": round(float(probs[best_label]) * 100.0, 1),
                        "source": "league_specific",
                    }
            if signal:
                print(f"  [LEAGUE-MODEL] {league_id}: {len(signal)} markets predicted")
                # Fill remaining markets from general V25 (markets not in league model)
                # fall through to general prediction below for missing ones

        def _temperature_scale(probs_dict: Dict[str, float], temperature: float = 1.5) -> Dict[str, float]:
            """
            Apply temperature scaling to soften overconfident model outputs.

            LightGBM often produces extreme probabilities (e.g., 0.999 / 0.001).
            Temperature scaling converts to log-odds, divides by T, then re-normalizes.
            T=1.0 → no change, T>1 → softer probabilities.

            Standard approach for post-hoc model calibration (Guo et al., 2017).

            V34: Reduced from 2.5 to 1.5 — V25 model is already calibrated via
            odds-aware training. Excessive flattening was destroying signal.
            """
            import math
            eps = 1e-7  # numerical stability
            n = len(probs_dict)

            # V34: Reduced temperature — odds-aware model is already calibrated
            # Binary markets (2-class) tend to be more overconfident in LGB
            if n <= 2:
                T = max(temperature, 1.5)   # was 2.0
            elif n == 3:
                T = max(temperature * 0.8, 1.2)  # was 1.5 — 3-way slightly less aggressive
            else:
                T = max(temperature * 0.6, 1.0)  # was 1.3 — 9-way (HTFT) already spread

            # Convert to log-odds and apply temperature
            labels = list(probs_dict.keys())
            log_odds = []
            for label in labels:
                p = max(eps, min(1.0 - eps, float(probs_dict[label])))
                log_odds.append(math.log(p) / T)

            # Softmax re-normalization
            max_lo = max(log_odds)
            exp_vals = [math.exp(lo - max_lo) for lo in log_odds]
            total = sum(exp_vals)

            scaled = {}
            for i, label in enumerate(labels):
                scaled[label] = exp_vals[i] / total

            return scaled

        calibrator = get_calibrator()
        _temperature = float(get_config().get('model_ensemble.temperature', 1.5))

        # Map (market_key, label) → calibrator market key
        _CAL_KEY_MAP: Dict[str, str] = {
            "ms_1": "ms_home", "ms_x": "ms_draw", "ms_2": "ms_away",
            "ou15_over": "ou15", "ou15_under": "ou15",
            "ou25_over": "ou25", "ou25_under": "ou25",
            "ou35_over": "ou35", "ou35_under": "ou35",
            "btts_yes": "btts", "btts_no": "btts",
            "ht_1": "ht_home", "ht_x": "ht_draw", "ht_2": "ht_away",
        }

        def _enrich_signal_entry(probs_dict: Dict[str, float], market_key: str = "") -> Dict[str, Any]:
            """Temperature scaling + Isotonic calibration pipeline."""
            scaled_probs = _temperature_scale(probs_dict, temperature=_temperature)

            # Isotonic calibration per outcome (if trained models exist)
            if market_key:
                calibrated = {}
                for label, prob in scaled_probs.items():
                    raw_key = f"{market_key}_{label}".lower().replace(" ", "_")
                    cal_key = _CAL_KEY_MAP.get(raw_key, raw_key)
                    calibrated[label] = calibrator.calibrate(cal_key, prob)
                total = sum(calibrated.values())
                if total > 0:
                    calibrated = {k: v / total for k, v in calibrated.items()}
                scaled_probs = calibrated

            best_label = max(scaled_probs, key=scaled_probs.__getitem__)
            best_prob = float(scaled_probs[best_label])
            return {
                "probs": scaled_probs,
                "raw_probs": probs_dict,
                "pick": best_label,
                "probability": best_prob,
                "confidence": round(best_prob * 100.0, 1),
            }

        # Core markets using dedicated methods (skip if league model already covered them)
        if "MS" not in signal:
            h, d, a = v25.predict_ms(feature_row)
            signal["MS"] = _enrich_signal_entry({"1": h, "X": d, "2": a}, "ms")
            print(f"  [V25-SIGNAL] MS → H={h:.4f} D={d:.4f} A={a:.4f}")
        else:
            print(f"  [LEAGUE-MODEL] MS → {signal['MS']['probs']}")

        if "OU25" not in signal:
            over25, under25 = v25.predict_ou25(feature_row)
            signal["OU25"] = _enrich_signal_entry({"Over": over25, "Under": under25}, "ou25")
            print(f"  [V25-SIGNAL] OU25 → O={over25:.4f} U={under25:.4f}")

        if "BTTS" not in signal:
            btts_y, btts_n = v25.predict_btts(feature_row)
            signal["BTTS"] = _enrich_signal_entry({"Yes": btts_y, "No": btts_n}, "btts")
            print(f"  [V25-SIGNAL] BTTS → Y={btts_y:.4f} N={btts_n:.4f}")

        # Additional markets via generic predict_market (skip if league model covered them)
        for model_key, label_map in [
            ("ou15", {"Over": 0, "Under": None}),
            ("ou35", {"Over": 0, "Under": None}),
            ("ht_result", {"1": 0, "X": 1, "2": 2}),
            ("ht_ou05", {"Over": 0, "Under": None}),
            ("ht_ou15", {"Over": 0, "Under": None}),
            ("htft", None),
            ("cards_ou45", {"Over": 0, "Under": None}),
            ("handicap_ms", {"1": 0, "X": 1, "2": 2}),
            ("odd_even", {"Odd": 0, "Even": None}),
        ]:
            out_key = str(self._V25_KEY_MAP.get(model_key, model_key.upper()))
            if out_key in signal:
                continue  # already predicted by league-specific model
            if not v25.has_market(model_key):
                continue
            raw = v25.predict_market(model_key, feature_row)
            if raw is None:
                continue

            if label_map is None:
                # HTFT — 9 combinations
                htft_labels = ["1/1", "1/X", "1/2", "X/1", "X/X", "X/2", "2/1", "2/X", "2/2"]
                probs_dict = {}
                for i, label in enumerate(htft_labels):
                    probs_dict[label] = float(raw[i]) if i < len(raw) else 0.0
                signal[out_key] = _enrich_signal_entry(probs_dict, model_key)
            elif len(label_map) == 2:
                # Binary market
                labels = list(label_map.keys())
                p = float(raw[0]) if len(raw) >= 1 else None
                if p is None:
                    print(f"  [V25-SIGNAL] {out_key} → EMPTY raw output, skipped")
                    continue
                signal[out_key] = _enrich_signal_entry({labels[0]: p, labels[1]: 1.0 - p}, model_key)
            elif len(label_map) == 3:
                # 3-class market
                labels = list(label_map.keys())
                probs_dict = {}
                for i, label in enumerate(labels):
                    if i >= len(raw):
                        print(f"  [V25-SIGNAL] {out_key} → insufficient probabilities in raw output")
                        break
                    probs_dict[label] = float(raw[i])
                else:
                    signal[out_key] = _enrich_signal_entry(probs_dict, model_key)

            if out_key in signal:
                print(f"  [V25-SIGNAL] {out_key} → {signal[out_key]['probs']}")

        print(f"  [V25-SIGNAL] Total markets with real predictions: {len(signal)}")
        if not signal:
            raise RuntimeError("V25 model produced ZERO market predictions — cannot continue")

        return signal

    @staticmethod
    def _prob_map(signal: Optional[Dict[str, Any]], market: str, defaults: Dict[str, float]) -> Dict[str, float]:
        """Extract normalised probabilities from signal.

        If the signal contains real model output for this market, use it.
        If the market is missing from the signal, log a warning and return
        the defaults as a LAST RESORT (so the pipeline doesn't crash).
        The defaults are ONLY used for non-core / secondary markets that
        may not have a trained model yet (e.g. CARDS, HCAP, OE).
        """
        market_payload = signal.get(market, {}) if isinstance(signal, dict) else {}
        probs = market_payload.get("probs", {}) if isinstance(market_payload, dict) else {}
        if not isinstance(probs, dict) or not probs:
            print(f"  ⚠️ [PROB_MAP] Market '{market}' NOT found in V25 signal — model output missing")
            return dict(defaults)
        out = {key: float(probs.get(key, value)) for key, value in defaults.items()}
        total = sum(out.values())
        if total <= 0:
            print(f"  ⚠️ [PROB_MAP] Market '{market}' has zero total probability")
            return dict(defaults)
        return {key: value / total for key, value in out.items()}

    @staticmethod
    def _is_cup_game(league_name: str) -> bool:
        """Detect cup/knockout competitions where home advantage is significantly weaker."""
        name = (league_name or "").lower()
        cup_keywords = (
            "kupa", "cup", "coupe", "copa", "coppa", "pokal",
            "trophy", "shield", "challenge",
            "ziraat", "süper kupa", "super cup",
        )
        return any(kw in name for kw in cup_keywords)

    @staticmethod
    def _best_prob_pick(prob_map: Dict[str, float]) -> Tuple[str, float]:
        if not prob_map:
            return "", 0.0
        pick = max(prob_map, key=prob_map.__getitem__)
        return pick, float(prob_map[pick])

    @staticmethod
    def _poisson_score_top5(home_xg: float, away_xg: float, max_goals: int = 5) -> List[Dict[str, Any]]:
        def poisson_p(lmbda: float, k: int) -> float:
            return math.exp(-lmbda) * (lmbda ** k) / math.factorial(k)

        scores: List[Tuple[str, float]] = []
        for home_goals in range(max_goals + 1):
            for away_goals in range(max_goals + 1):
                prob = poisson_p(home_xg, home_goals) * poisson_p(away_xg, away_goals)
                scores.append((f"{home_goals}-{away_goals}", prob))
        scores.sort(key=lambda item: item[1], reverse=True)
        return [
            {"score": score, "prob": round(prob, 4)}
            for score, prob in scores[:5]
        ]

    def _build_v25_prediction(
        self,
        data: MatchData,
        features: Dict[str, float],
        v25_signal: Dict[str, Any],
    ) -> FullMatchPrediction:
        prediction = FullMatchPrediction(
            match_id=data.match_id,
            home_team=data.home_team_name,
            away_team=data.away_team_name,
        )

        ms_probs = self._prob_map(v25_signal, "MS", {"1": 0.33, "X": 0.34, "2": 0.33})
        ou15_probs = self._prob_map(v25_signal, "OU15", {"Under": 0.5, "Over": 0.5})
        ou25_probs = self._prob_map(v25_signal, "OU25", {"Under": 0.5, "Over": 0.5})
        ou35_probs = self._prob_map(v25_signal, "OU35", {"Under": 0.5, "Over": 0.5})
        btts_probs = self._prob_map(v25_signal, "BTTS", {"No": 0.5, "Yes": 0.5})
        ht_probs = self._prob_map(v25_signal, "HT", {"1": 0.33, "X": 0.34, "2": 0.33})
        ht_ou05_probs = self._prob_map(v25_signal, "HT_OU05", {"Under": 0.5, "Over": 0.5})
        ht_ou15_probs = self._prob_map(v25_signal, "HT_OU15", {"Under": 0.5, "Over": 0.5})
        htft_probs = self._prob_map(
            v25_signal,
            "HTFT",
            {"1/1": 1 / 9, "1/X": 1 / 9, "1/2": 1 / 9, "X/1": 1 / 9, "X/X": 1 / 9, "X/2": 1 / 9, "2/1": 1 / 9, "2/X": 1 / 9, "2/2": 1 / 9},
        )
        oe_probs = self._prob_map(v25_signal, "OE", {"Even": 0.5, "Odd": 0.5})
        cards_probs = self._prob_map(v25_signal, "CARDS", {"Under": 0.5, "Over": 0.5})
        hcap_probs = self._prob_map(v25_signal, "HCAP", {"1": 0.33, "X": 0.34, "2": 0.33})

        # Cup game: dampen home advantage — model trained on league data overestimates home edge
        is_cup = self._is_cup_game(getattr(data, "league_name", "") or "")
        if is_cup:
            # Shift 8% of home probability toward away and draw (rotation, neutral venue effect)
            cup_transfer = ms_probs["1"] * 0.08
            ms_probs = {
                "1": ms_probs["1"] - cup_transfer,
                "X": ms_probs["X"] + cup_transfer * 0.4,
                "2": ms_probs["2"] + cup_transfer * 0.6,
            }
            total = sum(ms_probs.values())
            ms_probs = {k: v / total for k, v in ms_probs.items()}

        prediction.ms_home_prob = ms_probs["1"]
        prediction.ms_draw_prob = ms_probs["X"]
        prediction.ms_away_prob = ms_probs["2"]
        prediction.ms_pick, ms_top = self._best_prob_pick(ms_probs)
        prediction.ms_confidence = ms_top * 100.0

        prediction.dc_1x_prob = prediction.ms_home_prob + prediction.ms_draw_prob
        prediction.dc_x2_prob = prediction.ms_draw_prob + prediction.ms_away_prob
        prediction.dc_12_prob = prediction.ms_home_prob + prediction.ms_away_prob
        dc_probs = {"1X": prediction.dc_1x_prob, "X2": prediction.dc_x2_prob, "12": prediction.dc_12_prob}
        prediction.dc_pick, dc_top = self._best_prob_pick(dc_probs)
        prediction.dc_confidence = dc_top * 100.0

        prediction.over_15_prob = ou15_probs["Over"]
        prediction.under_15_prob = ou15_probs["Under"]
        prediction.ou15_pick = "1.5 Üst" if prediction.over_15_prob >= prediction.under_15_prob else "1.5 Alt"
        prediction.ou15_confidence = max(prediction.over_15_prob, prediction.under_15_prob) * 100.0

        prediction.over_25_prob = ou25_probs["Over"]
        prediction.under_25_prob = ou25_probs["Under"]
        prediction.ou25_pick = "2.5 Üst" if prediction.over_25_prob >= prediction.under_25_prob else "2.5 Alt"
        prediction.ou25_confidence = max(prediction.over_25_prob, prediction.under_25_prob) * 100.0

        prediction.over_35_prob = ou35_probs["Over"]
        prediction.under_35_prob = ou35_probs["Under"]
        prediction.ou35_pick = "3.5 Üst" if prediction.over_35_prob >= prediction.under_35_prob else "3.5 Alt"
        prediction.ou35_confidence = max(prediction.over_35_prob, prediction.under_35_prob) * 100.0

        prediction.btts_yes_prob = btts_probs["Yes"]
        prediction.btts_no_prob = btts_probs["No"]
        prediction.btts_pick = "KG Var" if prediction.btts_yes_prob >= prediction.btts_no_prob else "KG Yok"
        prediction.btts_confidence = max(prediction.btts_yes_prob, prediction.btts_no_prob) * 100.0

        prediction.ht_home_prob = ht_probs["1"]
        prediction.ht_draw_prob = ht_probs["X"]
        prediction.ht_away_prob = ht_probs["2"]
        prediction.ht_pick, ht_top = self._best_prob_pick(ht_probs)
        prediction.ht_confidence = ht_top * 100.0

        prediction.ht_over_05_prob = ht_ou05_probs["Over"]
        prediction.ht_under_05_prob = ht_ou05_probs["Under"]
        prediction.ht_ou_pick = "İY 0.5 Üst" if prediction.ht_over_05_prob >= prediction.ht_under_05_prob else "İY 0.5 Alt"

        prediction.ht_over_15_prob = ht_ou15_probs["Over"]
        prediction.ht_under_15_prob = ht_ou15_probs["Under"]
        prediction.ht_ou15_pick = "İY 1.5 Üst" if prediction.ht_over_15_prob >= prediction.ht_under_15_prob else "İY 1.5 Alt"

        prediction.ht_ft_probs = htft_probs

        prediction.odd_prob = oe_probs["Odd"]
        prediction.even_prob = oe_probs["Even"]
        prediction.odd_even_pick = "Tek" if prediction.odd_prob >= prediction.even_prob else "Çift"

        prediction.cards_over_prob = cards_probs["Over"]
        prediction.cards_under_prob = cards_probs["Under"]
        prediction.card_pick = "4.5 Üst" if prediction.cards_over_prob >= prediction.cards_under_prob else "4.5 Alt"
        prediction.cards_confidence = max(prediction.cards_over_prob, prediction.cards_under_prob) * 100.0

        prediction.handicap_home_prob = hcap_probs["1"]
        prediction.handicap_draw_prob = hcap_probs["X"]
        prediction.handicap_away_prob = hcap_probs["2"]
        prediction.handicap_pick, hcap_top = self._best_prob_pick(hcap_probs)
        prediction.handicap_confidence = hcap_top * 100.0

        # ── Score Prediction: Model-first, heuristic fallback ──────────
        ms_edge = prediction.ms_home_prob - prediction.ms_away_prob
        score_result = self._predict_score_with_model(features)
        if score_result is not None:
            # ML model predicted scores
            prediction.home_xg = score_result["ft_home"]
            prediction.away_xg = score_result["ft_away"]
            prediction.total_xg = round(prediction.home_xg + prediction.away_xg, 2)
            ht_home_xg = score_result["ht_home"]
            ht_away_xg = score_result["ht_away"]
            prediction.predicted_ft_score = f"{int(round(prediction.home_xg))}-{int(round(prediction.away_xg))}"
            prediction.predicted_ht_score = f"{int(round(ht_home_xg))}-{int(round(ht_away_xg))}"
        else:
            # Heuristic fallback (original formula)
            base_home_xg = max(0.25, (float(data.home_goals_avg or 1.3) + float(features.get("away_xga", data.away_conceded_avg) or 1.2)) / 2.0)
            base_away_xg = max(0.25, (float(data.away_goals_avg or 1.3) + float(features.get("home_xga", data.home_conceded_avg) or 1.2)) / 2.0)
            # ms_edge already computed above
            total_target = max(
                1.4,
                min(
                    4.8,
                    (float(features.get("league_avg_goals", 2.7)) * 0.55)
                    + ((float(data.home_goals_avg or 1.3) + float(data.away_goals_avg or 1.3)) * 0.45)
                    + ((prediction.over_25_prob - prediction.under_25_prob) * 1.15),
                ),
            )
            home_xg = max(0.2, base_home_xg + (ms_edge * 0.55) + ((prediction.btts_yes_prob - 0.5) * 0.18))
            away_xg = max(0.2, base_away_xg - (ms_edge * 0.55) + ((prediction.btts_yes_prob - 0.5) * 0.18))
            scale = total_target / max(home_xg + away_xg, 0.1)
            prediction.home_xg = round(home_xg * scale, 2)
            prediction.away_xg = round(away_xg * scale, 2)
            prediction.total_xg = round(prediction.home_xg + prediction.away_xg, 2)

        # Cup game: reduce xG by 20% — rotation + lower motivation + defensive tactics
        if is_cup:
            prediction.home_xg = round(prediction.home_xg * 0.80, 2)
            prediction.away_xg = round(prediction.away_xg * 0.80, 2)
            prediction.total_xg = round(prediction.home_xg + prediction.away_xg, 2)
            prediction.predicted_ft_score = f"{int(round(prediction.home_xg))}-{int(round(prediction.away_xg))}"
            prediction.predicted_ht_score = f"{int(round(prediction.home_xg * 0.45))}-{int(round(prediction.away_xg * 0.45))}"
        prediction.ft_scores_top5 = self._poisson_score_top5(prediction.home_xg, prediction.away_xg)

        # Score prediction: find the most likely scoreline consistent with the MS pick
        # Instead of just rounding xG (misleading), filter Poisson top scores by result direction
        ms_pick = prediction.ms_pick  # "1", "X", or "2"
        top5 = prediction.ft_scores_top5
        if top5 and ms_pick in ("1", "X", "2"):
            def _result_of(score_str: str) -> str:
                try:
                    h, a = map(int, score_str.split("-"))
                    if h > a: return "1"
                    if h < a: return "2"
                    return "X"
                except Exception:
                    return "?"

            # Filter to scorelines matching the predicted result
            matching = [s for s in top5 if _result_of(s["score"]) == ms_pick]
            if matching:
                best = matching[0]  # already sorted by probability desc
                h_str, a_str = best["score"].split("-")
                prediction.predicted_ft_score = best["score"]
                # Recalculate HT score proportionally from the FT pick
                h_val, a_val = int(h_str), int(a_str)
                prediction.predicted_ht_score = f"{int(round(h_val * 0.45))}-{int(round(a_val * 0.45))}"

        max_market_conf = max(
            prediction.ms_confidence,
            prediction.ou15_confidence,
            prediction.ou25_confidence,
            prediction.ou35_confidence,
            prediction.btts_confidence,
            prediction.ht_confidence,
            prediction.cards_confidence,
            prediction.handicap_confidence,
        )
        lineup_conf = max(0.0, min(1.0, float(getattr(data, "lineup_confidence", 0.0) or 0.0)))
        lineup_penalty = 12.0 if data.lineup_source == "none" else max(1.5, (1.0 - lineup_conf) * 8.0) if data.lineup_source == "probable_xi" else 0.0
        referee_penalty = 6.0 if not data.referee_name else 0.0
        parity_penalty = 8.0 if abs(ms_edge) < 0.08 else 0.0
        # Cup game penalty: model trained on league data has lower reliability for cup matches
        cup_penalty = 10.0 if is_cup else 0.0
        # Bookmaker margin penalty: high margin signals that even the market is uncertain
        bm_margin = 0.0
        odds_data = getattr(data, "odds_data", {}) or {}
        _h, _d, _a = float(odds_data.get("ms_h") or 0), float(odds_data.get("ms_d") or 0), float(odds_data.get("ms_a") or 0)
        if _h > 1.01 and _d > 1.01 and _a > 1.01:
            bm_margin = (1 / _h + 1 / _d + 1 / _a) - 1
        bookmaker_penalty = 12.0 if bm_margin > 0.20 else 6.0 if bm_margin > 0.15 else 0.0
        prediction.risk_score = round(min(100.0, max(10.0, 100.0 - max_market_conf + lineup_penalty + referee_penalty + parity_penalty + cup_penalty + bookmaker_penalty)), 1)
        if prediction.risk_score >= 78:
            prediction.risk_level = "EXTREME"
        elif prediction.risk_score >= 62:
            prediction.risk_level = "HIGH"
        elif prediction.risk_score >= 40:
            prediction.risk_level = "MEDIUM"
        else:
            prediction.risk_level = "LOW"
        prediction.is_surprise_risk = prediction.risk_level in {"HIGH", "EXTREME"} or prediction.ms_draw_prob >= 0.30
        prediction.surprise_type = "balanced_match_risk" if abs(ms_edge) < 0.08 else "draw_pressure" if prediction.ms_draw_prob >= 0.30 else ""
        prediction.risk_warnings = []
        if is_cup:
            prediction.risk_warnings.append("cup_game_home_advantage_reduced")
        if bookmaker_penalty > 0:
            prediction.risk_warnings.append(f"bookmaker_margin_high_{bm_margin*100:.0f}pct")
        if data.lineup_source == "probable_xi":
            prediction.risk_warnings.append("lineup_probable_not_confirmed")
            if lineup_conf < 0.65:
                prediction.risk_warnings.append("lineup_projection_low_confidence")
        if data.lineup_source == "none":
            prediction.risk_warnings.append("lineup_unavailable")
        if not data.referee_name:
            prediction.risk_warnings.append("missing_referee")
        if prediction.ms_draw_prob >= 0.30:
            prediction.risk_warnings.append("draw_probability_elevated")

        prediction.upset_score = int(round(max(0.0, min(100.0, (prediction.ms_draw_prob + min(prediction.ms_home_prob, prediction.ms_away_prob)) * 100.0))))
        prediction.upset_level = "HIGH" if prediction.upset_score >= 65 else "MEDIUM" if prediction.upset_score >= 45 else "LOW"
        prediction.upset_reasons = [prediction.surprise_type] if prediction.surprise_type else []
        surprise = self._build_surprise_profile(data, prediction)
        prediction.surprise_score = surprise["score"]
        prediction.surprise_comment = surprise["comment"]
        prediction.surprise_reasons = surprise["reasons"]
        prediction.surprise_breakdown = surprise.get("breakdown", [])
        # Auto-flag is_surprise_risk when score crosses 45 even if other paths didn't fire
        if surprise["score"] >= 45.0:
            prediction.is_surprise_risk = True

        prediction.team_confidence = round(max(35.0, min(95.0, 45.0 + (abs(ms_edge) * 85.0) + (abs(float(features.get("form_elo_diff", 0.0))) / 40.0))), 1)
        prediction.player_confidence = round(max(20.0, min(95.0, 38.0 + (float(features.get("home_key_players", 0.0)) + float(features.get("away_key_players", 0.0))) * 2.0 - (float(features.get("home_missing_impact", 0.0)) + float(features.get("away_missing_impact", 0.0))) * 22.0)), 1)
        prediction.odds_confidence = round(max(30.0, min(95.0, float(np.mean([prediction.ms_confidence, prediction.ou25_confidence, prediction.btts_confidence])))), 1)
        prediction.referee_confidence = 62.0 if data.referee_name else 35.0

        prediction.total_cards_pred = 4.8 if prediction.cards_over_prob >= prediction.cards_under_prob else 4.1
        prediction.total_corners_pred = round(8.8 + (prediction.over_25_prob - 0.5) * 2.5, 1)
        prediction.corner_pick = "9.5 Üst" if prediction.total_corners_pred >= 9.5 else "9.5 Alt"
        prediction.analysis_details = {
            "primary_model": "v25",
            "features_source": "v25.pre_match",
            "market_count": len([key for key in v25_signal.keys() if key != "value_bets"]),
            "lineup_source": data.lineup_source,
        }
        return prediction

    def _build_engine_breakdown(self, prediction: FullMatchPrediction) -> Dict[str, Any]:
        """
        Engine breakdown with backward-compatible flat scores + rich detail siblings.

        Shape:
          {
              team: 74.1, player: 55.7, odds: 55.2, referee: 62.0,   # legacy flat scores
              detail: { team: {score, label, ...}, player: {...}, ... }
          }
        """
        components = {
            "team": ("Takım modeli", float(prediction.team_confidence)),
            "player": ("Oyuncu / kadro modeli", float(prediction.player_confidence)),
            "odds": ("Oran piyasası", float(prediction.odds_confidence)),
            "referee": ("Hakem etkisi", float(prediction.referee_confidence)),
        }
        flat: Dict[str, Any] = {}
        detail: Dict[str, Any] = {}
        for key, (display, raw) in components.items():
            score = round(raw, 1)
            label, interpretation = self._confidence_label(score)
            flat[key] = score
            detail[key] = {
                "score": score,
                "label": label,
                "display_name": display,
                "interpretation": interpretation,
            }
        flat["detail"] = detail
        return flat