"""build_truth_reconciliation_gate_v1.py — TRUTH_RECONCILIATION_GATE_V1

P0-T3: 동일 지표가 파일마다 다른 값을 가지면 자동 FAIL.
감시 지표: prediction_match_rate_pct, t20_pass_rate, value_damage_pct_avg,
           gs_coverage_pct, portfolio_alpha_confidence, performance_readiness_score
허용 오차: 비율 지표 ±0.5%p, 금액 지표 ±1원
"""
from __future__ import annotations

import json
from datetime import datetime, timezone
from pathlib import Path
from typing import Any

ROOT = Path(__file__).resolve().parents[1]
TEMP = ROOT / "Temp"
DEFAULT_OUT = TEMP / "truth_reconciliation_gate_v1.json"

TOLERANCE_RATE = 0.5    # %p
TOLERANCE_KRW = 1.0     # 원

# 감시 지표: (정규화된 metric_id, json_pointer_list, 단위, 제외 파일 패턴)
# 위양성 방지: 같은 key명이 다른 개념에 쓰이는 파일은 명시 제외
MONITORED_METRICS: list[tuple[str, list[str], str, set[str]]] = [
    ("prediction_match_rate_pct",
     ["prediction_match_rate_pct", "t5_ap_combined"],
     "rate",
     # v5 = legacy v5.todo.batch 파일 (builder 없음), v7 = 다른 블렌드 점수
     {"prediction_accuracy_harness_v5", "smart_cash_recovery_v7"}),
    ("t20_pass_rate",
     ["t20_pass_rate"],                    # pass_rate_pct는 제외 (completion_gap과 혼동)
     "rate",
     {"completion_gap", "phase_checks"}),  # 완료기준 통과율 파일 제외
    ("value_damage_pct_avg",
     ["value_damage_pct_avg"],
     "rate",
     # 다른 목적함수 + 구버전 아카이브 파일 제외 (현재 파이프라인 외 레거시)
     {"dynamic_value_preservation", "cash_raise_value_optimizer",
      "cash_raise_value_preservation", "value_preserving_cash_raise_v1",
      "hts_sell_blueprint",
      "smart_cash_recovery_v7.json"}),   # v7 non-authoritative (2026-05-31 legacy)
    ("gs_strict_coverage_pct",             # gs_coverage_pct 대신 strict 전용 포인터
     ["gs_coverage_pct"],
     "rate",
     {"gs_native_coverage_lock"}),         # native coverage는 다른 개념
    ("portfolio_alpha_confidence",
     ["portfolio_alpha_confidence", "alpha_confidence"],
     "rate",
     set()),
    ("performance_readiness_score",
     ["performance_readiness_score", "blended_performance_readiness_score"],
     "rate",
     set()),
]


def _load(p: Path) -> dict[str, Any]:
    if not p.exists():
        return {}
    try:
        obj = json.loads(p.read_text(encoding="utf-8"))
        return obj if isinstance(obj, dict) else {}
    except Exception:
        return {}


def _extract(d: dict[str, Any], pointers: list[str]) -> float | None:
    for ptr in pointers:
        v = d.get(ptr)
        if v is not None:
            try:
                f = float(v)
                if f != 0.0 or ptr in d:
                    return f
            except (TypeError, ValueError):
                pass
    return None


def main() -> int:
    # 모든 Temp JSON 로드
    json_files = list(TEMP.glob("*.json"))
    # 제외: 보고서/golden/binary
    exclude_patterns = {"formula_golden", "formula_behavioral", "formula_gas_parity", "engine_audit_2026"}
    candidates = [f for f in json_files if not any(ex in f.name for ex in exclude_patterns)]

    observations: dict[str, list[dict[str, Any]]] = {m[0]: [] for m in MONITORED_METRICS}

    for f in candidates:
        d = _load(f)
        if not d:
            continue
        rel = str(f.relative_to(ROOT))
        for metric_id, pointers, unit, exclude_patterns in MONITORED_METRICS:
            # 제외 패턴 파일 스킵
            if any(ep in f.name for ep in exclude_patterns):
                continue
            val = _extract(d, pointers)
            if val is not None:
                observations[metric_id].append({"file": rel, "value": val})

    conflicts: list[dict[str, Any]] = []
    for metric_id, pointers, unit, _ in MONITORED_METRICS:
        obs = observations[metric_id]
        if len(obs) < 2:
            continue
        values = [o["value"] for o in obs]
        min_v, max_v = min(values), max(values)
        tol = TOLERANCE_RATE if unit == "rate" else TOLERANCE_KRW
        if (max_v - min_v) > tol:
            conflicts.append({
                "metric_id": metric_id,
                "min": min_v,
                "max": max_v,
                "spread": round(max_v - min_v, 4),
                "tolerance": tol,
                "unit": unit,
                "observations": sorted(obs, key=lambda x: x["value"]),
            })

    gate = "PASS" if not conflicts else "FAIL"
    result = {
        "formula_id": "TRUTH_RECONCILIATION_GATE_V1",
        "gate": gate,
        "conflict_count": len(conflicts),
        "conflicts": conflicts,
        "monitored_metrics": [m[0] for m in MONITORED_METRICS],
        "excluded_per_metric": {m[0]: list(m[3]) for m in MONITORED_METRICS if m[3]},
        "files_scanned": len(candidates),
        "generated_at": datetime.now(timezone.utc).isoformat(),
    }

    DEFAULT_OUT.parent.mkdir(parents=True, exist_ok=True)
    DEFAULT_OUT.write_text(json.dumps(result, ensure_ascii=False, indent=2), encoding="utf-8")
    summary = {k: v for k, v in result.items() if k != "conflicts"}
    print(json.dumps(summary, indent=2, ensure_ascii=False))
    if gate == "PASS":
        print("TRUTH_RECONCILIATION_GATE_V1_PASS")
    else:
        print(f"TRUTH_RECONCILIATION_GATE_V1_FAIL ({len(conflicts)} conflicts)")
        for c in conflicts:
            print(f"  {c['metric_id']}: spread={c['spread']} (tol={c['tolerance']})")
            for o in c["observations"]:
                print(f"    {o['file']}: {o['value']}")
    return 0 if gate == "PASS" else 1


if __name__ == "__main__":
    raise SystemExit(main())