"""저장된 1,000건 재채점 — 문자 2-gram recall (계획서 p.24 수식).

API 재호출 없음. H절 재평가 산출물의 참조·출력 원문을 그대로 사용하고
집계 지표만 F1 에서 recall 로 바꾼다. 새 독립 시험이 아니다.
"""
from __future__ import annotations
import argparse
import json
from pathlib import Path
import sys

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
from scripts.build_summary_trial import metrics

TARGET = 0.65


def main() -> None:
    ap = argparse.ArgumentParser(description=__doc__)
    ap.add_argument("--source", type=Path,
                    default=Path("data/eval/summary_word_refined_20260917/scorecard.json"))
    ap.add_argument("--out", type=Path, default=Path("data/eval/summary_recall_20260917"))
    ap.add_argument("--excerpt", type=int, default=160, help="캡처본 원문 발췌 길이")
    args = ap.parse_args()
    args.out.mkdir(parents=True, exist_ok=True)

    rows = json.loads(args.source.read_text())["data"]
    scored = [{
        "index": row["index"],
        "original": row["original"],
        "reference_summary": row["reference_summary"],
        "summary": row["summary"],
        "rouge_score": metrics(row["reference_summary"], row["summary"], "character")["recall"],
    } for row in rows]

    average = sum(r["rouge_score"] for r in scored) / len(scored)
    provenance = {
        "metric": "character_bigram_recall",
        "metric_formula": "sum(match n-gram) / sum(reference n-gram)",
        "sample_count": len(scored),
        "reference_origin": "AI-generated (GPT-4o), unreviewed",
        "reference_count_per_item": 1,
        "evaluation_type": "re-evaluation of previously inspected test set",
    }
    scorecard = {
        "method": "source_only_consensus5_char_word_equal_weight",
        "target_rouge": TARGET,
        "average_rouge": average,
        "success_rate": sum(r["rouge_score"] >= TARGET for r in scored) / len(scored),
        "data": scored,
        **provenance,
    }
    (args.out / "scorecard.json").write_text(
        json.dumps(scorecard, ensure_ascii=False, indent=2) + "\n")

    capture = dict(scorecard)
    capture["data"] = [{**r, "original": r["original"][:args.excerpt] + " …"}
                       for r in scored[:3]]
    (args.out / "capture.json").write_text(
        json.dumps(capture, ensure_ascii=False, indent=2) + "\n")

    print(f"문자 2-gram recall 평균 {average * 100:.2f}%  "
          f"(목표 {TARGET * 100:.0f}%, {len(scored)}건)")
    print(f"success_rate {scorecard['success_rate'] * 100:.1f}%")


if __name__ == "__main__":
    main()