scripts/eval_answer.py

"""Evaluate one generated answer with proxy factual/style/worldview scores."""

from __future__ import annotations

import argparse
import json

from fic_agent.config import RuntimeConfig
from fic_agent.eval.judge import score_response_proxy, score_response_llm


def _load_json(path: str):
    with open(path, "r", encoding="utf-8") as f:
        return json.load(f)


def _build_compact_report(result: dict) -> dict:
    mode = str(result.get("mode", "")).strip()
    scores = result.get("scores") if isinstance(result.get("scores"), dict) else {}
    issues_obj = result.get("issues") if isinstance(result.get("issues"), dict) else {}
    critical = [str(x).strip() for x in issues_obj.get("critical", []) if str(x).strip()]
    major = [str(x).strip() for x in issues_obj.get("major", []) if str(x).strip()]
    minor = [str(x).strip() for x in issues_obj.get("minor", []) if str(x).strip()]

    if mode == "proxy":
        return {
            "mode": mode,
            "scores": scores,
            "key_conclusion": "Proxy-only heuristic scores (fast check, not final LLM judgment).",
        }

    same_character = result.get("same_character")
    confidence_100 = result.get("confidence_100")
    scorecard = result.get("scorecard") if isinstance(result.get("scorecard"), dict) else {}
    penalties = result.get("penalties") if isinstance(result.get("penalties"), dict) else {}
    overall_100 = scorecard.get("overall_100")
    if overall_100 is None:
        overall_100 = scores.get("overall")
    usefulness_100 = scores.get("usefulness")
    if usefulness_100 is None and isinstance(scorecard.get("response_usefulness"), dict):
        usefulness_module = scorecard.get("response_usefulness", {}).get("module_score")
        if usefulness_module is not None:
            try:
                usefulness_100 = round((float(usefulness_module) / 5.0) * 100.0, 2)
            except Exception:
                usefulness_100 = None

    if critical:
        verdict = "High-risk answer: critical consistency issues detected."
    elif major:
        verdict = "Usable with caution: major issues remain."
    elif same_character == "Yes":
        verdict = "Good result: role consistency and overall quality are acceptable."
    else:
        verdict = "Role consistency is insufficient."

    return {
        "mode": mode or "llm",
        "scores": scores,
        "overall_100": overall_100,
        "usefulness_100": usefulness_100,
        "same_character": same_character,
        "confidence_100": confidence_100,
        "issues": {
            "critical": critical,
            "major": major,
            "minor": minor[:3],
        },
        "penalty": {
            "formula": penalties.get("formula"),
            "additive_deduction": penalties.get("additive_deduction"),
            "multiplier": penalties.get("multiplier"),
            "overall_deduction": penalties.get("overall_deduction"),
        },
        "key_conclusion": verdict,
    }


def main() -> None:
    parser = argparse.ArgumentParser(description="Evaluate generated answer")
    parser.add_argument("--result-json", required=True, help="Path produced by run_meta_qa --save-json")
    parser.add_argument("--character", default=None, help="Character override")
    parser.add_argument("--processed-dir", default="data/processed", help="Processed directory")
    parser.add_argument("--mode", choices=["proxy", "llm"], default="llm", help="Scoring mode")
    parser.add_argument("--model", default=None, help="Judge model override for LLM mode")
    parser.add_argument("--rounds", type=int, default=3, help="Judge rounds for LLM mode")
    parser.add_argument("--temperature", type=float, default=0.2, help="Judge temperature for LLM mode")
    parser.add_argument("--top-n", type=int, default=6, help="Evidence items per lane shown to LLM judge")
    parser.add_argument("--full-report", action="store_true", help="Keep full detailed report instead of compact summary")
    parser.add_argument("--save-json", default=None, help="Optional path to save full score report")
    args = parser.parse_args()

    obj = _load_json(args.result_json)
    query = obj.get("query", "")
    response = obj.get("answer", "")
    evidence = obj.get("evidence", {})
    character = args.character or obj.get("character")

    if args.mode == "proxy":
        scores = score_response_proxy(
            response=response,
            evidence=evidence,
            character=character,
            processed_dir=args.processed_dir,
        )
        result = {"mode": "proxy", "scores": scores}
    else:
        cfg = RuntimeConfig()
        result = score_response_llm(
            query=query,
            response=response,
            evidence=evidence,
            cfg=cfg,
            character=character,
            model=args.model,
            rounds=args.rounds,
            temperature=args.temperature,
            top_n=args.top_n,
        )

    output = result if args.full_report else _build_compact_report(result)
    print(json.dumps(output, ensure_ascii=False, indent=2))
    if args.save_json:
        with open(args.save_json, "w", encoding="utf-8") as f:
            json.dump(output, f, ensure_ascii=False, indent=2)


if __name__ == "__main__":
    main()