#!/usr/bin/env python3

"""Evaluate benchmark rows against a Git Baseline Book and target file."""

from __future__ import annotations

import argparse

import json

import sys

from pathlib import Path

ROOT = Path(__file__).resolve().parents[1]

sys.path.insert(0, str(ROOT / "scripts"))

from oakbench.baseline_book import load_baseline_book, load_targets  # noqa: E402

from oakbench.scorecard import evaluate_targets, load_jsonl  # noqa: E402

def parse_args() -> argparse.Namespace:

    parser = argparse.ArgumentParser(description=__doc__)

    parser.add_argument("jsonl", nargs="+", type=Path, help="candidate row JSONL")

    parser.add_argument("--baseline-book", required=True, type=Path)

    parser.add_argument("--targets", type=Path, default=ROOT / "config" / "targets.json")

    parser.add_argument("--bootstrap-samples", type=int, default=2000)

    parser.add_argument("--confidence", type=float, default=0.95)

    parser.add_argument("--json", action="store_true")

    return parser.parse_args()

def main() -> int:

    args = parse_args()

    rows = load_jsonl(args.jsonl)

    targets = load_targets(args.targets)

    baseline = load_baseline_book(args.baseline_book)

    results = evaluate_targets(

        rows,

        targets,

        baseline,

        bootstrap_samples=args.bootstrap_samples,

        confidence=args.confidence,

    )

    if args.json:

        print(json.dumps(results, sort_keys=True))

    else:

        print("| Target | Verdict | Ratio | Goal | n | Notes |")

        print("| --- | --- | ---: | ---: | ---: | --- |")

        for row in results:

            ratio = row.get("ratio")

            ratio_text = "" if ratio is None else f"{ratio:.3f}"

            notes = "; ".join(str(item) for item in row.get("messages", []))

            print(

                f"| `{row['target_id']}` | {row['verdict']} | {ratio_text} | "

                f"{row['goal_value']:.3f} | {row['n']} | {notes} |"

            )

    return 1 if any(row["verdict"] in {"RED", "INVALID"} for row in results) else 0

if __name__ == "__main__":

    raise SystemExit(main())