#!/usr/bin/env python3

"""Compare matched τ³ baseline, generic-review, and Shield summaries."""

from __future__ import annotations

import argparse

import json

from pathlib import Path

from typing import Any

def parse_args() -> argparse.Namespace:

    parser = argparse.ArgumentParser()

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

    parser.add_argument("--generic-review", required=True)

    parser.add_argument("--shield", required=True)

    return parser.parse_args()

def load(path: str) -> dict[str, Any]:

    return json.loads(Path(path).read_text(encoding="utf-8"))

def keyed(value: dict[str, Any]) -> dict[tuple[str, int], float]:

    return {

        (str(row["taskId"]), int(row["trial"])): float(row["reward"])

        for row in value["simulations"]

        if row.get("reward") is not None

    }

def stats(name: str, value: dict[str, Any]) -> str:

    count = int(value["simulationCount"])

    average = value.get("averageReward")

    passes = int(value["passCount"])

    reward_text = "n/a" if average is None else f"{float(average):.4f}"

    pass_text = "n/a" if count == 0 else f"{passes / count:.1%}"

    return f"{name:16} n={count:4d}  reward={reward_text:>6}  pass={pass_text}"

def paired_delta(left: dict[str, Any], right: dict[str, Any]) -> tuple[float, int]:

    left_rows = keyed(left)

    right_rows = keyed(right)

    common = sorted(left_rows.keys() & right_rows.keys())

    if not common:

        raise ValueError("Runs have no matching task/trial pairs")

    return sum(right_rows[key] - left_rows[key] for key in common) / len(common), len(common)

def main() -> None:

    args = parse_args()

    baseline = load(args.baseline)

    generic = load(args.generic_review)

    shield = load(args.shield)

    signatures = {

        (value["domain"], value["agentModel"], value["userModel"], value["seed"], value["numTrials"])

        for value in (baseline, generic, shield)

    }

    if len(signatures) != 1:

        raise ValueError("Runs are not comparable: domain/model/seed/trial settings differ")

    vs_baseline, baseline_pairs = paired_delta(baseline, shield)

    vs_generic, generic_pairs = paired_delta(generic, shield)

    print(stats("baseline", baseline))

    print(stats("generic review", generic))

    print(stats("OpenClasp Shield", shield))

    print()

    print(f"Shield − baseline:       {vs_baseline:+.4f} reward ({baseline_pairs} paired runs)")

    print(f"Shield − generic review: {vs_generic:+.4f} reward ({generic_pairs} paired runs)")

if __name__ == "__main__":

    main()