"""Character error rate (CER) between a reference baseline and a candidate.

CER = levenshtein(reference, hypothesis) / len(reference).
Used to gate work-reduction changes: keep a change only if CER <= 0.005.

Usage:
    python scripts/cer_divergence.py baseline_outputs/ candidate_outputs/
"""
from __future__ import annotations

import argparse
import json
from pathlib import Path


def _levenshtein(a: str, b: str) -> int:
    if a == b:
        return 0
    if not a:
        return len(b)
    if not b:
        return len(a)
    prev = list(range(len(b) + 1))
    for i, ca in enumerate(a, 1):
        cur = [i]
        for j, cb in enumerate(b, 1):
            cost = 0 if ca == cb else 1
            cur.append(min(prev[j] + 1, cur[j - 1] + 1, prev[j - 1] + cost))
        prev = cur
    return prev[-1]


def cer(reference: str, hypothesis: str) -> float:
    if not reference:
        return 0.0 if not hypothesis else 1.0
    return _levenshtein(reference, hypothesis) / len(reference)


def cer_over_dirs(baseline_dir: Path, candidate_dir: Path) -> dict:
    per_file = {}
    for ref_path in sorted(baseline_dir.glob("*.txt")):
        cand_path = candidate_dir / ref_path.name
        ref = ref_path.read_text(encoding="utf-8")
        hyp = cand_path.read_text(encoding="utf-8") if cand_path.exists() else ""
        per_file[ref_path.name] = cer(ref, hyp)
    mean = sum(per_file.values()) / len(per_file) if per_file else 0.0
    return {"mean_cer": mean, "max_cer": max(per_file.values(), default=0.0), "per_file": per_file}


def main() -> None:
    parser = argparse.ArgumentParser(description="CER divergence between two output dirs.")
    parser.add_argument("baseline_dir", type=Path)
    parser.add_argument("candidate_dir", type=Path)
    args = parser.parse_args()
    print(json.dumps(cer_over_dirs(args.baseline_dir, args.candidate_dir), indent=2))


if __name__ == "__main__":
    main()