scripts/run-ocr-benchmark.py

#!/usr/bin/env python3
"""
Score OCR predictions against a ground-truth annotation file.

This gives us a stable benchmark before wiring OCR into production indexing.
"""

from __future__ import annotations

import argparse
import json
import sys
from pathlib import Path


ROOT_DIR = Path(__file__).resolve().parents[1]
BACKEND_DIR = ROOT_DIR / "backend"
if str(BACKEND_DIR) not in sys.path:
    sys.path.insert(0, str(BACKEND_DIR))

from ocr_ground_truth import build_benchmark_report, load_ground_truth, load_predictions  # noqa: E402


def build_parser() -> argparse.ArgumentParser:
    parser = argparse.ArgumentParser(description="Score OCR predictions against Search-UI ground truth.")
    parser.add_argument("--ground-truth", required=True, type=Path, help="Ground-truth JSON file")
    parser.add_argument("--predictions", required=True, type=Path, help="Predictions JSON file")
    parser.add_argument(
        "--report",
        type=Path,
        default=ROOT_DIR / "docs" / "reports" / "ocr-benchmark-report.json",
        help="Where to write the report JSON",
    )
    return parser


def main() -> int:
    args = build_parser().parse_args()
    annotations = load_ground_truth(args.ground_truth)
    predictions = load_predictions(args.predictions)
    report = build_benchmark_report(annotations, predictions)

    args.report.parent.mkdir(parents=True, exist_ok=True)
    args.report.write_text(json.dumps(report, indent=2, ensure_ascii=False), encoding="utf-8")

    summary = report["summary"]
    print(
        "OCR benchmark:"
        f" segments={summary['segments']}"
        f" exact_match_rate={summary['exact_match_rate']}"
        f" avg_similarity={summary['average_normalized_edit_similarity']}"
        f" junk_rate={summary['junk_prediction_rate']}"
        f" missing_rate={summary['missing_prediction_rate']}"
    )
    print(f"Report: {args.report}")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())