ocr-projects-archive / ocr-tech /tools /audit_limbu_book_missing_line_review_queue.py
prosperious's picture
Add 1000 files (ocr-tech/tests/test_audit_limbu_book_missed_line_review_coverage.py ...)
d82db76 verified
Raw History Blame Contribute Delete
12.8 kB
#!/usr/bin/env python3
"""Audit a Limbu book missing-line review queue artifact."""
from __future__ import annotations
import argparse
import csv
import json
import platform
from collections import Counter
from datetime import UTC, datetime
from pathlib import Path
from typing import Any
from ocrtech.manifest import sha256_file
REPORT_JSON = "book-missing-line-review-queue.json"
REVIEW_TSV = "book-missing-line-review-queue.tsv"
REVIEW_MD = "book-missing-line-review-queue.md"
REVIEW_HTML = "book-missing-line-review-dashboard.html"
COMMAND_TXT = "command.txt"
AUDIT_JSON = "book-missing-line-review-queue-audit.json"
REQUIRED_TSV_FIELDS = {
"book_page_number",
"book_page_id",
"book_page_output_dir",
"book_document_dir",
"review_priority",
"review_category",
"suggested_review_action",
"candidate_row_id",
"candidate_sample_id",
"candidate_line_id",
"candidate_structural_role",
"candidate_confidence",
"max_selected_iou",
"candidate_bbox",
"candidate_text",
"best_selected_text",
"crop_path",
"crop_sha256",
"page_context_path",
"page_context_sha256",
"review_status",
"corrected_text",
"reviewer",
"notes",
}
def _read_json_object(path: Path, issues: list[str], *, label: str) -> dict[str, Any] | None:
if not path.is_file():
issues.append(f"{label} is missing: {path}")
return None
try:
payload = json.loads(path.read_text(encoding="utf-8"))
except json.JSONDecodeError as exc:
issues.append(f"{label} is invalid JSON: {exc}")
return None
if not isinstance(payload, dict):
issues.append(f"{label} must contain a JSON object")
return None
return payload
def _read_tsv(path: Path, issues: list[str]) -> tuple[list[dict[str, str]], list[str]]:
if not path.is_file():
issues.append(f"review TSV is missing: {path}")
return [], []
try:
with path.open("r", encoding="utf-8", newline="") as handle:
reader = csv.DictReader(handle, delimiter="\t")
if reader.fieldnames is None:
issues.append(f"review TSV has no header: {path}")
return [], []
return [dict(row) for row in reader], list(reader.fieldnames)
except OSError as exc:
issues.append(f"review TSV could not be read: {type(exc).__name__}: {exc}")
return [], []
def _check_hash(path: Path, expected: object, issues: list[str], *, label: str) -> str | None:
if not path.is_file():
issues.append(f"{label} is missing: {path}")
return None
actual = sha256_file(path)
if not isinstance(expected, str) or not expected:
issues.append(f"{label} missing recorded SHA-256")
elif actual != expected:
issues.append(f"{label} SHA-256 mismatch: recorded={expected} actual={actual} path={path}")
return actual
def _path_from_report(report: dict[str, Any], key: str, issues: list[str]) -> Path:
value = report.get(key)
if not isinstance(value, str) or not value:
issues.append(f"report missing {key}")
return Path("")
return Path(value)
def _audit_rows(rows: list[dict[str, str]], fieldnames: list[str], report: dict[str, Any], issues: list[str]) -> dict[str, Any]:
missing_fields = sorted(REQUIRED_TSV_FIELDS - set(fieldnames))
if missing_fields:
issues.append(f"review TSV missing required fields: {missing_fields}")
row_ids: set[str] = set()
category_counts: Counter[str] = Counter()
priority_counts: Counter[str] = Counter()
pending_count = 0
accepted_count = 0
crop_count = 0
page_context_count = 0
for index, row in enumerate(rows, start=2):
row_id = str(row.get("candidate_row_id") or "").strip()
if not row_id:
issues.append(f"review TSV row {index} missing candidate_row_id")
elif row_id in row_ids:
issues.append(f"review TSV duplicate candidate_row_id: {row_id}")
row_ids.add(row_id)
page_id = str(row.get("book_page_id") or "").strip()
candidate_sample_id = str(row.get("candidate_sample_id") or "").strip()
if not page_id:
issues.append(f"review TSV row {index} missing book_page_id")
if candidate_sample_id != page_id:
issues.append(
f"review TSV row {index} candidate_sample_id does not match book_page_id: "
f"{candidate_sample_id!r} != {page_id!r}"
)
status = str(row.get("review_status") or "").strip().lower()
if status == "pending":
pending_count += 1
elif status in {"accepted", "approved", "corrected", "verified", "done"}:
accepted_count += 1
if not str(row.get("corrected_text") or "").strip():
issues.append(f"review TSV row {index} accepted without corrected_text")
else:
issues.append(f"review TSV row {index} has unsupported review_status: {status!r}")
category = str(row.get("review_category") or "").strip()
if not category:
issues.append(f"review TSV row {index} missing review_category")
else:
category_counts[category] += 1
priority = str(row.get("review_priority") or "").strip()
if not priority:
issues.append(f"review TSV row {index} missing review_priority")
else:
priority_counts[priority] += 1
crop_path = Path(str(row.get("crop_path") or ""))
expected_crop_sha = str(row.get("crop_sha256") or "")
if not crop_path.is_file():
issues.append(f"review TSV row {index} crop_path is missing: {crop_path}")
else:
crop_count += 1
actual_crop_sha = sha256_file(crop_path)
if actual_crop_sha != expected_crop_sha:
issues.append(
f"review TSV row {index} crop SHA-256 mismatch: "
f"recorded={expected_crop_sha} actual={actual_crop_sha} path={crop_path}"
)
page_context_path_text = str(row.get("page_context_path") or "").strip()
page_context_sha = str(row.get("page_context_sha256") or "").strip()
if page_context_path_text or page_context_sha:
page_context_path = Path(page_context_path_text)
if not page_context_path.is_file():
issues.append(f"review TSV row {index} page_context_path is missing: {page_context_path}")
else:
page_context_count += 1
actual_context_sha = sha256_file(page_context_path)
if actual_context_sha != page_context_sha:
issues.append(
f"review TSV row {index} page context SHA-256 mismatch: "
f"recorded={page_context_sha} actual={actual_context_sha} path={page_context_path}"
)
if report.get("review_row_count") != len(rows):
issues.append(f"review_row_count mismatch: recorded={report.get('review_row_count')} actual={len(rows)}")
recorded_category_counts = report.get("category_counts")
if recorded_category_counts != dict(sorted(category_counts.items())):
issues.append(f"category_counts mismatch: recorded={recorded_category_counts} actual={dict(sorted(category_counts.items()))}")
recorded_priority_counts = report.get("priority_counts")
if recorded_priority_counts != dict(sorted(priority_counts.items())):
issues.append(f"priority_counts mismatch: recorded={recorded_priority_counts} actual={dict(sorted(priority_counts.items()))}")
recorded_page_context_count = report.get("page_context_count")
if recorded_page_context_count is not None and recorded_page_context_count != page_context_count:
issues.append(f"page_context_count mismatch: recorded={recorded_page_context_count} actual={page_context_count}")
return {
"row_count": len(rows),
"pending_count": pending_count,
"accepted_count": accepted_count,
"crop_count": crop_count,
"page_context_count": page_context_count,
"category_counts": dict(sorted(category_counts.items())),
"priority_counts": dict(sorted(priority_counts.items())),
}
def audit_queue(queue_dir: str | Path, audit_dir: str | Path | None = None) -> dict[str, Any]:
queue_root = Path(queue_dir)
issues: list[str] = []
warnings: list[str] = []
report_path = queue_root / REPORT_JSON
report = _read_json_object(report_path, issues, label="book missing-line queue report")
if report is None:
audit = _audit_report(queue_root, report_path, issues, warnings, {})
_write_audit(audit, Path(audit_dir) if audit_dir is not None else queue_root)
return audit
if report.get("tool") != "tools/prepare_limbu_book_missing_line_review_queue.py":
issues.append(f"unexpected report tool: {report.get('tool')!r}")
if report.get("out") != str(queue_root):
issues.append(f"report out mismatch: recorded={report.get('out')!r} actual={str(queue_root)!r}")
book_run = _path_from_report(report, "book_run", issues)
candidate_review_file = _path_from_report(report, "candidate_review_file", issues)
review_tsv = _path_from_report(report, "review_queue_tsv", issues)
review_md = _path_from_report(report, "review_queue_md", issues)
review_dashboard_value = report.get("review_dashboard")
command = _path_from_report(report, "command", issues)
_check_hash(book_run, report.get("book_run_sha256"), issues, label="book-run.json")
_check_hash(candidate_review_file, report.get("candidate_review_file_sha256"), issues, label="candidate review file")
_check_hash(review_tsv, report.get("review_queue_tsv_sha256"), issues, label="review TSV")
_check_hash(review_md, report.get("review_queue_md_sha256"), issues, label="review Markdown")
if isinstance(review_dashboard_value, str) and review_dashboard_value:
_check_hash(Path(review_dashboard_value), report.get("review_dashboard_sha256"), issues, label="review dashboard")
else:
warnings.append("report does not include a review dashboard")
_check_hash(command, report.get("command_sha256"), issues, label="command.txt")
rows, fieldnames = _read_tsv(review_tsv, issues)
row_summary = _audit_rows(rows, fieldnames, report, issues)
recorded_dashboard_cards = report.get("review_dashboard_card_count")
if recorded_dashboard_cards is not None and recorded_dashboard_cards != row_summary.get("row_count"):
issues.append(
"review_dashboard_card_count mismatch: "
f"recorded={recorded_dashboard_cards} actual={row_summary.get('row_count')}"
)
boundary = str(report.get("boundary") or "")
if "review evidence only" not in boundary:
warnings.append("report boundary does not explicitly say review evidence only")
if any(row.get("review_status") != "pending" for row in rows):
warnings.append("review queue contains non-pending rows; regenerate a fresh queue after review before treating report hashes as current")
audit = _audit_report(queue_root, report_path, issues, warnings, row_summary)
_write_audit(audit, Path(audit_dir) if audit_dir is not None else queue_root)
return audit
def _audit_report(
queue_dir: Path,
report_path: Path,
issues: list[str],
warnings: list[str],
row_summary: dict[str, Any],
) -> dict[str, Any]:
return {
"created_at_utc": datetime.now(UTC).isoformat(),
"machine": platform.node(),
"tool": "tools/audit_limbu_book_missing_line_review_queue.py",
"queue_dir": str(queue_dir),
"report": str(report_path),
"report_sha256": sha256_file(report_path) if report_path.is_file() else None,
"audit_passed": not issues,
"operational_review_queue_safe": not issues and row_summary.get("accepted_count", 0) == 0,
"issues": issues,
"warnings": warnings,
**row_summary,
}
def _write_audit(audit: dict[str, Any], out: Path) -> None:
out.mkdir(parents=True, exist_ok=True)
(out / AUDIT_JSON).write_text(json.dumps(audit, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
def main(argv: list[str] | None = None) -> int:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--queue-dir", type=Path, required=True)
parser.add_argument("--audit-dir", type=Path)
args = parser.parse_args(argv)
result = audit_queue(args.queue_dir, args.audit_dir)
print(f"audit_passed={result['audit_passed']} issues={len(result['issues'])}")
return 0 if result["audit_passed"] else 1
if __name__ == "__main__":
raise SystemExit(main())