Download audit.py from pranasara-intelligence-lab/sleep-discovery-sample: direct link, hf CLI and curl.
- Browser
- Download file 5.82 kB
-
https://huggingface.co/datasets/pranasara-intelligence-lab/sleep-discovery-sample/resolve/main/audit.py
- Command line
-
hf download hf://datasets/pranasara-intelligence-lab/sleep-discovery-sample/audit.py
-
curl -L -o audit.py https://huggingface.co/datasets/pranasara-intelligence-lab/sleep-discovery-sample/resolve/main/audit.py
5.82 kB
| #!/usr/bin/env python3 | |
| """Deterministic title/metadata audit; never a clinical classification.""" | |
| import hashlib | |
| import html | |
| import json | |
| import re | |
| import unicodedata | |
| from collections import Counter, defaultdict | |
| from pathlib import Path | |
| ROOT = Path(__file__).resolve().parent | |
| DATA = ROOT / 'sleep-discovery.jsonl' | |
| EXPECTED_SHA256 = '1db7da5e2d54812b5ce9eee93615f2f8815e6e9d529bc69cd1b2cdabbdf521d3' | |
| FIELDS = ('id', 'doi', 'titles', 'author_names', 'journal_names', 'publisher', 'crossref_type', 'publication_date_parts', 'doi_url', 'query_memberships', 'source_endpoint', 'retrieved_at_utc') | |
| RELEVANCE = re.compile(r'\b(?:sleep\w*|insomnia|circadian|apnea|apnoea|sono)\b', re.I) | |
| def normalized_title(title): | |
| text = unicodedata.normalize('NFKC', html.unescape(title)).casefold() | |
| return ' '.join(re.findall(r'\w+', text)) | |
| def main(): | |
| before = DATA.read_bytes() | |
| digest = hashlib.sha256(before).hexdigest() | |
| if digest != EXPECTED_SHA256: | |
| raise ValueError('Pinned sample checksum mismatch; audit aborted') | |
| rows = [json.loads(line) for line in before.decode('utf-8').splitlines()] | |
| if len(rows) != 39: | |
| raise ValueError('Expected exactly 39 records') | |
| doi_groups, title_groups = defaultdict(set), defaultdict(set) | |
| years, queries = Counter(), Counter() | |
| audits = [] | |
| for row in rows: | |
| ident = row['id'] | |
| doi_groups[row['doi'].casefold()].add(ident) | |
| titles = row.get('titles', []) | |
| for title in titles: | |
| title_groups[normalized_title(title)].add(ident) | |
| dates = row.get('publication_date_parts', []) | |
| years[str(dates[0][0]) if dates and dates[0] else 'missing'] += 1 | |
| queries.update(set(row.get('query_memberships', []))) | |
| flags = [] | |
| if row['doi'].casefold().startswith('10.31525/ct1-'): | |
| flags.append({'code': 'candidate_registry_record', 'reason': 'DOI prefix 10.31525/ct1-; registry identity requires verification'}) | |
| for title in titles: | |
| if re.search(r'\b(?:rat|canine)\b', title, re.I): | |
| flags.append({'code': 'title_animal_model', 'reason': 'Title explicitly contains Rat or Canine; not a verified study classification'}) | |
| if re.match(r'^0442\b', title): | |
| flags.append({'code': 'candidate_conference_abstract', 'reason': 'Title starts with numeric conference-style marker 0442; publication type requires verification'}) | |
| if re.search(r'</?[A-Za-z][^>]*>', title): | |
| flags.append({'code': 'title_markup', 'reason': 'HTML/XML-like tag detected'}) | |
| if re.search(r'<|>', title): | |
| flags.append({'code': 'title_literal_angle_bracket', 'reason': 'Literal angle bracket present; this is not necessarily markup'}) | |
| if len(title) > 150: | |
| flags.append({'code': 'title_long', 'reason': 'Title length exceeds 150 characters; formatting/repetition merits inspection'}) | |
| if len(re.sub(r'[^A-Za-z]', '', title)) >= 20 and title.upper() == title: | |
| flags.append({'code': 'title_all_caps', 'reason': 'At least 20 alphabetic characters, all uppercase'}) | |
| if title != title.strip() or re.search(r'\s{2,}', title): | |
| flags.append({'code': 'title_whitespace', 'reason': 'Leading/trailing or repeated whitespace'}) | |
| evidence = sorted({match.group(0).casefold() for title in titles for match in RELEVANCE.finditer(title)}) | |
| audits.append({'id': ident, 'title_relevance_terms': evidence, 'direct_title_relevance': bool(evidence), 'flags': flags}) | |
| duplicates = lambda groups: [{'value': key, 'record_ids': sorted(ids)} for key, ids in sorted(groups.items()) if len(ids) > 1] | |
| doi_values = [r['doi'].casefold() for r in rows] | |
| after = hashlib.sha256(DATA.read_bytes()).hexdigest() | |
| if digest != after: | |
| raise RuntimeError('Source changed during audit') | |
| report = { | |
| 'audit_method': 'deterministic title/metadata rules only; no semantic, clinical, expert or publisher credibility review', | |
| 'record_count': len(rows), 'metric_denominator': '39 records unless explicitly stated otherwise; query memberships may overlap', | |
| 'sha256_before': digest, 'sha256_after': after, 'source_unchanged': True, | |
| 'field_completeness': {field: {'missing_key_count': sum(field not in r for r in rows), 'empty_value_count': sum(field in r and not r[field] for r in rows), 'nonempty_count': sum(bool(r.get(field)) for r in rows)} for field in FIELDS}, | |
| 'doi_duplicate_record_count': len(doi_values) - len(set(doi_values)), | |
| 'doi_duplicate_groups': duplicates(doi_groups), 'normalized_title_duplicate_groups': duplicates(title_groups), | |
| 'title_normalization': 'HTML-unescape, Unicode NFKC, casefold, retain Unicode word tokens separated by one space; compare across different record IDs', | |
| 'publication_year_distribution': dict(sorted(years.items())), 'query_membership_counts': dict(sorted(queries.items())), | |
| 'direct_title_relevance_count': sum(r['direct_title_relevance'] for r in audits), | |
| 'flag_record_counts': dict(sorted(Counter(code for r in audits for code in {f['code'] for f in r['flags']}).items())), | |
| 'crossref_type_counts': dict(Counter(r['crossref_type'] for r in rows)), | |
| 'classification_limit': 'Crossref journal-article is a source metadata type, not clinical publication classification. All candidate flags require verification; no records removed.', | |
| 'record_audits': audits, | |
| } | |
| (ROOT / 'quality-report.json').write_text(json.dumps(report, ensure_ascii=False, indent=2) + '\n', encoding='utf-8') | |
| print(json.dumps({k: report[k] for k in ('record_count', 'direct_title_relevance_count', 'flag_record_counts', 'source_unchanged')}, indent=2)) | |
| if __name__ == '__main__': | |
| main() | |