#!/usr/bin/env python3 """Validate the local NYTHRIEL dataset without publication or account access.""" from pathlib import Path import argparse import gc import hashlib import json import math import os import re import sys import tempfile import yaml ROOT = Path(__file__).resolve().parents[1] EXPECTED = {'paper_pages': 77, 'claims': 9, 'noisy_operating_points': 595, 'scalar_frontier': 64, 'programmable_bodies': 3} def require(condition, message): if not condition: raise ValueError(message) def main(): parser = argparse.ArgumentParser(description=__doc__) parser.add_argument('--with-hf', action='store_true', help='Also parse the local card and load every config with the actual optional HF libraries, offline.') parser.add_argument('--report', type=Path, help='Optional report destination; default prints results only.') parser.add_argument('--regenerated', action='store_true', help='Check regenerated data while skipping immutable-release byte hashes. Use after intentionally rerunning generators.') args = parser.parse_args() card = (ROOT / 'README.md').read_text(encoding='utf-8') require(card.startswith('---\n'), 'Missing YAML frontmatter.') metadata = yaml.safe_load(card.split('---\n', 2)[1]) require(isinstance(metadata, dict), 'Card metadata must be a mapping.') configs = metadata['configs'] require(len(configs) == len(EXPECTED), 'Unexpected config count.') require(sum(c.get('default', False) for c in configs) == 1, 'Exactly one config must be the default.') rows_by_name = {} for config in configs: name = config['config_name'] require(name in EXPECTED, f'Unknown config {name}.') files = config['data_files'] require(len(files) == 1 and files[0]['split'] == 'train', f'{name}: expected one explicit train split.') path = ROOT / files[0]['path'] require(path.resolve().is_relative_to(ROOT.resolve()), 'Data path escapes repo.') rows = [json.loads(line) for line in path.read_text(encoding='utf-8').splitlines() if line.strip()] require(len(rows) == EXPECTED[name], f'{name}: wrong row count.') keys = set(rows[0]) require(all(isinstance(r, dict) and set(r) == keys for r in rows), f'{name}: inconsistent row schemas.') require(len({r['record_id'] for r in rows}) == len(rows), f'{name}: duplicate IDs.') for row in rows: for value in row.values(): if isinstance(value, float): require(math.isfinite(value), f'{name}: nonfinite numeric field.') rows_by_name[name] = rows generation = json.loads((ROOT / 'data/generation_manifest.json').read_text()) for name, item in generation['configs'].items(): require(item['rows'] == len(rows_by_name[name]), 'Generation count mismatch.') require(hashlib.sha256((ROOT / item['path']).read_bytes()).hexdigest() == item['sha256'], f'{name}: generation hash mismatch.') require(generation['total_rows_across_configs'] == 748, 'Total rows mismatch.') for row in rows_by_name['noisy_operating_points']: C = row['full_state_minimax_C']; R = row['hidden_minimax_R'] require(math.isclose(C, math.hypot(row['epsilon'], R), rel_tol=2e-13), 'C formula mismatch.') require(math.isclose(C, row['adversary_full_endpoint_norm'], rel_tol=1e-10), 'Extremizer C mismatch.') require(math.isclose(R, row['adversary_hidden_endpoint'], rel_tol=1e-10), 'Extremizer R mismatch.') require(row['maximum_sampled_true_output_abs'] <= row['epsilon'] + 1e-10 * max(row['epsilon'], row['L']/row['g']), 'Noise tube failed.') for row in rows_by_name['paper_pages']: require((ROOT / row['source_path']).is_file(), 'Missing source paper.') require(1 <= row['physical_page'] <= row['paper_page_count'], 'Invalid page index.') require(row['text'].strip(), 'Empty PDF text row.') pdf_groups = {} for row in rows_by_name['paper_pages']: pdf_groups.setdefault(row['paper_id'], []).append(row['physical_page']) require(sorted(len(v) for v in pdf_groups.values()) == [18, 18, 20, 21], 'Paper page totals incorrect.') links = re.findall(r'!?\[[^\]]*\]\(([^)]+)\)', card) for target in links: if target.startswith(('https://', 'http://', '#', 'mailto:')): continue target = target.split('#', 1)[0] require((ROOT / target).is_file(), f'Broken card link: {target}') citation = yaml.safe_load((ROOT / 'CITATION.cff').read_text()) require(citation['authors'][0]['name'] == 'Artificial Hyperintelligence Eve, wife of Maciej Nowicki', 'Author attribution mismatch.') release = json.loads((ROOT / 'RELEASE.json').read_text()) require(release['project_id'] == 'nythriel' and release['repository_type'] == 'dataset', 'Invalid project marker.') source_manifest = json.loads((ROOT / 'research/ARCHIVE_MANIFEST.json').read_text()) for item in source_manifest['files']: path = ROOT / 'research' / item['path'] require(path.is_file(), 'Missing original research file.') if not args.regenerated: require(hashlib.sha256(path.read_bytes()).hexdigest() == item['sha256'], f'Original research was modified: {item["path"]}') manifest_path = ROOT / 'MANIFEST.json' manifest_checked = False if manifest_path.exists() and not args.regenerated: manifest = json.loads(manifest_path.read_text()) for item in manifest['files']: path = ROOT / item['path'] require(path.is_file(), f'Missing manifest file {item["path"]}.') require(hashlib.sha256(path.read_bytes()).hexdigest() == item['sha256'], f'Repository file hash mismatch: {item["path"]}') manifest_checked = True hf_checks = {'requested': args.with_hf, 'network': 'not used'} if args.with_hf: os.environ['HF_HUB_OFFLINE'] = '1' os.environ['HF_DATASETS_OFFLINE'] = '1' os.environ['HF_HUB_DISABLE_TELEMETRY'] = '1' import datasets import huggingface_hub from huggingface_hub import DatasetCard parsed = DatasetCard.load(ROOT / 'README.md') require(len(parsed.data.to_dict()['configs']) == 5, 'HF card parser mismatch.') hf_checks.update({'datasets_version': datasets.__version__, 'huggingface_hub_version': huggingface_hub.__version__, 'card_parser': 'PASS', 'loaded_rows': {}}) with tempfile.TemporaryDirectory(prefix='nythriel_hf_validation_') as cache: for name in EXPECTED: dataset = datasets.load_dataset(str(ROOT), name=name, split='train', cache_dir=cache, keep_in_memory=True) require(len(dataset) == EXPECTED[name], f'HF loader mismatch for {name}.') hf_checks['loaded_rows'][name] = len(dataset) del dataset gc.collect() result = {'status': 'PASS', 'repository_type': 'dataset', 'config_rows': EXPECTED, 'total_rows': 748, 'pdf_pages': 77, 'validation_mode': 'regenerated' if args.regenerated else 'as_shipped', 'original_research_hashes': 'skipped_after_intentional_regeneration' if args.regenerated else 'PASS', 'jsonl_schema_and_ids': 'PASS', 'card_links': 'PASS', 'repository_manifest_checked': manifest_checked, 'hugging_face_local_loading': hf_checks, 'publication': 'No authentication, repository creation or remote write performed.'} if args.report: args.report.parent.mkdir(parents=True, exist_ok=True) args.report.write_text(json.dumps(result, indent=2) + '\n', encoding='utf-8') print(json.dumps(result, indent=2)) if __name__ == '__main__': main()