Blame for export_data.py

#!/usr/bin/env python3
"""Export recorded research results to deterministic Hugging Face JSON Lines tables.

This script transforms saved evidence only. It does not rerun or alter experiments.
"""
from __future__ import annotations
import argparse
import hashlib
import json
import math
import re
import statistics
from pathlib import Path

METHODS = ('frozen_syntax', 'one_shot', 'recursive', 'semantic_bfs')

def read_json(path):
    return json.loads(path.read_text(encoding='utf-8'))

def scalar_fields(value, prefix=''):
    """Retain named scalar leaves, never variable-length arrays."""
    result = {}
    for key, item in value.items():
        name = prefix + key
        if isinstance(item, dict):
            result.update(scalar_fields(item, name + '_'))
        elif item is None or isinstance(item, (str, bool, int, float)):
            result[name] = item
    return result

def ratio(a, b):
    return a / b if b else None

def canonical_json(value):
    return json.dumps(value, sort_keys=True, ensure_ascii=False, allow_nan=False)

def write_rows(path, rows):
    path.write_text(''.join(canonical_json(row) + '\n' for row in rows), encoding='utf-8')

def close(actual, expected):
    if isinstance(expected, float):
        assert math.isclose(actual, expected, rel_tol=1e-12, abs_tol=1e-12), (actual, expected)
    else:
        assert actual == expected, (actual, expected)

def export(source: Path, output: Path):
    summary = read_json(source / 'results/SUMMARY.json')
    tests = read_json(source / 'results/TEST_RESULTS.json')
    paths = sorted((source / 'results').glob('case_*.json'))
    paths.sort(key=lambda p: tuple(map(int, re.fullmatch(r'case_(\d+)_(\d+)\.json', p.name).groups())))
    cases, generations, source_cases = [], [], []
    expected_snapshots = set()
    case_keys = set()
    for path in paths:
        case = read_json(path)
        width, seed = case['width'], case['seed']
        assert (width, seed) not in case_keys, 'Duplicate case key'
        case_keys.add((width, seed))
        assert path.name == f'case_{width}_{seed}.json'
        history = case['history']
        assert [h['generation'] for h in history] == list(range(1, len(history)+1))
        targets = read_json(source / f'results/targets_{width}_{seed}.json')
        row = {
            'case_id': f'n{width}_seed{seed}',
            'width': width,
            'seed': seed,
            'source_case_path': path.relative_to(source).as_posix(),
            'source_case_sha256': hashlib.sha256(path.read_bytes()).hexdigest(),
            'scheduled_generations': len(history),
            'nonempty_updates': sum(h['upgrade']['accepted'] > 0 for h in history),
            'final_rules': case['final_rules'],
            'final_automaton_states': case['final_automaton_states'],
            'training_wall_seconds': case['training_wall_seconds'],
            'targets_total': len(targets),
        }
        row.update(scalar_fields(case['final_receipt'], 'final_receipt_'))
        row.update(scalar_fields(case['final_stats'], 'final_stats_'))
        for method in METHODS:
            data = case['methods'][method]
            for budget in ('fixed_products', 'fixed_prefix'):
                row.update(scalar_fields(data[budget], f'{method}_{budget}_'))
            row[f'{method}_median_prefix_seconds'] = data['median_prefix_seconds']
            for metric in ('known_semantics', 'targets_solved', 'wall_seconds', 'semantic_products'):
                row[f'{method}_wall_probe_median_{metric}'] = statistics.median(p[metric] for p in data['wall_probes'])
        for branch in ('continue_self_updates', 'freeze_meta_only'):
            row.update(scalar_fields(case['causal_fork'][branch], 'causal_' + branch + '_'))
        for base in ('frozen_syntax', 'one_shot', 'semantic_bfs'):
            row[f'ratio_recursive_to_{base}_fixed_products_semantics'] = ratio(
                case['methods']['recursive']['fixed_products']['known_semantics'],
                case['methods'][base]['fixed_products']['known_semantics'])
            row[f'speedup_recursive_vs_{base}_prefix'] = ratio(
                case['methods'][base]['median_prefix_seconds'],
                case['methods']['recursive']['median_prefix_seconds'])
        row['ratio_causal_continue_to_freeze_additional_semantics'] = ratio(
            case['causal_fork']['continue_self_updates']['additional_semantics'],
            case['causal_fork']['freeze_meta_only']['additional_semantics'])
        cases.append(row)
        source_cases.append(case)
        prior_rule_count = 0
        for item in history:
            generation = item['generation']
            snapshot_path = source / f'snapshots/n{width}_seed{seed}_g{generation}.json'
            expected_snapshots.add(snapshot_path.name)
            snapshot = read_json(snapshot_path)
            guard = snapshot['guard']
            assert all(len(g) == width for g in snapshot['generators'])
            assert snapshot['known_semantics'] == item['growth']['known_semantics']
            assert len(snapshot['rules']) == item['upgrade']['rules_total']
            assert len(snapshot['rules']) - prior_rule_count == item['upgrade']['accepted']
            assert len(guard['prefixes']) == item['upgrade']['automaton_states']
            assert guard['sha256'] == item['upgrade']['after_hash']
            assert snapshot['stats'] == item['stats']
            prior_rule_count = len(snapshot['rules'])
            gen = {
                'case_id': row['case_id'], 'width': width, 'seed': seed,
                'generation': generation,
                'source_snapshot_path': snapshot_path.relative_to(source).as_posix(),
                'source_snapshot_sha256': hashlib.sha256(snapshot_path.read_bytes()).hexdigest(),
                'known_semantics': snapshot['known_semantics'],
                'rules_total': len(snapshot['rules']),
                'automaton_states': len(guard['prefixes']),
                'generator_count': len(snapshot['generators']),
                'accepted_new_rules': item['upgrade']['accepted'],
                'nonempty_update': item['upgrade']['accepted'] > 0,
                'guard_sha256': guard['sha256'],
                'before_guard_sha256': item['upgrade']['before_hash'],
                'upgrade_wall_seconds': item['upgrade']['wall_seconds'],
            }
            gen.update(scalar_fields(item['growth'], 'growth_'))
            gen.update(scalar_fields(item['stats'], 'stats_'))
            generations.append(gen)
        assert prior_rule_count == row['final_rules']
    assert set(p.name for p in (source/'snapshots').glob('*.json')) == expected_snapshots
    assert len(cases) == summary['case_count'] == 24
    assert len(generations) == tests['accepted_snapshots'] == 192
    assert len({(g['width'],g['seed'],g['generation']) for g in generations}) == len(generations)
    close(sum(c['final_rules'] for c in cases), summary['total_final_rules'])
    close(sum(c['nonempty_updates'] for c in cases), summary['total_nonempty_updates'])
    for method in METHODS:
        recorded = summary['methods'][method]
        close(statistics.mean(c[f'{method}_fixed_products_known_semantics'] for c in cases), recorded['mean_semantics_fixed_products'])
        close(sum(c[f'{method}_fixed_products_targets_solved'] for c in cases), recorded['total_targets_solved'])
        close(sum(c['targets_total'] for c in cases), recorded['targets_total'])
        close(statistics.mean(c[f'{method}_fixed_prefix_semantic_products'] for c in cases), recorded['mean_products_fixed_prefix'])
        close(statistics.median(c[f'{method}_median_prefix_seconds'] for c in cases), recorded['median_prefix_seconds'])
    for base in ('frozen_syntax', 'one_shot', 'semantic_bfs'):
        close(statistics.mean(c[f'ratio_recursive_to_{base}_fixed_products_semantics'] for c in cases), summary['paired'][base]['mean_discovery_ratio'])
        close(statistics.median(c[f'speedup_recursive_vs_{base}_prefix'] for c in cases), summary['paired'][base]['median_prefix_speedup'])
    close(statistics.mean(c['ratio_causal_continue_to_freeze_additional_semantics'] for c in cases), summary['causal_additional_discovery_mean_ratio'])
    output.mkdir(parents=True, exist_ok=True)
    write_rows(output/'cases.jsonl', cases)
    write_rows(output/'generations.jsonl', generations)
    report = {
        'cases':len(cases), 'generations':len(generations),
        'nonempty_updates':sum(g['nonempty_update'] for g in generations),
        'final_rules_summed_across_cases':sum(c['final_rules'] for c in cases),
        'causal_continue_additional_semantics':sum(c['causal_continue_self_updates_additional_semantics'] for c in cases),
        'causal_freeze_additional_semantics':sum(c['causal_freeze_meta_only_additional_semantics'] for c in cases),
        'all_recorded_aggregate_checks_passed':True,
        'source_files_unchanged':True,
        'tables':{name:{'rows':len(rows),'columns':len(rows[0]),'sha256':hashlib.sha256((output/name).read_bytes()).hexdigest()} for name,rows in [('cases.jsonl',cases),('generations.jsonl',generations)]},
    }
    (output/'EXPORT_VALIDATION.json').write_text(json.dumps(report,indent=2,sort_keys=True)+'\n',encoding='utf-8')
    return report

def main():
    root=Path(__file__).resolve().parent
    parser=argparse.ArgumentParser(description=__doc__)
    parser.add_argument('--source',type=Path,default=root/'research')
    parser.add_argument('--output',type=Path,default=root/'data')
    args=parser.parse_args()
    print(json.dumps(export(args.source.resolve(),args.output.resolve()),indent=2,sort_keys=True))

if __name__=='__main__':
    main()