7deadlysinstest / scripts /validate.py
drewgrant616's picture
Upload 45 files
81b8593 verified
Raw History Blame Contribute Delete
6.5 kB
"""Validate schema, relationships, numerical results, reproducibility and SHA-256."""
import argparse
import hashlib
import json
from pathlib import Path
from jsonschema import Draft202012Validator
from generate import BASE, IDS, SNAPSHOT, build, classify, score
def sha(path):
return hashlib.sha256(path.read_bytes()).hexdigest()
def inventory():
# Reports are derived outputs and cannot recursively hash themselves.
return sorted(p for p in BASE.rglob('*') if p.is_file() and
not any(x in p.relative_to(BASE).parts for x in ['reports','__pycache__','.venv','.cache']) and
p.name not in ['SHA256SUMS','VALIDATION.md'])
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--write-hashes',action='store_true',help='Regenerate manifest only after reviewing intentional changes')
args = parser.parse_args()
schema = json.loads((BASE/'schema.json').read_text(encoding='utf-8'))
expected = build()
counts = {}
for name, generated in expected.items():
validator = Draft202012Validator(schema['$defs'][name])
path = BASE/'data'/f'{name}.jsonl'
assert path.read_bytes().endswith(b'\n')
rows = [json.loads(line) for line in path.read_text(encoding='utf-8').splitlines()]
for row in rows: validator.validate(row)
assert len({r['id'] for r in rows}) == len(rows), name
assert rows == generated, f'{name}: stale or non-reproducible data'
counts[name] = len(rows)
q = expected['questions']
assert len({x['text'] for x in q}) == 70
assert len(q) == 91
for mode, n, rev, minimum, version in [('quick',3,1,2,'quick-2'),('full',10,3,8,'full-1')]:
items = [x for x in q if x['mode']==mode]
assert [x['position'] for x in items] == list(range(1,len(items)+1))
for sid in IDS:
group = [x for x in items if x['dimension_id']==sid]
assert len(group)==n and sum(x['reverse'] for x in group)==rev
rows = [x for x in expected['scoring_cases'] if x['mode']==mode]
fingerprints = []
for row in rows:
assert row['version']==version and row['minimum_per_dimension']==minimum
answer = {a['question_id']:a['raw'] for a in row['answers']}
assert len(answer)==len(row['answers'])
assert set(answer)<=set(x['question_id'] for x in items)
assert row['addressed_count']==len(answer)
assert [s['id'] for s in row['scores']]==IDS
assert row['result_eligible'] == (row['coverage_complete'] and (mode=='quick' or len(answer)==70))
assert row['sensitivity']['changed']<=row['sensitivity']['checked']
for r in row['possible_completed_ranges']: assert 0<=r['low']<=r['high']<=100
fingerprints.append(json.dumps(row['answers'],sort_keys=True))
assert len(set(fingerprints))==len(rows), f'{mode}: duplicate answer fixture'
midpoint = next(x for x in rows if x['id']==f'{mode}:adjusted-uniform-2')
assert all(s['value']==50 and s['count']==n for s in midpoint['scores'])
assert midpoint['computed_profile']['title']=='An even constellation'
assert midpoint['computed_character']['title']=='A glorious mix'
assert midpoint['sensitivity']=={'checked':len(items)*2,'changed':0}
skipped = next(x for x in rows if x['id']==f'{mode}:minimum-with-skips')
assert skipped['result_eligible'] and all(s['count']==minimum for s in skipped['scores'])
omitted = next(x for x in rows if x['id']==f'{mode}:unaddressed')
assert omitted['coverage_complete'] and omitted['result_eligible']==(mode=='quick')
# Hand-specified branch expectations independently guard strict/inclusive boundaries.
boundary_expectations = {
'below-25':'A quieter side','at-25':'An even constellation',
'below-75':'An even constellation','at-75':'Many sides, strongly felt',
'spread-at-12-5':'An even constellation','spread-above-12-5':'Pride',
'top-below-50':'No strong leading side','top-at-50':'Pride',
'tie-at-6-25':'Pride + Envy','tie-outside-6-25':'Pride',
'stable-source-order':'Envy + Wrath','three-leaders':'A little more complex',
'quiet-before-even':'A quieter side','broad-before-even':'Many sides, strongly felt'}
for row in expected['classification_cases']:
assert row['expected_profile']['title']==boundary_expectations[row['id'].split(':')[1]]
quick = {x['question_id']:x for x in q if x['mode']=='quick'}
for x in (x for x in q if x['mode']=='full' and x['source_item_id'] in quick):
for field in ['text','reverse','theme','dimension_id']:
assert x[field]==quick[x['source_item_id']][field]
manifest = BASE/'SHA256SUMS'
if args.write_hashes:
manifest.write_text(''.join(f'{sha(p)} {p.relative_to(BASE).as_posix()}\n' for p in inventory()),encoding='utf-8',newline='\n')
entries = [line.split(' ',1) for line in manifest.read_text(encoding='utf-8').splitlines()]
assert {name for _,name in entries}=={p.relative_to(BASE).as_posix() for p in inventory()}, 'Manifest file coverage mismatch'
for digest, name in entries: assert sha(BASE/name)==digest, f'Hash mismatch: {name}'
report = {'status':'PASS','schema_version':SNAPSHOT['schema_version'],'snapshot_date':SNAPSHOT['snapshot_date'],
'row_counts':counts,'total_rows':sum(counts.values()),'distinct_question_texts':70,
'hash_files_checked':len(entries),'manifest_exclusions':['reports/','__pycache__/','.venv/','.cache/','SHA256SUMS','VALIDATION.md'],
'checks':['JSON Schema types and required keys','unique IDs','no duplicate attempt inputs within modes',
'mode/version correspondence','ordered question positions','question reuse foreign keys',
'reverse counts','answer domain and joins','independent hand-specified boundary expectations',
'coverage versus addressed eligibility','ranges and sensitivity','deterministic regeneration','SHA-256 coverage and contents'],
'publication_status':'local_only','rights_status':'no_additional_license_granted_pending_owner_review'}
(BASE/'reports').mkdir(exist_ok=True)
(BASE/'reports/validation.json').write_text(json.dumps(report,indent=2)+'\n',encoding='utf-8',newline='\n')
print(json.dumps(report,indent=2))
if __name__ == '__main__': main()