ai-labs / validate_evidence.py
ProCreations's picture
Publish MSNN exact native reproduction
0cf1bfb verified
Raw
History Blame Contribute Delete
1.3 kB
#!/usr/bin/env python3
import json,os,subprocess,sys,tempfile
from pathlib import Path
r=Path(__file__).resolve().parent
a=json.loads((r/'outputs/results.json').read_text());b=json.loads((r/'replay_a/results.json').read_text());c=json.loads((r/'replay_b/results.json').read_text());e=json.loads((r/'EVIDENCE_MATRIX.json').read_text())
assert a==b==c and a['all_gates_pass'] and a['summary']=={'passed':33,'total':33}
assert len(e['claims'])==6 and [x['assessment'] for x in e['claims']]==['verified','verified','verified','verified','falsified','verified']
assert all(x['registered_system_executed'] and x['destructive_control_executed'] for x in e['claims'])
with tempfile.TemporaryDirectory() as t:
q=Path(t)/'fresh';subprocess.run([sys.executable,'-W','error',str(r/'reproduce.py'),'--bundle-root',str(r),'--output-dir',str(q)],check=True,env={**os.environ,'PYTHONDONTWRITEBYTECODE':'1','PYTHONHASHSEED':'0','PYTHONWARNINGS':'error'},stdout=subprocess.DEVNULL)
assert {p.name for p in q.iterdir()}=={p.name for p in (r/'outputs').iterdir()}
assert all((q/p.name).read_bytes()==p.read_bytes() for p in (r/'outputs').iterdir())
print('evidence validated: 6 exact claims, 33 gates, native estimator, exhaustive MCAR factor, complete tables, biclique, controls, and four byte-identical runs')