Bind synthetic edge cases and GPU comparisons to original CPU oracle

models/male-cns-lif/conditioning-acceleration-v1/audit_integrity.py

@@ -0,0 +1,49 @@
+"""Read-only rehash of the saved Phase4AR shutdown inventory and registered sources."""
+import argparse
+from datetime import datetime, timezone
+import hashlib
+import json
+from pathlib import Path
+import subprocess
+import time
+
+INVENTORY = Path('/home/qishizhe/phase4ar-user-shutdown-20260930T203113Z/saved-files.sha256.jsonl')
+INVENTORY_SHA = 'd7b675b0d0d4647d8fe2fcf9616b9a599e5bc3c15565e604e0d66a046afb835a'
+REG = Path('/data/digital-drosophila/experiments/phase4ar-normal-conditioning/protocol/preregistered.json')
+REG_SHA = 'b63e6281b6505ca3b60acb467dd0b3fae274cf513dcbfd653649ba52e9ba9740'
+
+def sha(path):
+    h = hashlib.sha256()
+    with Path(path).open('rb') as f:
+        for chunk in iter(lambda: f.read(8 << 20), b''): h.update(chunk)
+    return h.hexdigest()
+
+p = argparse.ArgumentParser()
+p.add_argument('--output', type=Path, required=True)
+a = p.parse_args()
+assert not a.output.exists() and a.output.parent.is_dir()
+start = time.perf_counter()
+assert sha(INVENTORY) == INVENTORY_SHA and sha(REG) == REG_SHA
+records = [json.loads(line) for line in INVENTORY.read_text().splitlines()]
+differences = []
+for row in records:
+    path = Path(row['path'])
+    if not path.is_file() or path.stat().st_size != row['bytes'] or sha(path) != row['sha256']:
+        differences.append(str(path))
+reg = json.loads(REG.read_text())
+source_mismatches = [path for path, expected in reg['source_sha256'].items() if sha(path) != expected]
+head = subprocess.check_output(['git', '-C', '/data/digital-drosophila', 'rev-parse', 'HEAD'], text=True).strip()
+status = subprocess.check_output(['git', '-C', '/data/digital-drosophila', 'status', '--porcelain'], text=True)
+summary = dict(status='PASS' if not differences and not source_mismatches and status == '' else 'FAIL',
+    created_utc=datetime.now(timezone.utc).isoformat(), files_checked=len(records),
+    bytes_checked=sum(r['bytes'] for r in records), inventory_sha256=INVENTORY_SHA,
+    registration_sha256=REG_SHA, registered_science_commit=reg['source_code_commit'],
+    sources_checked=len(reg['source_sha256']), file_mismatches=differences,
+    source_mismatches=source_mismatches, original_git_head=head, original_git_status=status,
+    wall_seconds=time.perf_counter()-start,
+    scope='Saved Phase4AR experiment/control files and registered source files; NOT a new Phase1-4AQ numerical regression',
+    training_epochs_cpu_verified_unchanged=1033, probes_completed_unchanged=138,
+    original_experiment_resumed=False)
+with a.output.open('x') as f: json.dump(summary, f, indent=2)
+print(json.dumps(summary), flush=True)
+raise SystemExit(0 if summary['status'] == 'PASS' else 1)