Files
MeltMind/tools/compare_g1_band25_reviews.py

82 lines
4.5 KiB
Python

"""Compare two frozen anonymous CSVs without opening any identity/pose mapping."""
from pathlib import Path
import collections
import csv
import datetime
import hashlib
import json
import shutil
import subprocess
ROOT=Path('artifacts/s2-g1-band25-2026-11-15')
DEST=Path('artifacts/s2-g1-band25-review-comparison-2026-11-15')
CLAUDE='dfe2c65b9b8afe583f43dd6df8e0afdf520aa1d0'
MOLD='36f2462eb95ad0d67274a49ba5c431fe728fab2f'
EXPECTED={
'mold/frozen-review.csv':'6479eebdfc09629c7188f18f888494c5d9bea8dd160d92763813b15ef813ce18',
'mold/reviewer-attestation.md':'f317b9c4f790943b45607452c9c381624b17ad5f6fee6b009658415fe4c38085',
'claude/band25-review.csv':'93096734186c1247e46e98f26c7c9b0f7a4987c8d8d588e8147110ae3880be2e',
'claude/reviewer-attestation.md':'bc5df5dbcf02176b5292dc455aaf9fae34ef14bf65208645d9a7bfd4319af5a4',
}
def sha(p):return hashlib.sha256(Path(p).read_bytes()).hexdigest()
def main():
DEST.mkdir(exist_ok=False)
sources={
'mold':ROOT/'custodian/mold-frozen-received/extracted',
'claude':ROOT/'custodian/claude-frozen-received',
}
for name,folder in sources.items():
out=DEST/'sources'/name;out.mkdir(parents=True)
files=['frozen-review.csv','reviewer-attestation.md','frozen-review.bundle'] if name=='mold' else ['band25-review.csv','reviewer-attestation.md','tools/dims.py','tools/inv.py','tools/web2.py']
for file in files:
target=out/file;target.parent.mkdir(parents=True,exist_ok=True);shutil.copy2(folder/file,target)
commit_source=folder.parent/'original-commit.txt' if name=='mold' else folder/'original-commit.txt'
shutil.copy2(commit_source,out/'original-commit.txt')
for name,digest in EXPECTED.items():assert sha(DEST/'sources'/name)==digest,name
blank=list(csv.DictReader((ROOT/'blind-package/blank-review.csv').open()))
ids=[r['anonymous_pair_id'] for r in blank]
assert len(ids)==len(set(ids))==25
reviews={}
for who,file in [('mold','frozen-review.csv'),('claude','band25-review.csv')]:
rows=list(csv.DictReader((DEST/'sources'/who/file).open()))
assert len(rows)==25 and {r['anonymous_pair_id'] for r in rows}==set(ids)
assert len({r['anonymous_pair_id'] for r in rows})==25
assert all(r['relation'] in {'unrelated','UNRESOLVED','rigid / same part'} and r['local_evidence'].strip() and r['reviewer'].strip() and r['review_version'].strip() for r in rows)
reviews[who]={r['anonymous_pair_id']:r for r in rows}
fields=['relation','local_evidence','resolvable','unresolved_reason','reviewer','review_version']
output=[]
for pid in ids:
a,b=reviews['mold'][pid],reviews['claude'][pid]
same=a['relation']==b['relation']
row=dict(anonymous_pair_id=pid,agreement=same,
adjudication_status='CONCORDANT_REVIEW_ONLY' if same else 'CLAUDE_RULE_ADJUDICATION_PENDING')
for who,source in [('mold',a),('claude',b)]:
row.update({who+'_'+field:source[field] for field in fields})
output.append(row)
with (DEST/'comparison.csv').open('w') as f:
writer=csv.DictWriter(f,fieldnames=list(output[0]),lineterminator='\n');writer.writeheader();writer.writerows(output)
disagreement=[r for r in output if not r['agreement']]
assert len(disagreement)==3
matrix=collections.Counter((r['mold_relation'],r['claude_relation']) for r in output)
summary=dict(status='TWO_FROZEN_REVIEWS_COMPARED_ADJUDICATION_PENDING',
created_at=datetime.datetime.now(datetime.timezone(datetime.timedelta(hours=8))).isoformat(),
anonymous_input_commit='3864d34141a393b957a09937e609ad827ea21c16',
mold_commit=MOLD,claude_commit=CLAUDE,pairs=25,
matrix=[dict(mold=a,claude=b,count=n) for (a,b),n in sorted(matrix.items())],
disagreement_ids=[r['anonymous_pair_id'] for r in disagreement],
same_unresolved_ids=[r['anonymous_pair_id'] for r in output if r['mold_relation']==r['claude_relation']=='UNRESOLVED'],
comparison_preserves_original_rows_verbatim=True,identity_mapping_read=False,
custody_join_performed=False,components_frozen=False,splits_created=False,
geometry_or_registration_rerun=False,scripts_from_reviewers_executed=False,
original_authors_preserved=True,source_sha256={str(p.relative_to(DEST)):sha(p) for p in sorted((DEST/'sources').rglob('*')) if p.is_file()},
comparison_sha256=sha(DEST/'comparison.csv'))
(DEST/'receipt.json').write_text(json.dumps(summary,indent=2)+'\n')
print(json.dumps({k:v for k,v in summary.items() if k!='source_sha256'},ensure_ascii=False))
if __name__=='__main__':main()