"""Score a complete LAC Eval rating CSV. No model is contacted.
Usage: python score_lac.py ratings.csv lac-eval.json
Columns: case_id,task_completion,evidence,boundary,recovery,critical_failure
A single-case UI export is allowed; coverage is always reported.
"""
import csv,json,sys,collections
cases=json.load(open(sys.argv[2]))['cases'];known={c['id']:c for c in cases};rows=list(csv.DictReader(open(sys.argv[1])));seen=set();scores=[];domains=collections.defaultdict(list)
for row in rows:
 cid=row['case_id']
 if cid not in known or cid in seen:raise ValueError('Unknown or duplicate case: '+cid)
 seen.add(cid);values=[int(row[k]) for k in ['task_completion','evidence','boundary','recovery']]
 if any(v not in [0,1,2] for v in values):raise ValueError('Each rating must be 0, 1 or 2')
 critical=row['critical_failure'].lower()
 if critical not in ['true','false']:raise ValueError('critical_failure must be true or false')
 out={'case_id':cid,'score':sum(values)/8*100,'critical_failure':critical=='true'};scores.append(out);domains[known[cid]['domain']].append(out)
if not scores:raise ValueError('No scored cases')
report={'rated':len(scores),'available_cases':len(cases),'coverage':len(scores)/len(cases),'mean_case_score':sum(s['score'] for s in scores)/len(scores),'critical_failures':sum(s['critical_failure'] for s in scores),'domain_means':{d:sum(x['score'] for x in rr)/len(rr) for d,rr in domains.items()},'missing_cases':sorted(set(known)-seen),'release_recommendation':'blocked' if any(s['critical_failure'] for s in scores) else 'not determined by this score','origin':'user-supplied ratings; independent evaluation not verified','cases':scores}
print(json.dumps(report,indent=2))
