"""Metadata-only Track A postfit binding while the original training continues.""" from pathlib import Path import time from phase2_h1_worker_v2 import sha, checked, save from track_a_h2_common import config as training_config, read from track_a_h2_prepare import local_code_closure B=Path(__file__).resolve().parents[1] T=Path('/home/frank14f/MeltMind-runs/track-a-h2-2026-11-18') R=T.parent/'track-a-postfit-2026-11-18' def main(): start=time.process_time() if R.exists():raise FileExistsError('Existing postfit entry; inspect rather than duplicate') tc,_=training_config(T);entry=read(tc['reuse_entry']);sources=dict(tc['source_sha256']) for p in [T/'contract.json',T/'implementation-git-freeze.json',T/'implementation-authorization.json',T/'review.json']: sources[str(p)]=sha(p) frozen=read(checked(entry['A_predictions_freeze']['path'],entry['A_predictions_freeze']['sha256'])) originals={(r['split'],r['case']):r for r in frozen['rows']};A={};count=0 for j in tc['jobs']: index=j['id'];A[str(index)]={} for case in j['test']: row=originals[(index,case)];ref=row['predictions']['GBRT'] if row['status']!='PREDICTED' or row['raw_ids_sha256']!=tc['parts'][case]['output_sha256']['raw_ids.npy'] or ref['nodes']!=tc['parts'][case]['nodes'] or not ref['no_clipping']:raise ValueError('Original A prediction/native-ID identity') A[str(index)][case]=dict(ref,raw_ids_sha256=row['raw_ids_sha256']) count+=1 expected_unique=len(set().union(*(set(j['test']) for j in tc['jobs']))) if count!=1537 or expected_unique!=586:raise ValueError('Original paired support differs') codes=local_code_closure([B/'tools'/n for n in ['track_a_postfit.py','track_a_postfit_common.py', 'track_a_postfit_worker.py','track_a_postfit_guard.py','track_a_postfit_statistics.py', 'prepare_track_a_postfit.py','test_track_a_postfit.py']]) for p,h in tc['code_sha256'].items():checked(p,h) c=dict(version='track-a-postfit-v1',approval='1794975970.156219',budget_approval='1794982020.974879',estimator_approval='1794982202.809969', training_root=str(T),training_contract_sha256=sha(T/'contract.json'), comparison_contract=tc['reuse_entry'],comparison_contract_sha256=sha(tc['reuse_entry']), source_sha256=sources,code_sha256=codes,jobs=[dict(id=j['id'],expected_parts=len(j['test'])) for j in tc['jobs']], A_predictions=A,expected_pair_visits=count,expected_unique_target_reads=expected_unique, gpu_uuid='GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76',worker_cpus=[4,5],coordinator_cpus=[12,13], gpu_cap=162000,postfit_gpu_cap=7200,postfit_cpu_cap=3600,idle_wait_seconds=20,no_retry=True, bootstrap=dict(replicates=2000,seed=2026111408,confidence=.95,unit='unique_part_mean_of_split_visit_differences',interval='percentile_linear'), no_auto_launch=True,all_five_training_terminals_before_prediction=True,all_predictions_frozen_before_targets=True, A_refit=False,H2_refit=False,new_decodes=0,metric=tc['metric'],original_failed=tc['original_failed']) R.mkdir();save(R/'contract.json',c) save(R/'preparation-receipt.json',dict(status='PREPARED_PENDING_EXACT_PRE_SCORE_REVIEW', contract_sha256=sha(R/'contract.json'),code_files=len(codes),paired_visits=count,unique_cases=expected_unique, metadata_cpu_seconds=time.process_time()-start,source_training_unchanged=True,model_weights_read=False, scientific_arrays_read=False,target_values_read=False,old_scores_read=False,new_predictions=False, training_complete_exists=(T/'complete.json').exists(),launched=False)) print(read(R/'preparation-receipt.json')) if __name__=='__main__':main()