"""Labelled S1 amendment; original pilot and scientific sources remain immutable.""" import json from pathlib import Path import numpy as np import torch from phase2_h1_worker_v2 import checked from phase2_h2_data import config as pilot_config BASE=Path('/home/frank14f/optane/MeltMind') def read(p):return json.loads(Path(p).read_text()) def config(root): c=read(root/'contract.json') if c['approval']!='1794901608.131369' or c['mode']!='H2_S1_FOLD0_LAST200':raise PermissionError('Unapproved S1 scope') for p,h in {**c['code_sha256'],**c['source_sha256']}.items():checked(BASE/p,h) old=pilot_config(Path(c['pilot_root'])) if c['epochs']!=200 or c['seed']!=0 or c['fold']!=0 or c['total_h2_gpu_seconds']!=144000:raise ValueError('Frozen scientific recipe changed') if c['prior_pilot_gpu_seconds']+c['gpu_seconds']>c['total_h2_gpu_seconds']:raise ValueError('Cumulative budget exceeds cap') return c,old def gate(field_high,last_high): return bool((field_high is not None and np.isfinite(field_high) and field_high<.005) or (last_high is not None and np.isfinite(last_high) and last_high<0)) def selected_checkpoint(completed_epochs): if completed_epochs!=200:raise ValueError('Only final epoch200 may be evaluated') return 'last-epoch200.pt' def stats_from_pilot(c,old): stats=read(checked(c['standardization_path'],c['standardization_sha256'])) if stats['parts']!=old['folds'][0]['learning'] or stats['nodes']!=old['folds'][0]['learning_nodes']:raise ValueError('Normalization learning domain mismatch') if len(stats['mean'])!=19 or len(stats['scale'])!=19 or not np.isfinite(stats['scale']).all() or min(stats['scale'])<=0:raise ValueError('Invalid fixed normalization') return stats