Archive MaleCNS research sources, provenance, bibliography and immutable data inventory

models/male-cns-lif/late-kc-audit-v1/phase4w_describe.py

@@ -0,0 +1,46 @@
+"""Descriptive summaries after preregistered diagnostics; no candidate selection."""
+from pathlib import Path
+import json
+import pandas as pd
+import numpy as np
+import yaml
+from phase4r import save
+
+cfg=yaml.safe_load(Path(__file__).with_name('phase4w.yaml').read_text());exp=Path(cfg['experiment_root'])
+assert not (exp/'checksums.sha256').exists(),'Refusing sealed output changes'
+src=pd.read_parquet(exp/'audit/incoming-source-metadata.parquet')
+meta=pd.read_parquet(exp/'audit/watched-metadata.parquet')
+top=[];shadow=[]
+for p in sorted((exp/'drive').glob('*-actual-edge-arrivals.parquet')):
+    trial=p.name.removesuffix('-actual-edge-arrivals.parquet');df=pd.read_parquet(p)
+    df=df[(df.arrival_step>=int(cfg['late_start_ms']/.1)) & (df.effective_weight>0)]
+    grouped=df.groupby(['target_neuron_idx','source_neuron_idx','source_category'],as_index=False).agg(
+        positive_drive=('effective_weight','sum'),events=('edge_id','size'),raw_synapse_count=('raw_synapse_count','first'))
+    grouped['positive_fraction']=grouped.positive_drive/grouped.groupby('target_neuron_idx').positive_drive.transform('sum')
+    grouped=grouped.merge(src,left_on='source_neuron_idx',right_on='neuron_idx',how='left').drop(columns='neuron_idx')
+    grouped['trial']=trial;top.append(grouped)
+for p in sorted((exp/'drive').glob('*-shadow.parquet')):
+    df=pd.read_parquet(p);late=df[df.step>=int(cfg['late_start_ms']/.1)]
+    row={'file':p.name,'neuron_idx':int(df.neuron_idx.iloc[0]),'max_abs_rounding_residual':float(df.shadow_residual.abs().max()),
+        'mean_abs_rounding_residual':float(df.shadow_residual.abs().mean())}
+    row.update({col+'_late_mean':float(late[col].mean()) for col in df.columns if col.startswith('g_')})
+    shadow.append(row)
+tops=pd.concat(top,ignore_index=True);tops.to_parquet(exp/'audit/late-positive-sources.parquet',index=False)
+pd.DataFrame(shadow).to_parquet(exp/'audit/shadow-summary.parquet',index=False)
+replay=pd.read_parquet(exp/'replay/results.parquet');drive=pd.read_parquet(exp/'drive/summary.parquet')
+lines=['# Phase 4W mechanism summary','',
+    'Descriptive audit; no learning, model parameter change, or candidate selection. Six previously seen odors, one fixed seed each; not independent population generalization.','',
+    '| neuron_idx | body_id | actual annotation type |','|---:|---:|---|']
+for _,r in meta.iterrows():lines.append(f'| {r.neuron_idx} | {r.body_id} | {r.annotation_type} |')
+lines += ['', '| neuron_idx | source class | fraction of positive late arrivals |','|---:|---|---:|']
+late=drive[drive.period=='late'].groupby(['neuron_idx','source_category']).positive_arrival_sum.sum()
+for (node,cat),value in late.items():
+    den=late.loc[node].sum();lines.append(f'| {node} | {cat} | {100*value/den if den else 0:.4f}% |')
+lines += ['', '| neuron_idx | isolated replay | late spike count range |','|---:|---|---:|']
+for (node,mode),g in replay.groupby(['neuron_idx','mode']):lines.append(f'| {node} | {mode} | {g.late_spikes.min()}–{g.late_spikes.max()} |')
+lines += ['', 'Replay holds all other recorded drive and the measured applied APL signal fixed. Removing a source class is a conditional single-cell test, not a closed-loop whole-brain intervention.',
+    'Drive fractions refer to delivered positive synaptic input, not fractions of biological causality or of net g. Inhibition and APL are separately accounted.',
+    'Ranked ancestry uses positive edges only; inhibitory influences are not claimed absent. Ranking is a heuristic; all path edges have real source events and arrival<=target-spike time.',
+    f"Float64 category shadow max absolute residual versus actual float32 g: {max(x['max_abs_rounding_residual'] for x in shadow):.9g}. This does not alter the exact CPU replay or actual queue checks."]
+(exp/'audit/MECHANISM.md').write_text('\n'.join(lines)+'\n',encoding='utf-8')
+print('\n'.join(lines))