Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261014T113716Z-figure1-overview-training-corpus/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/tools/figure1_overview_20261014.py

@@ -0,0 +1,151 @@
+"""Original, editable biological overview glyphs inspired by the supplied panel d."""
+from pathlib import Path
+import hashlib
+import json
+import os
+import sys
+import numpy as np
+if os.environ.get('CUDA_VISIBLE_DEVICES') != '':
+    raise SystemExit('CPU-only: CUDA_VISIBLE_DEVICES must be empty')
+D=Path(__file__).resolve().parents[1]
+M=D.parents[1]
+A=D/'data/figure1_revision_20261014/overview'
+O=D/'figures/figure1_revision_20261014/overview'
+sys.path.insert(0,str(D/'tools'))
+import style
+import matplotlib.pyplot as plt
+from matplotlib.patches import PathPatch, Circle, FancyArrowPatch, Rectangle
+from matplotlib.path import Path as MPath
+INK='#222222'; BLUE='#168FE5'; PALE='#D7ECF9'; GOLD='#D9B23F'
+CELL=['#4C7195','#9ABACF','#C98A4B','#6B977B','#A3BE97','#9576A2']
+
+
+def sha(p):
+    with Path(p).open('rb') as f:return hashlib.file_digest(f,'sha256').hexdigest()
+
+
+def organic(ax,x,y,rx,ry,colour,edge='none',lw=.8,phase=0,alpha=1):
+    t=np.linspace(0,2*np.pi,145)
+    r=1+.045*np.sin(3*t+phase)+.025*np.cos(5*t-phase)
+    xy=np.column_stack([x+rx*r*np.cos(t),y+ry*r*np.sin(t)])
+    ax.add_patch(PathPatch(MPath(xy,closed=True),facecolor=colour,edgecolor=edge,lw=lw,alpha=alpha,zorder=2))
+
+
+def cell(ax,x,y,r,c,phase=0,nucleus=True):
+    organic(ax,x,y,r,r*.91,c,phase=phase)
+    if nucleus:organic(ax,x-r*.12,y+r*.02,r*.40,r*.38,'#FFFFFF',phase=phase+1,alpha=.53)
+
+
+def cloud(ax,x,y,scale=1,noise=False):
+    coords=[(-.45,.20),(-.19,.36),(.10,.30),(.41,.20),(-.53,-.11),(-.24,-.02),(.05,.02),(.35,-.08),(-.31,-.32),(.02,-.30),(.37,-.34)]
+    for j,(dx,dy) in enumerate(coords):
+        cell(ax,x+dx*scale,y+dy*scale,.115*scale, '#C8CDD1' if noise else CELL[j%len(CELL)],phase=j*.7,nucleus=not noise)
+
+
+def model(ax,x,y,adapted=False,scale=1):
+    organic(ax,x,y,.56*scale,.51*scale,PALE,edge='#93BAD4',lw=1.0,phase=.8)
+    pts=np.array([[-.30,.19],[0,.29],[.29,.14],[-.30,-.17],[0,-.03],[.27,-.20],[.02,-.31]])*scale
+    for a,b in [(0,1),(0,3),(0,4),(1,2),(1,4),(2,4),(2,5),(3,4),(3,6),(4,5),(4,6),(5,6)]:
+        ax.plot([x+pts[a,0],x+pts[b,0]],[y+pts[a,1],y+pts[b,1]],color='#86B6D6',lw=.9,zorder=3)
+    for i,(dx,dy) in enumerate(pts):cell(ax,x+dx,y+dy,.078*scale,BLUE,phase=i,nucleus=False)
+    if adapted:
+        for i,(dx,dy) in enumerate([(.47,.34),(.57,.05),(.44,-.37)]):
+            ax.plot([x+.20*scale,x+dx*scale],[y+dy*.5*scale,y+dy*scale],color='#9D8946',lw=.9,zorder=3)
+            cell(ax,x+dx*scale,y+dy*scale,.095*scale,GOLD,phase=i,nucleus=False)
+
+
+def arrow(ax,start,end,via=None,label=None,label_xy=None):
+    if via:
+        verts=[start,*via,end]
+        path=MPath(verts,[MPath.MOVETO]+[MPath.LINETO]*(len(verts)-1))
+        patch=FancyArrowPatch(path=path,arrowstyle='-|>',mutation_scale=12,lw=1.05,color=INK,zorder=1)
+    else:patch=FancyArrowPatch(start,end,arrowstyle='-|>',mutation_scale=12,lw=1.05,color=INK,zorder=1)
+    ax.add_patch(patch)
+    if label:
+        lx,ly=label_xy or ((start[0]+end[0])/2,(start[1]+end[1])/2+.18)
+        ax.text(lx,ly,label,ha='center',va='center',fontsize=10.5,color=INK,
+                bbox=dict(facecolor='white',edgecolor='none',pad=1.5))
+
+
+def bars(ax,x,y,w=.7,h=.42):
+    vals=[.16,.55,.9,.36,.71,.18,.43,.78,.29]
+    for j,v in enumerate(vals):ax.add_patch(Rectangle((x-w/2+j*w/len(vals),y),w/len(vals)*.67,h*v,facecolor='#B89D4D',edgecolor='none',zorder=3))
+    ax.plot([x-w/2,x+w/2],[y,y],color='#888888',lw=.6)
+
+
+def main():
+    A.mkdir(parents=True,exist_ok=True);O.mkdir(parents=True,exist_ok=True)
+    fig,ax=plt.subplots(figsize=(12.3,5.7));ax.axis('off');ax.set_xlim(-.1,12.3);ax.set_ylim(-.6,5.1);ax.set_aspect('equal')
+    for x,num,title in [(.12,'1','Pretrain on single cells'),(4.45,'2','Adapt on measured pairs'),(8.5,'3','Generate from new bulk')]:
+        ax.text(x,4.94,num,fontsize=16,fontweight='bold',va='top')
+        ax.text(x+.30,4.94,title,fontsize=14,va='top',fontweight='bold')
+    # Training: same-sample member counts define the conditioning pseudobulk.
+    cloud(ax,.90,3.72,.92)
+    ax.text(.90,2.94,'33.14M\ntraining cells',ha='center',va='top',fontsize=12.5,linespacing=1.35)
+    arrow(ax,(1.48,3.73),(2.12,3.73))
+    organic(ax,2.87,3.65,.58,.83,'#F3F3F0',edge='#B4B6B2',phase=2)
+    bars(ax,2.87,3.87,.71,.40)
+    ax.text(2.87,4.46,'Same sample',ha='center',fontsize=11)
+    ax.text(2.87,3.68,'Pseudobulk',ha='center',fontsize=10.5)
+    cell(ax,2.87,3.31,.19,CELL[3],phase=1.4)
+    ax.text(2.87,2.99,'Member cell',ha='center',fontsize=10.5)
+    arrow(ax,(2.66,2.78),(2.22,2.16),label='condition + target',label_xy=(3.10,2.35))
+    model(ax,2.00,1.60)
+    ax.text(2.00,.86,'Pretrained PBISC',ha='center',fontsize=13,fontweight='bold')
+    ax.text(2.00,.49,'Learn gene value and detection',ha='center',fontsize=10.5,color='#555555')
+    # Adaptation: pairing is represented by a light common envelope, not a Venn overlap.
+    organic(ax,6.22,3.67,1.20,.64,'#F2F5F3',phase=1)
+    bars(ax,5.63,3.55,.73,.52)
+    cloud(ax,6.83,3.75,.66)
+    ax.text(6.22,4.46,'Paired samples · same donor',ha='center',fontsize=11)
+    ax.text(5.63,3.04,'Measured bulk',ha='center',fontsize=10.5)
+    ax.text(6.83,3.04,'scRNA-seq',ha='center',fontsize=10.5)
+    arrow(ax,(6.22,2.81),(6.22,2.19),label='LoRA adaptation',label_xy=(7.28,2.53))
+    arrow(ax,(2.66,1.60),(5.54,1.60),label='pretrained weights',label_xy=(4.09,1.83))
+    model(ax,6.22,1.60,adapted=True)
+    ax.text(6.22,.86,'Adapted PBISC',ha='center',fontsize=13,fontweight='bold')
+    ax.scatter([5.40],[.48],s=30,c=GOLD,zorder=5)
+    ax.text(5.56,.48,'Trainable adapters',ha='left',va='center',fontsize=10.5,color='#555555')
+    # Inference: new measured bulk is shared; N_gen noise draws produce cells.
+    bars(ax,9.25,3.60,.80,.55)
+    ax.text(9.25,4.46,'New measured bulk',ha='center',fontsize=10.5)
+    cloud(ax,11.35,3.79,.65,noise=True)
+    ax.text(11.35,4.46,'Gaussian noise',ha='center',fontsize=10.5)
+    arrow(ax,(9.43,3.35),(9.94,2.95))
+    arrow(ax,(11.18,3.30),(10.76,2.95))
+    model(ax,10.35,2.49,adapted=True,scale=.9)
+    arrow(ax,(6.96,1.60),(9.69,2.49),via=[(8.54,1.60),(8.54,2.49)],label='reuse model',label_xy=(7.82,1.83))
+    arrow(ax,(10.35,1.94),(10.35,1.34),label='denoise',label_xy=(11.20,1.67))
+    cloud(ax,10.35,.85,1.01)
+    ax.text(10.35,.18,'Generated single cells',ha='center',fontsize=12.5,fontweight='bold')
+    ax.plot([.12,12.08],[-.13,-.13],color='#D7D7D7',lw=.7)
+    ax.text(6.10,-.43,'Held-out paired evaluation: cell composition · disease signals · rare populations',
+            ha='center',va='center',fontsize=11.6,color='#444444')
+    fig.subplots_adjust(left=.01,right=.99,top=.99,bottom=.01)
+    for ext in ['png','pdf','svg']:fig.savefig(O/f'experimental_overview.{ext}',dpi=260)
+    plt.close(fig)
+    detail=('Original schematic redrawn with editable organic vector circles, muted fills, thin dark arrows and open white space, '
+            'inspired by the supplied reference panel d. No scientific objects or numeric results were copied from that paper. '
+            '33,141,380 is the verified unique training-cell universe. A same-sample K-cell recipe supplies pseudobulk conditioning '
+            'to each member target cell. Pretrained weights initialize LoRA adaptation on measured bulk–scRNA pairs. '
+            'New measured bulk and independent Gaussian-noise draws generate cells, without supplied cell-type quotas. '
+            'Cell colours and model-node glyphs are schematic, not counts, measured populations or a literal architecture. '
+            'The main model uses no cell-type/disease input labels; Figure2 oracle/prediction-hint ablations are separate. '
+            'Generated annotations are post hoc. Conditioning recipe K and generated-cell number N_gen are distinct. '
+            'Held-out scRNA is evaluation truth, never shown as an inference input. Rare-population endpoints are evaluation '
+            'objectives, not a claim of completed source-cluster recovery; Figure6 bulk-only published-finding reproduction remains TODO.')
+    src=[Path(__file__),M/'runs/f01_6k_1x_seed20260627/ALLCELL_DATA_CONTRACT.json',
+         D/'data/figure_assignment_20261013/AUTHOR_SCOPE.md',D/'PBISC-Diffusion_color_palette_20261013.md',
+         Path('/home/sj_server_1/PBISC/paper-release-20260928/manuscript/scripts/build_figure1_architecture_nature.py')]
+    result={'schema':'pbisc.figure1-overview.20261014.v1','slides':[{'kind':'content','title':'Experimental overview',
+        'section':'Figure 1','deck_group':'Figure 1','section_label':'Fig. 1a','layout':'full','keywords':[],
+        'figures':[str((O/'experimental_overview.png').relative_to(D))],
+        'footnote':'Schematic. K defines the pseudobulk recipe; N_gen defines generated-cell count. Held-out scRNA is used only for evaluation.',
+        'footnote_full':detail,'sources':'; '.join(str(p) for p in src)}],
+        'reference_style':{'path':'/home/sj_server_1/.codex/attachments/15a605fb-2cca-4162-9f87-354b7c57cac6/codex-clipboard-98e511c6-7c97-416f-9054-2e6db758e44e.png','interpretation':'panel d: organic flat circles, minimal black arrows, white background; existing method palette retained'},
+        'assets':[str(p) for p in O.glob('*')], 'sources':[{'path':str(p),'sha256':sha(p)} for p in src],
+        'new_training':False,'new_inference':False}
+    (A/'result.json').write_text(json.dumps(result,indent=2)+'\n')
+
+
+if __name__=='__main__':main()

paper-releases/pbisc-diffusion/updates/20261014T113716Z-figure1-overview-training-corpus/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/tools/figure1_train_corpus_20261014.py

@@ -0,0 +1,290 @@
+#!/usr/bin/env python3
+"""Exact training-source composition; metadata-only, bounded streaming.
+
+The source support chart uses all 41,395,184 corpus cells. This derivative
+counts the actual train split, including the unlabelled overlay, once per
+cell. It never estimates training composition by rescaling source shares.
+"""
+from __future__ import annotations
+
+import ast
+from collections import Counter
+import csv
+import gzip
+import hashlib
+import json
+from pathlib import Path
+import re
+
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+from matplotlib.patches import Rectangle
+import numpy as np
+import pandas as pd
+import pyarrow.parquet as pq
+
+D = Path(__file__).resolve().parents[1]
+M = D.parents[1]
+MS = Path('/home/sj_server_1/PBISC/paper-release-20260928/manuscript')
+C = MS/'source_data/corpus'
+M7 = M.parent/'model_7_no_annotation_sh_hwang/data'
+E2 = M7/'e2_full_source/metadata/obs_sc.parquet'
+BLOCK = Path('/home/sj_server_1/PBISC/PBISC-data/single-cell-data-block-blood/metadata/obs_sc.parquet')
+DATA = D/'data/figure1_revision_20261014/train_corpus'
+FIG = D/'figures/figure1_revision_20261014/train_corpus'
+KEY = ['dataset_id','donor_id','sample_id']
+INK, SUB = '#1A1A1A', '#626262'
+SOURCES, ASSETS = {}, []
+
+
+def sha(path):
+    h=hashlib.sha256()
+    with Path(path).open('rb') as f:
+        for chunk in iter(lambda:f.read(1<<20),b''): h.update(chunk)
+    return h.hexdigest()
+
+
+def source(path, role):
+    path=Path(path)
+    SOURCES[str(path)]={'path':str(path),'host':'server_1','sha256':sha(path),
+                        'bytes':path.stat().st_size,'role':role}
+    return path
+
+
+def doc(path,role): return json.loads(source(path,role).read_text())
+
+
+def table(path,role): return pd.read_csv(source(path,role))
+
+
+def output_csv(name,rows):
+    p=DATA/(name+'.csv')
+    with p.open('w',newline='') as f:
+        w=csv.DictWriter(f,fieldnames=list(rows[0]));w.writeheader();w.writerows(rows)
+    ASSETS.append(str(p))
+
+
+def output_json(name,data):
+    p=DATA/(name+'.json');p.write_text(json.dumps(data,indent=2)+'\n');ASSETS.append(str(p));return p
+
+
+def geo(acc):
+    rows={}
+    p=source(C/'geo'/f'{acc}_series_matrix.txt.gz','original compact GEO donor metadata')
+    with gzip.open(p,'rt') as f:
+        for line in f:
+            if not line.startswith('!Sample_'):continue
+            k,*v=line.rstrip('\n').split('\t');rows.setdefault(k,[]).append([x.strip('"') for x in v])
+    t=pd.DataFrame({'gsm':rows['!Sample_geo_accession'][0],'title':rows['!Sample_title'][0]})
+    for v in rows.get('!Sample_characteristics_ch1',[]):
+        for i,s in enumerate(v):
+            if ': ' in s:
+                k,x=s.split(': ',1);t.loc[i,k.strip()]=x.strip()
+    for i,v in enumerate(rows.get('!Sample_description',[])):t[f'description_{i}']=v
+    return t
+
+
+def aggregate():
+    summ=doc(C/'corpus_summary.json','all-source totals used for independent replay checks')
+    g=table(C/'corpus_cellgroups.csv','cell-level condition and split counts, including overlay')
+    samples=table(C/'corpus_samples.csv','compact registered sample and split table')
+    ds=table(C/'corpus_datasets.csv','independent train totals by dataset')
+    srcct=table(C/'corpus_celltypes.csv','independent all-source L1/L2 counts')
+    contract=doc(M/'runs/f01_6k_1x_seed20260627/ALLCELL_DATA_CONTRACT.json','completed training all-cell coverage contract')
+    idx=doc(M7/'e2_full_index/dataset_manifest.json','training corpus fingerprint and registered split')
+    si=pd.read_parquet(source(M7/'e2_full_index/metadata/sample_index.parquet','exact sample-to-train/validation/test assignment'))
+    assert not si.duplicated(KEY).any()
+    for col in KEY:si[col]=si[col].astype(str)
+    assert set(si.sample_key)==set(samples.sample_key)
+    sc=samples.merge(si[['sample_key','n_cells','split']],on='sample_key',suffixes=('_table','_index'),validate='one_to_one')
+    assert (sc.n_cells_table==sc.n_cells_index).all() and (sc.split_table==sc.split_index).all()
+    source(C/'README.md','source-condition interpretation and biological-donor correction rules')
+    source(MS/'scripts/corpus/build_corpus_tables.py','source-condition and donor harmonization implementation')
+    colorcode=source(MS/'scripts/corpus/build_figure1_corpus.py','preserved biological category colors and L1 merge rules')
+    maps={}
+    for n in ast.parse(colorcode.read_text()).body:
+        if isinstance(n,ast.Assign):
+            for t in n.targets:
+                if isinstance(t,ast.Name) and t.id in ['COND_COLOR','L1_COLOR','L1_ORDER','L1_MERGE']:
+                    maps[t.id]=ast.literal_eval(n.value)
+    # Canonical donor corrections match the corpus table builder. GSE274680
+    # source keys are per visit; actual subjects are retrieved through GSM.
+    g274=geo('GSE274680');subjects=g274['subject id'].fillna(g274['subject'])
+    assert subjects.notna().all()
+    gsm_subject=dict(zip(g274.gsm,subjects))
+    g132=geo('GSE132802');dress_dx=dict(zip(g132.description_0.str.strip(),g132['disease diagnosis']))
+    donor_correction={}
+    counts=Counter();sample_counts=Counter();shards=[];overlay=Counter()
+    files=sorted(E2.glob('*.parquet'));assert len(files)==432
+    for fi,ep in enumerate(files):
+        source(ep,'corpus sample membership and original row identity metadata')
+        epq=pq.ParquetFile(ep)
+        ov=ep.name.endswith('__strict_overlay.parquet')
+        is274=ep.name.startswith('061_A')
+        bp=None if ov else BLOCK/ep.name
+        if bp:
+            source(bp,'matched original source L1 annotation metadata')
+            bpq=pq.ParquetFile(bp)
+            assert epq.metadata.num_rows==bpq.metadata.num_rows
+            bcols=['source_row_index','celltype_l1']+(['source_file'] if is274 else [])
+            bb=iter(bpq.iter_batches(batch_size=65536,columns=bcols,use_threads=False))
+        local=Counter();nrows=0
+        for eb in epq.iter_batches(batch_size=65536,columns=KEY+['source_row_index'],use_threads=False):
+            e=eb.to_pandas()
+            for col in KEY:e[col]=e[col].astype(str)
+            if ov:e['celltype_l1']='No label (overlay)'
+            else:
+                b=next(bb).to_pandas()
+                assert len(b)==len(e) and np.array_equal(b.source_row_index.to_numpy(),e.source_row_index.to_numpy())
+                e['celltype_l1']=b.celltype_l1.astype(str).to_numpy()
+                if is274:
+                    e['source_file']=b.source_file.astype(str).to_numpy()
+                    for r in e[KEY+['source_file']].drop_duplicates().itertuples(index=False):
+                        k=tuple(r[:3]);gsm=re.search(r'(GSM\d+)_',r.source_file).group(1)
+                        corrected=gsm_subject[gsm]
+                        assert k not in donor_correction or donor_correction[k]==corrected
+                        donor_correction[k]=corrected
+            # Aggregate per sample before the compact join to avoid per-cell
+            # string-heavy joined tables. No expression values are accessed.
+            a=e.groupby(KEY+['celltype_l1'],observed=True).size().rename('cells').reset_index()
+            a=a.merge(si[KEY+['split','sample_key']],on=KEY,how='left',validate='many_to_one')
+            assert a.split.notna().all()
+            for r in a.itertuples(index=False):
+                counts[(r.split,r.celltype_l1)]+=int(r.cells)
+                local[r.split]+=int(r.cells);sample_counts[r.sample_key]+=int(r.cells)
+                if ov:overlay[r.split]+=int(r.cells)
+            nrows+=len(e)
+        assert nrows==epq.metadata.num_rows
+        if bp:assert next(bb,None) is None
+        shards.append({'shard':ep.name,'n_cells':nrows,'train':local['train'],'val':local['val'],'test':local['test'],'overlay':ov})
+        if fi%40==0:print(f'metadata {fi+1}/{len(files)}',flush=True)
+    assert sum(counts.values())==summ['cells']
+    all_l1=Counter()
+    for (s,c),n in counts.items():all_l1[c]+=n
+    assert dict(all_l1)==summ['celltype_l1_cells']
+    assert dict(srcct.groupby('celltype_l1').n_cells.sum())==dict(all_l1)
+    assert all(sample_counts[r.sample_key]==r.n_cells for r in si.itertuples(index=False))
+    split_totals={s:sum(n for (ss,c),n in counts.items() if ss==s) for s in ['train','val','test']}
+    assert split_totals==summ['split_cells']
+    assert sum(overlay.values())==summ['overlay_cells']
+    assert split_totals['train']==contract['coverage_after_topup']['train_universe_cells']==33141380
+    assert contract['coverage_after_topup']['train_cells_covered_after_topup']==33141380
+    assert idx['corpus_fingerprint']==contract['pairing']['source_corpus_fingerprint']
+    train=g[g.split=='train'];N=int(train.n_cells.sum())
+    assert N==33141380==int(ds.train_cells.sum())==int(si.loc[si.split=='train','n_cells'].sum())
+    cond={k:int(v) for k,v in train.groupby('condition').n_cells.sum().items()}
+    cells=Counter()
+    for (s,c),n in counts.items():
+        if s=='train':cells[maps['L1_MERGE'].get(c,c)]+=n
+    assert sum(cells.values())==N and set(cells)==set(maps['L1_ORDER'])
+    donors=set();all_donors=set();ds_train=set()
+    for r in si.itertuples(index=False):
+        k=(r.dataset_id,r.donor_id,r.sample_id)
+        if r.dataset_id.startswith('061_A'):dn=donor_correction[k]
+        elif r.dataset_id.startswith('300b_A'):dn='healthy volunteer' if dress_dx[r.donor_id]=='Healthy' else 'DRESS patient'
+        else:dn=r.donor_id
+        all_donors.add((r.dataset_id,dn))
+        if r.split=='train':donors.add((r.dataset_id,dn));ds_train.add(r.dataset_id)
+    assert len(all_donors)==summ['donors']==3199
+    nh=N-cond['Healthy'];risk=cond['ACPA+ at risk of RA']
+    cmv=int(train.loc[train.cmv_seropositive_healthy,'n_cells'].sum())
+    iv=int(train.loc[(train.condition=='Healthy')&(train.handling=='in vitro'),'n_cells'].sum())
+    z={'n_cells':N,'datasets':len(ds_train),'samples':int((si.split=='train').sum()),'donors':len(donors),
+       'healthy_cells':cond['Healthy'],'disease_or_at_risk_cells':nh,'at_risk_cells':risk,
+       'disease_history_without_at_risk_cells':nh-risk,'condition_cells':cond,'celltype_cells':dict(cells),
+       'healthy_cmv_seropositive_cells':cmv,'healthy_in_vitro_cells':iv,'overlay_cells':dict(overlay),
+       'split_cells':split_totals,'corpus_fingerprint':idx['corpus_fingerprint'],'maps':maps,
+       'source_total_cells':summ['cells'],'all_source_donors':len(all_donors),
+       'checks':{'source_row_alignment_all_base_shards':True,'all_source_l1_replay_matches_saved_table':True,
+                 'all_sample_counts_replay_matches_index':True,'all_split_counts_match_saved_tables':True,
+                 'donor_correction_replay_matches_saved_total':True,'all_train_covered_by_final_run_contract':True}}
+    output_csv('celltype_by_split',[{'split':s,'celltype_l1':c,'n_cells':n,'share_of_split':n/split_totals[s]}
+                                   for (s,c),n in sorted(counts.items())])
+    output_csv('training_celltype_counts',[{'celltype':c,'n_cells':cells[c],'training_fraction':cells[c]/N,'color':maps['L1_COLOR'][c]}
+                                         for c in maps['L1_ORDER']])
+    output_csv('training_condition_counts',[{'condition':c,'n_cells':cond[c],'training_fraction':cond[c]/N,
+                  'disease_detail_fraction':cond[c]/nh if c!='Healthy' else '',
+                  'color':maps['COND_COLOR'][c]} for c in summ['conditions_order']])
+    output_csv('shard_checks',shards)
+    output_csv('training_dataset_counts',[{'dataset_prefix':r.dataset_prefix,'dataset':r.short_name,'train_cells':int(r.train_cells)} for r in ds.itertuples(index=False)])
+    return z
+
+
+def donut(ax,counts,colors,top,bottom):
+    ax.pie(counts,colors=colors,startangle=90,counterclock=False,
+           wedgeprops={'width':.29,'edgecolor':'white','linewidth':1.4})
+    ax.text(0,.06,top,ha='center',va='center',fontsize=22,fontweight='bold')
+    ax.text(0,-.18,bottom,ha='center',va='center',fontsize=12,color=SUB)
+    ax.set_aspect('equal')
+
+
+def plot(z):
+    N=z['n_cells'];cond=z['condition_cells'];cts=z['celltype_cells'];maps=z['maps'];nh=z['disease_or_at_risk_cells']
+    fig=plt.figure(figsize=(12.3,5.6))
+    fig.text(.028,.965,f"33,141,380 training cells  ·  {z['datasets']} datasets  ·  {z['samples']:,} samples  ·  {z['donors']:,} donor identities",fontsize=16,fontweight='bold')
+    fig.text(.028,.887,'Healthy versus disease / at-risk',fontsize=14,fontweight='bold')
+    fig.text(.355,.887,'Disease / at-risk detail',fontsize=14,fontweight='bold')
+    fig.text(.687,.887,'Training cell-type composition',fontsize=14,fontweight='bold')
+    a=fig.add_axes([.035,.461,.275,.415]);b=fig.add_axes([.367,.461,.275,.415]);c=fig.add_axes([.700,.461,.275,.415])
+    donut(a,[cond['Healthy'],nh],[maps['COND_COLOR']['Healthy'],'#94776F'],'33.14M','training cells')
+    detail=sorted((k for k in cond if k!='Healthy'),key=cond.get,reverse=True)
+    donut(b,[cond[k] for k in detail],[maps['COND_COLOR'][k] for k in detail],f'{nh/1e6:.3f}M','disease / at-risk')
+    ct=maps['L1_ORDER'];donut(c,[cts[k] for k in ct],[maps['L1_COLOR'][k] for k in ct],'33.14M','training cells')
+    fig.text(.028,.431,f"Healthy: {cond['Healthy']:,}  ({cond['Healthy']/N*100:.1f}%)",fontsize=12.4,fontweight='bold',color='#52647D')
+    fig.text(.028,.384,f"Disease / at-risk: {nh:,}  ({nh/N*100:.1f}%)",fontsize=11.5,color='#765B54')
+    fig.text(.028,.305,'Healthy donor status includes:',fontsize=12,fontweight='bold')
+    fig.text(.028,.259,f"• {z['healthy_in_vitro_cells']/1e6:.3f}M in-vitro cytokine/PBS or IL-36 cells\n• {z['healthy_cmv_seropositive_cells']/1e6:.3f}M CMV-seropositive healthy cells\n• Post-vaccination draws",fontsize=10.8,va='top',linespacing=1.6,color=SUB)
+    fig.text(.028,.130,'ACPA+ at risk is not confirmed RA.',fontsize=11,fontweight='bold')
+    fig.text(.028,.082,f"Disease/history excluding at-risk: {z['disease_history_without_at_risk_cells']/N*100:.2f}%\nAt-risk alone: {z['at_risk_cells']/N*100:.2f}% of training cells",fontsize=10.5,va='top',linespacing=1.4,color=SUB)
+    for i,k in enumerate(detail):
+        y=.433-i*.035
+        fig.patches.append(Rectangle((.350,y-.010),.010,.019,transform=fig.transFigure,color=maps['COND_COLOR'][k],lw=0))
+        fig.text(.368,y,k,fontsize=10.5,va='center')
+        fig.text(.653,y,f'{cond[k]:,} · {cond[k]/nh*100:.1f}%',fontsize=10.3,color=SUB,va='center',ha='right')
+    for i,k in enumerate(ct):
+        y=.433-i*.035
+        fig.patches.append(Rectangle((.685,y-.010),.010,.019,transform=fig.transFigure,color=maps['L1_COLOR'][k],lw=0))
+        fig.text(.703,y,k,fontsize=10.5,va='center')
+        pct=cts[k]/N*100;pp=f'{pct:.1f}%' if pct>=.1 else '<0.1%'
+        fig.text(.987,y,f'{cts[k]:,} · {pp}',fontsize=10.3,color=SUB,va='center',ha='right')
+    fig.text(.355,.026,'Each training source cell is counted once; not recipe-weighted exposure.',fontsize=10.8,color=SUB)
+    for ext in ['png','pdf','svg']:
+        p=FIG/('training_corpus_composition.'+ext);fig.savefig(p,dpi=220,facecolor='white');ASSETS.append(str(p))
+    plt.close(fig)
+    return str((FIG/'training_corpus_composition.png').relative_to(D))
+
+
+def main():
+    DATA.mkdir(parents=True,exist_ok=True);FIG.mkdir(parents=True,exist_ok=True)
+    plt.rcParams.update({'font.family':'sans-serif','font.sans-serif':['Arial','Liberation Sans'],
+        'font.size':14,'pdf.fonttype':42,'svg.fonttype':'none','figure.facecolor':'white','text.color':INK})
+    z=aggregate();figure=plot(z)
+    source(Path(__file__).resolve(),'reproducible exact training split metadata aggregation and figure builder')
+    ver=output_json('verified_data',z)
+    roster=output_json('source_hashes',{'schema':'pbisc.training-corpus-source-hashes.v1','sources':list(SOURCES.values())})
+    source(roster,'complete 432 E2 plus 404 matched annotation source metadata hash roster and compact contracts')
+    source(ver,'exact training-source statistics and independent numeric checks')
+    bindings=[SOURCES[str(ver)],SOURCES[str(roster)],SOURCES[str(Path(__file__).resolve())]]
+    note=(f"Exact training-source composition from the same 33,141,380-cell universe covered by the final f01_6k_1x_seed20260627 run. "
+          f"22 datasets, {z['samples']:,} samples, {z['donors']:,} biological donor identities after the original corpus GEO corrections. "
+          f"Healthy {z['healthy_cells']:,} ({z['healthy_cells']/z['n_cells']*100:.4f}%); disease/at-risk {z['disease_or_at_risk_cells']:,}. "
+          "Condition counts use saved cell-level groups filtered to train, never sample-majority labels. Cell types are recomputed from matching corpus and source annotation metadata, with row identity checked for every base shard and exact sample membership joined to the registered train split. "
+          f"All {z['overlay_cells']['train']:,} train overlay cells remain included as Unlabelled. Source-cell replay across all splits matches the independently saved 41,395,184-cell L1 summary exactly. "
+          "ACPA+ is risk, not confirmed RA; Post-COVID is historical/post-infection; Cytopenia is not equated with confirmed malignancy. Healthy includes stimulation, post-vaccination and CMV-seropositive donors. Cell/condition labels are descriptive, not final base-model inputs. Unique source-cell proportions do not describe recipe-weighted target exposure. "
+          "This panel does not assert biological donor-disjoint internal validation; the original 27-donor cross-split caveat remains.")
+    panel={'key':'main_figure1_training_corpus','kind':'content','title':'Pretraining data: 33.14 million training cells',
+           'subtitle':'Actual training-split composition, with each source cell counted once',
+           'figure':figure,'figures':[figure],'section':'Figure 1','section_label':'Figure 1 · main panel',
+           'keywords':[],'footnote':'Exact training split; disease / at-risk includes ACPA+ risk. Counts describe unique source cells, not recipe-weighted exposure.',
+           'note':note,'footnote_full':note,'palette':'biological','source_refs':[b['path'] for b in bindings],'source_bindings':bindings}
+    output_json('hashes',{'sources':list(SOURCES.values()),'outputs':[{'path':p,'sha256':sha(p)} for p in ASSETS]})
+    result={'schema':'pbisc.figure1.training-corpus.v1','panels':[panel],'slides':[panel],
+            'assets':ASSETS+[str(DATA/'result.json')],'sources':list(SOURCES.values()),'verified_data':z,
+            'new_training':False,'new_inference':False,'raw_expression_payload_read':False}
+    (DATA/'result.json').write_text(json.dumps(result,indent=2)+'\n')
+    print(json.dumps({'cells':z['n_cells'],'healthy':z['healthy_cells'],'donors':z['donors'],
+                      'assets':len(result['assets']),'sources':len(SOURCES),'result':str(DATA/'result.json')}),flush=True)
+
+
+if __name__=='__main__':main()

paper-releases/pbisc-diffusion/updates/20261014T113716Z-figure1-overview-training-corpus/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/tools/verify_figure1_master_20261014.py

@@ -0,0 +1,67 @@
+"""Verify source hashes, page preservation, embedded pictures and rendered bounds."""
+from pathlib import Path
+from datetime import datetime,timezone
+import hashlib
+import json
+import sys
+import fitz
+from pptx import Presentation
+D=Path(__file__).resolve().parents[1]
+A=D/'data/figure1_revision_20261014';Q=A/'qa'
+sys.path.insert(0,str(D/'tools'))
+import verify_expanded_master_20261013 as H
+H.QA=Q
+
+
+def sha(path):
+    with Path(path).open('rb') as f:return hashlib.file_digest(f,'sha256').hexdigest()
+
+
+def main():
+    Q.mkdir(exist_ok=True)
+    spec=json.loads((A/'spec.json').read_text());m=json.loads((A/'manifest.json').read_text())
+    records=spec['slides'];pp=D/spec['deck']['output_pptx'];pf=D/spec['deck']['output_pdf']
+    prs=Presentation(pp);pdf=fitz.open(pf);fail=[]
+    counts={'spec':len(records),'pptx':len(prs.slides),'pdf':len(pdf)}
+    if set(counts.values())!={m['total_slides']}:fail.append(['page_counts',counts])
+    if sorted(s['parent_154_page'] for s in records if 'parent_154_page' in s)!=list(range(1,155)):fail.append(['parent_coverage'])
+    if sorted(s['original_slide'] for s in records if 'original_slide' in s)!=list(range(1,77)):fail.append(['original_coverage'])
+    for p,b in m['sources'].items():
+        if sha(p)!=b['sha256']:fail.append(['source_hash',p])
+    for p,h in m['protected_sources'].items():
+        if sha(p)!=h:fail.append(['protected_hash',p])
+    original=json.loads(Path(m['parent_spec']).read_text())['slides']
+    for s in records:
+        if 'parent_154_page' in s and s['parent_154_page']!=7 and s.get('figures',[])!=original[s['parent_154_page']-1].get('figures',[]):
+            fail.append(['parent_picture_replaced',s['parent_154_page']])
+    images=[];fonts=set()
+    for n,(slide,s) in enumerate(zip(prs.slides,records),1):
+        images.append(H._image_proofs(slide,s,n,fail))
+        for shape in slide.shapes:
+            if shape.left < -2 or shape.top < -2 or shape.left+shape.width > prs.slide_width+2 or shape.top+shape.height > prs.slide_height+2:
+                fail.append(['shape_bounds',n,shape.name])
+            for tf in H._text_frames(shape):
+                for paragraph in tf.paragraphs:
+                    for run in paragraph.runs:fonts.update(H._font_names(run,paragraph))
+    if fonts!={'Arial'}:fail.append(['font',sorted(fonts)])
+    for n,page in enumerate(pdf,1):
+        for block in page.get_text('dict')['blocks']:
+            for line in block.get('lines',[]):
+                for span in line.get('spans',[]):
+                    x0,y0,x1,y1=span['bbox']
+                    if x0<-.75 or y0<-.75 or x1>page.rect.width+.75 or y1>page.rect.height+.75:
+                        fail.append(['pdf_text_bounds',n,span['text']])
+    renders=H._render(pdf,records)
+    m['output']['pdf']={'path':str(pf),'sha256':sha(pf)}
+    (A/'manifest.json').write_text(json.dumps(m,indent=2)+'\n')
+    report=dict(schema='pbisc.figure1-revision.qa.v1',checked_at=datetime.now(timezone.utc).isoformat(),
+                status='automated_pass_visual_pending' if not fail else 'failed',automated_checks_pass=not fail,
+                failures=fail,counts=counts,parent154_represented=True,modified_parent_pages=[7],original76_preserved=True,
+                newpages=m['newpages'],groups=m['groups'],image_checks=images,renders=renders,fonts=sorted(fonts),
+                artifacts=m['output'],visual_review={'status':'pending'})
+    (Q/'VERIFICATION.json').write_text(json.dumps(report,indent=2)+'\n')
+    print(json.dumps({'pass':not fail,'counts':counts,'failures':fail}))
+    if fail:raise SystemExit(1)
+
+
+if __name__=='__main__':main()

paper-releases/pbisc-diffusion/updates/20261014T113716Z-figure1-overview-training-corpus/diffusion/metadata/request.json

@@ -0,0 +1,337 @@
+{
+  "assets": [
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/AGENTS.md",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/ara/trace/exploration_tree.yaml",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/ara/trace/pm_reasoning_log.yaml",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/ara/trace/sessions/2026-10-14_001.yaml",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/ara/trace/sessions/session_index.yaml",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261014_figure1.pdf",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261014_figure1.pptx",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/README.md",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/README.md",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/manifest.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/overview/result.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/page_map.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/VERIFICATION.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact01.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact02.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact03.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact04.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact05.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact06.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact07.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact08.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact09.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact10.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact11.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact12.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/contact13.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/slide001.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/slide006.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/slide007.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/slide008.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/qa/slide009.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/render.log",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "code",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/render.py",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/spec.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/VERIFICATION.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/celltype_by_split.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/hashes.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/result.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/shard_checks.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/source_hashes.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/training_celltype_counts.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/training_condition_counts.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/training_dataset_counts.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure1_revision_20261014/train_corpus/verified_data.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/figure_assignment_20261013/STRENGTHENING_PROPOSAL.md",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "figure",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figure1_revision_20261014/overview/experimental_overview.pdf",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "figure",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figure1_revision_20261014/overview/experimental_overview.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "figure",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figure1_revision_20261014/overview/experimental_overview.svg",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "figure",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figure1_revision_20261014/train_corpus/training_corpus_composition.pdf",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "figure",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figure1_revision_20261014/train_corpus/training_corpus_composition.png",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "figure",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/figures/figure1_revision_20261014/train_corpus/training_corpus_composition.svg",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "code",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/build_figure1_master_20261014.py",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "code",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/figure1_overview_20261014.py",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "code",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/figure1_train_corpus_20261014.py",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "code",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/tools/verify_figure1_master_20261014.py",
+      "state": "completed"
+    }
+  ],
+  "inference_locations": [],
+  "project_id": "pbisc-diffusion",
+  "summary": "Figure1a organic vector redesign and exact33.14M training-corpus mainFigure1b;155-page deck, no model experiments; metadata-only aggregation and fullsource preservation."
+}