Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261015T131236Z-corpus-five-marker-codetection/diffusion/dataset/server_1/model8/plans/corpus_macrophage_markers_20261015/summary.csv

@@ -0,0 +1,8 @@
+dataset,flagged,cells,samples,dataset_scoped_donors,C1QA_detected,C1QA_fraction,C1QA_mean_raw_counts,CD163_detected,CD163_fraction,CD163_mean_raw_counts,CD68_detected,CD68_fraction,CD68_mean_raw_counts,WARS1_detected,WARS1_fraction,WARS1_mean_raw_counts,LGALS3_detected,LGALS3_fraction,LGALS3_mean_raw_counts,exactly_0_detected,exactly_1_detected,exactly_2_detected,exactly_3_detected,exactly_4_detected,exactly_5_detected,all_five_detected,all_five_detected_samples,all_five_detected_dataset_scoped_donors,c1qa_plus_at_least_three_others,c1qa_plus_at_least_three_others_samples,c1qa_plus_at_least_three_others_dataset_scoped_donors
+ALL,False,33219,1398,779,32958,0.9921430506637767,12.234684969445198,1818,0.05472771606610675,0.09467473433878203,32607,0.9815768084529938,6.9743520274541675,28867,0.8689906378879557,3.788584846021855,30962,0.932056955356874,5.272735482705681,10,62,532,5693,25603,1319,1319,580,204,26903,1384,765
+006,False,2726,439,439,2677,0.9820249449743214,11.379677182685253,7,0.0025678650036683784,0.003301540719002201,2724,0.9992663242846662,12.546955245781364,2664,0.9772560528246516,7.42516507703595,2555,0.9372707263389581,4.842993396918562,0,0,16,252,2451,7,7,5,5,2458,434,434
+009,False,3058,86,86,3032,0.9914977109221713,12.782210595160235,198,0.06474820143884892,0.1105297580117724,2992,0.9784172661870504,5.9905166775670375,2636,0.8620013080444735,3.287115761935906,2877,0.9408109875735775,5.497383911052975,1,7,49,514,2347,140,140,56,56,2483,86,86
+065,False,25268,696,77,25115,0.9939449105588095,12.456545828716163,1446,0.05722653158144689,0.09755421877473484,24774,0.9804495804970714,5.973563400348267,21482,0.8501662181415228,3.1704923223048915,23923,0.9467706189646984,5.553110653791356,4,38,425,4400,19353,1048,1048,451,75,20389,696,77
+200,False,2167,177,177,2134,0.9847715736040609,9.950622981079833,167,0.07706506691278264,0.15366866635902168,2117,0.9769266266728196,13.022150438394092,2085,0.9621596677434241,7.128749423165667,1607,0.7415782187355792,2.2270419935394554,5,17,42,527,1452,124,124,68,68,1573,168,168
+ALL,True,903,351,351,899,0.9955703211517165,19.383167220376524,131,0.1450719822812846,0.2469545957918051,892,0.9878183831672204,10.091915836101883,851,0.9424141749723145,6.210409745293466,841,0.9313399778516057,5.904761904761905,0,1,4,96,693,109,109,86,86,801,339,339
+006,True,903,351,351,899,0.9955703211517165,19.383167220376524,131,0.1450719822812846,0.2469545957918051,892,0.9878183831672204,10.091915836101883,851,0.9424141749723145,6.210409745293466,841,0.9313399778516057,5.904761904761905,0,1,4,96,693,109,109,86,86,801,339,339

paper-releases/pbisc-diffusion/updates/20261015T131236Z-corpus-five-marker-codetection/diffusion/dataset/server_1/model8/plans/corpus_macrophage_markers_20261015/verify_candidates.py

@@ -0,0 +1,128 @@
+#!/usr/bin/env python3
+"""Independent physical-row, gene-axis, and direct sparse-row replay checks."""
+from pathlib import Path
+import argparse
+import json
+import numpy as np
+import pandas as pd
+import pyarrow.parquet as pq
+import zarr
+import h5py
+from check_candidates import ROOT, M7, E2, BLOCK, GENES, IDS, COLS, receipt, write_json, sha
+
+ap = argparse.ArgumentParser()
+ap.add_argument('--selected-genes', required=True)
+ap.add_argument('--training-config', required=True)
+args = ap.parse_args()
+v = pd.read_parquet(ROOT / 'candidate_marker_counts.parquet')
+var = pd.read_parquet(E2 / 'metadata/var.parquet')
+assert var.iloc[COLS].gene_id.tolist() == IDS
+assert var.iloc[COLS].panel_index.tolist() == COLS.tolist()
+assert var.iloc[COLS].gene_symbol.tolist() == ['C1QA', 'CD163', 'CD68', 'WARS', 'LGALS3']
+sel = pd.read_csv(args.selected_genes)
+assert set(IDS).issubset(set(sel.gene_id))
+assert len(sel) == 6000 and not sel.gene_id.duplicated().any()
+cfg = json.loads(Path(args.training_config).read_text())['data']
+assert cfg['n_genes'] == 6000
+assert sha(cfg['gene_selection_manifest']) == cfg['gene_selection_manifest_sha256']
+assert sha(cfg['gene_positions_path']) == cfg['gene_positions_sha256']
+manifest = json.loads(Path(cfg['gene_selection_manifest']).read_text())
+assert Path(manifest['selected_genes_path']) == Path(args.selected_genes)
+assert sha(args.selected_genes) == manifest['selected_genes_sha256']
+assert np.array_equal(np.load(cfg['gene_positions_path']), sel.panel_position)
+assert set(COLS).issubset(set(sel.panel_position))
+assert var.iloc[sel.panel_position.to_numpy()].gene_id.tolist() == sel.gene_id.tolist()
+physical_checks, direct_checks, raw_source_checks = [], [], []
+first_by_dataset, boundary_by_dataset = set(), set()
+raw_source_by_dataset = set()
+for sid, d in v.groupby('shard_id', sort=True):
+    rows = d.shard_row.to_numpy()
+    ep = E2 / 'metadata/obs_sc.parquet' / (sid + '.parquet')
+    bp = BLOCK / 'metadata/obs_sc.parquet' / (sid + '.parquet')
+    e = pq.read_table(ep, columns=['row_index', 'source_row_index']).to_pandas()
+    b = pq.read_table(bp, columns=['row_index', 'source_row_index', 'shard_id']).to_pandas()
+    assert np.array_equal(e.iloc[rows].row_index, rows)
+    assert set(b.iloc[rows].shard_id) == {sid}
+    assert np.array_equal(e.iloc[rows].source_row_index, d.source_row_index)
+    assert np.array_equal(b.iloc[rows].source_row_index, d.source_row_index)
+    zp = BLOCK / 'counts/sc_counts' / (sid + '.zarr')
+    g = zarr.open_group(str(zp), mode='r')
+    assert g.attrs['shard_id'] == sid
+    assert g.attrs['shape'][0] == len(e) == len(b)
+    assert g.attrs['shape'][1] == len(var) == 20097
+    physical_checks.append({'shard_id': sid, 'selected_rows': len(d),
+                            'e2_row_index_equals_physical_row': True,
+                            'historical_source_row_index_is_not_csr_row': int((b.iloc[rows].row_index.to_numpy() != rows).sum()),
+                            'e2_and_historical_source_identity_match': True,
+                            'metadata_and_expression_shard_id_equal': True})
+    prefix = sid[:3]
+    chosen = set()
+    if prefix not in first_by_dataset:
+        chosen.update([d.index[0], d.index[-1]])
+        pos = d[d.all_five_detected]
+        if len(pos): chosen.add(pos.index[0])
+        first_by_dataset.add(prefix)
+    if prefix not in boundary_by_dataset:
+        ptr = np.asarray(g['indptr'][:])
+        boundary = (ptr[rows] // 1000000) != ((ptr[rows+1]-1) // 1000000)
+        if boundary.any():
+            chosen.add(d.index[np.flatnonzero(boundary)[0]])
+            boundary_by_dataset.add(prefix)
+    for ix in sorted(chosen):
+        r = v.loc[ix]
+        p = np.asarray(g['indptr'][int(r.shard_row):int(r.shard_row)+2])
+        a, b = int(p[0]), int(p[1])
+        indices = np.asarray(g['indices'][a:b])
+        vals = np.asarray(g['data'][a:b])
+        counts = np.array([vals[indices == j].sum() for j in COLS])
+        assert np.array_equal(counts, r[GENES].to_numpy(dtype=np.int64))
+        assert int(vals.sum()) == int(r.total_counts_20097)
+        direct_checks.append({'shard_id': sid, 'shard_row': int(r.shard_row),
+                              'five_counts': counts.tolist(), 'row_sum': int(vals.sum()),
+                              'crosses_chunk_boundary': a // 1000000 != (b-1) // 1000000})
+        if prefix not in raw_source_by_dataset:
+            hp = Path('/home/sj_server_1/gblee') / g.attrs['source_path']
+            with h5py.File(hp, 'r') as hf:
+                matrix_key = g.attrs['source_count_matrix']
+                vg = hf['raw/var' if matrix_key.startswith('raw/') else 'var']
+                raw_ids = [x.decode() if isinstance(x, bytes) else str(x) for x in vg[vg.attrs['_index']][:]]
+                assert len(raw_ids) == len(set(raw_ids))
+                rawpos = {gid: j for j, gid in enumerate(raw_ids)}
+                mx = hf[matrix_key]
+                assert mx.attrs['encoding-type'] == 'csr_matrix'
+                srcrow = int(r.source_row_index)
+                p = mx['indptr'][srcrow:srcrow+2]
+                ri, rv = mx['indices'][int(p[0]):int(p[1])], mx['data'][int(p[0]):int(p[1])]
+                rc = np.array([rv[ri == rawpos[gid]].sum() for gid in IDS])
+                assert np.array_equal(rc, counts)
+                canonical_positions = np.array([rawpos[gid] for gid in var.gene_id if gid in rawpos])
+                projected_sum = rv[np.isin(ri, canonical_positions)].sum()
+                assert projected_sum == int(r.total_counts_20097)
+            raw_source_checks.append({'path': str(hp), 'bytes': hp.stat().st_size, 'mtime_ns': hp.stat().st_mtime_ns,
+                                      'source_matrix': matrix_key, 'source_row_index': srcrow,
+                                      'shard_id': sid, 'shard_row': int(r.shard_row),
+                                      'original_h5ad_five_counts': rc.tolist(), 'canonical_projected_sum': int(projected_sum),
+                                      'five_counts_and_projected_sum_match': True})
+            raw_source_by_dataset.add(prefix)
+assert len(physical_checks) == 69 and sum(x['selected_rows'] for x in physical_checks) == 34122
+assert len(first_by_dataset) == 4
+assert len(raw_source_by_dataset) == 4
+assert len(direct_checks) >= 8
+data = {'status': 'PASS', 'physical_row_alignment': physical_checks, 'independent_direct_row_replays': direct_checks,
+        'original_h5ad_source_row_replays': raw_source_checks,
+        'historical_row_index_note': 'BLOCK metadata retains original H5AD offsets after QC. E2 row_index resets to contiguous CSR row position; source_row_index is the shared identity join. Direct original-H5AD row replay confirms the projected five counts and total for one cell per dataset.',
+        'gene_axis_raw_positions': COLS.tolist(), 'raw_gene_ids': IDS,
+        'raw_gene_symbols': var.iloc[COLS].gene_symbol.tolist(),
+        'sources': [receipt(E2 / 'metadata/var.parquet'), receipt(args.selected_genes), receipt(args.training_config),
+                    receipt(cfg['gene_selection_manifest']), receipt(cfg['gene_positions_path']),
+                    receipt(M7.parent / 'scripts/build_e2_full_source.py'),
+                    receipt(M7.parent / 'scripts/prepare_e2_celltype_latent_analysis.py'),
+                    receipt(ROOT / 'check_candidates.py'), receipt(ROOT / 'verify_candidates.py')]}
+write_json(ROOT / 'INDEPENDENT_VERIFICATION.json', data)
+r = json.loads((ROOT / 'RESULT.json').read_text())
+r['independent_verification'] = receipt(ROOT / 'INDEPENDENT_VERIFICATION.json')
+r['code'] = receipt(ROOT / 'check_candidates.py')
+r['gene_axis_and_training_panel_sources'] = data['sources'][:5]
+write_json(ROOT / 'RESULT.json', r)
+print(json.dumps({'status': 'PASS', 'physical_rows_checked': len(v), 'shards': len(physical_checks),
+                  'direct_row_replays': len(direct_checks), 'boundary_datasets': sorted(boundary_by_dataset)}))

paper-releases/pbisc-diffusion/updates/20261015T131236Z-corpus-five-marker-codetection/diffusion/metadata/inference_locations.json

@@ -0,0 +1,5 @@
+{
+  "locations": [],
+  "project_id": "pbisc-diffusion",
+  "update_id": "20261015T131236Z-corpus-five-marker-codetection"
+}

paper-releases/pbisc-diffusion/updates/20261015T131236Z-corpus-five-marker-codetection/diffusion/metadata/request.json

@@ -0,0 +1,115 @@
+{
+  "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/2026-10-15_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/plans/corpus_macrophage_markers_20261015/INDEPENDENT_VERIFICATION.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/PROTOCOL.md",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/REPORT.md",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/RESULT.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "data",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/candidate_marker_counts.parquet",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "code",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/check_candidates.py",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/expression_source_receipts.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "report",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/extraction_plan.json",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/sample_support.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "data",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/selected_training_candidates.parquet",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/source_label_summary.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "table",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/summary.csv",
+      "state": "completed"
+    },
+    {
+      "host": "server_1",
+      "kind": "code",
+      "source": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/corpus_macrophage_markers_20261015/verify_candidates.py",
+      "state": "completed"
+    }
+  ],
+  "inference_locations": [],
+  "project_id": "pbisc-diffusion",
+  "summary": "CPU-only exact training C1Q-candidate five-marker co-detection audit. User markers C1QA/CD163/CD68/WARS1/LGALS3. Verified1,319 unflagged+109flagged all-positive; scoped candidate subset, not confirmed macrophages/full-corpus prevalence. Source metadata and figures unchanged."
+}