Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261008T022410Z-deconv-scsemi-deck/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores/ye2026_sepsis/scsemi_budget_20261002/fold3/README.md

@@ -0,0 +1,73 @@
+# score_ft.py output: ye2026_sepsis, 16 scored donors
+
+Arms: scsemi_n5__raw, scsemi_n10__raw, scsemi_n20__raw, scsemi_n40__raw.  Every arm is scored on the same donors against the same observed reference.
+
+## Inputs and fixed contracts
+- Observed reference: `/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/reference.npz` (sha bac12da8a241), built from datadir:/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/data/ye2026_sepsis;
+  observed cells sum exactly to the features pseudobulk of every donor. Observed labels: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/observed_labels/labels/labels.parquet.
+- Generated cells: `generated_counts` read out once with Poisson(seed 20260916) over the **whole NPZ in row order**
+  (`poisson_readout.poisson_integerise`), then the scored donors' rows. These integer counts are used for labels, embedding,
+  depth and the contrast. Pseudobulk uses the summed `generated_counts` (rates), as `score_generated.py`.
+- Labels: frozen f01common scANVI (`predict_scanvi.py`, CPU, zero query epochs), six-type gate ['T cell', 'B cell', 'Monocyte', 'NK cell', 'DC', 'Other']; query genes =
+  the 1,006-gene annotation list minus genes the dataset's sc assay did not measure in every sample (1006 genes).
+- Scoring genes: 6k panel genes measured by both assays in every sample of the dataset (5975 genes).
+
+## Definitions
+1. **Composition**: fraction of each donor's cells per six-type label (all generated cells; all observed QC cells).
+   `comp_l1` = sum over the six labels of |generated - observed| (immicro_common.l1). `comp_l1_obs_loo_null` = L1 of the
+   donor's observed composition to the mean of the *other scored donors'* observed compositions. `composition_top1` =
+   fraction of donors whose generated composition is closest (L1) to their own observed composition.
+2. **Pseudobulk**: `pb_r` = Pearson r of CPM-log1p(summed generated_counts) vs CPM-log1p(observed pseudobulk) on the scoring
+   genes (evaluate_realbulk.cpm_log1p / paired_r). `pb_top1` = the donor's generated pseudobulk is closest (Pearson) to its own
+   observed pseudobulk among the 16 scored donors (immicro_common.retrieval = evaluate_realbulk.top1 arithmetic).
+3. **Precision / coverage / density** (stage_b_embed.py functions): per-cell CP10K over the scoring genes + log1p; PCA-50
+   (randomized, random_state 20260925) and k = 10 real-real k-NN balls fitted on **R1 of the scored donors only**.
+   Reference rule (independent of fold and arm): per donor, cells in canonical order (library_id+barcode), one
+   permutation from `default_rng([20260925, uint64(sha256(donor)[:8])])`; R1 = first n, R2 = next n,
+   n = min(150, n_cells // 2) (2400 R1 cells here). Generated query: per donor n cells (the donor's R1 size),
+   without replacement, `default_rng([20260926, same key])`. Precision = fraction of generated cells inside any R1 ball
+   (Kynkaanniemi 2019); coverage = fraction of R1 cells whose ball holds a generated cell; density = Naeem 2020.
+   Per-donor precision = that donor's generated cells; per-donor coverage = that donor's R1 cells (the pooled values are
+   their n-weighted means; plain means when every donor has 150 reference cells). R2 (held-out real cells of the same donors) is the
+   perfect-generator control: precision 0.9967, coverage 1.0000.
+   **Depth matched**: stage_b_embed depth matching run pairwise (R1 vs this arm): per donor and within-donor rank of detected
+   genes, both cells are thinned by UMI sampling without replacement to the smaller detected-gene count; PCA kept from
+   un-thinned R1, radii recomputed on thinned R1; fresh rng(20260929) per arm. R2 control: precision
+   0.9971, coverage 1.0000.
+4. **Monocyte %** = 100 x Monocyte fraction (all cells) per donor; **median UMI per cell** = median over cells of the
+   Poisson-read (generated) / observed counts summed over the scoring genes (per donor, and pooled).
+5. **Group contrast** (T, B, Monocyte, NK): case groups ['adult_sepsis', 'ped_sepsis'] (n = 13) vs control groups ['adult_HC', 'adult_ICU_ctrl', 'ped_HC', 'ped_ICU_ctrl'] (n = 3); rule: explicit --case-groups / --control-groups; groups from features groups.
+   Genes per type = scoring genes detected (count > 0) in >= 1 % of the scored donors' observed
+   cells of that type. Primary statistic `mean_cp10k`: per-cell CP10K over the scoring genes; per donor the mean over its
+   cells of the type (donors with >= 1 such cell); group mean = mean over the group's donors (donor-balanced);
+   logFC = log2(case mean + 1) - log2(control mean + 1). Secondary `pseudobulk_log1p_cpm`: the SLE benchmark evaluator's
+   statistic (donor x type summed counts, log1p CPM over the scoring genes, mean case - mean control;
+   evaluate_generated_vs_truth.group_log_fold_change). Both are computed on the generated cells (their frozen labels) and on
+   the observed cells; `r_generated` = Pearson r of the two logFC vectors. Broadcast baseline `r_broadcast`: the same
+   observed vector against the logFC of the input bulk (condition NPZ `bulk_counts` of the arm, scoring genes, per-donor
+   CP10K resp. log1p CPM, group means). `partial_r_generated_given_broadcast`: r of the OLS residuals of observed and
+   generated logFC on the bulk logFC (residual_specificity.py).
+6. **Top-50 DEG overlap** (contrast.csv rows `statistic = welch_top50_deg`; summary `deg_top50`): the SLE benchmark's score (`total_top50_overlap` of
+   `bulk2sc-paper/scripts/evaluate_generated_vs_truth.py`, Welch backend: the "top-50 DEG 81 / 84" of the final model), run
+   with the evaluator's own functions on the same donor x type pseudobulks as the contrast. Per cell type, observed and
+   generated separately: donor x type summed counts on all 5975 scoring genes (donors of the two groups with
+   >= 1 cell of the type) -> `welch_bh_deg` (log1p CPM, two-sided Welch t-test case vs control, Benjamini-Hochberg over the
+   genes with a defined p-value; needs >= 2 donors per group) -> `top_deg_genes` (FDR < 0.05, descending |logFC|,
+   first 50; shorter when fewer genes pass). `top50_overlap` = |observed list & generated list|; `top50_jaccard`
+   the evaluator's Jaccard. Summary `sum` = total over the scored types (the quantity reported as 81), `max` = the sum of
+   the observed list lengths (the "of 262"), `sum_4type` = T + B + Monocyte + NK. A type is scored when its observed test
+   is defined; an arm whose generated test is undefined for a scored type gets overlap 0. Types: T cell, B cell, Monocyte, NK cell, DC
+   (aggregates: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/deg_reference.npz).
+   **Broadcast analogue** (`top50_overlap_broadcast`, as `sle_broadcast.py` `broadcast_input_bulk`; not part of the
+   published benchmark): the same test on the arm's condition bulk_counts (scoring genes, the 13 + 3
+   donors), one whole-PBMC top-50 list, compared with every type's observed list.
+
+## Files
+- `per_donor.csv`: one row per arm x donor (all per-donor quantities above).
+- `summary.json`: per arm the donor means / pooled values, provenance (NPZ sha, receipt check, generation metadata,
+  Poisson nnz of the whole NPZ, annotation contract), and the reference block (PCA, R2 control).
+- `contrast.csv`: per arm x cell type x statistic (`mean_cp10k`, `pseudobulk_log1p_cpm`: logFC r; `welch_top50_deg`:
+  top-50 DEG overlap, T / B / Monocyte / NK / DC; columns of the other kind are empty).
+- `arms/<arm>/labels/` (predict_scanvi.py labels + manifest), `arms/<arm>/generated_cells.parquet` (NPZ row, donor, label,
+  UMI on the scoring genes, P/C subsample flag), `arms/<arm>/deg_top50_lists.csv` (observed, generated and broadcast
+  top-50 gene lists, Ensembl ids, rank order).

paper-releases/pbisc-diffusion/updates/20261008T022410Z-deconv-scsemi-deck/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores/ye2026_sepsis/scsemi_budget_20261002/fold4/README.md

@@ -0,0 +1,73 @@
+# score_ft.py output: ye2026_sepsis, 16 scored donors
+
+Arms: scsemi_n5__raw, scsemi_n10__raw, scsemi_n20__raw, scsemi_n40__raw.  Every arm is scored on the same donors against the same observed reference.
+
+## Inputs and fixed contracts
+- Observed reference: `/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/reference.npz` (sha bac12da8a241), built from datadir:/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/data/ye2026_sepsis;
+  observed cells sum exactly to the features pseudobulk of every donor. Observed labels: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/observed_labels/labels/labels.parquet.
+- Generated cells: `generated_counts` read out once with Poisson(seed 20260916) over the **whole NPZ in row order**
+  (`poisson_readout.poisson_integerise`), then the scored donors' rows. These integer counts are used for labels, embedding,
+  depth and the contrast. Pseudobulk uses the summed `generated_counts` (rates), as `score_generated.py`.
+- Labels: frozen f01common scANVI (`predict_scanvi.py`, CPU, zero query epochs), six-type gate ['T cell', 'B cell', 'Monocyte', 'NK cell', 'DC', 'Other']; query genes =
+  the 1,006-gene annotation list minus genes the dataset's sc assay did not measure in every sample (1006 genes).
+- Scoring genes: 6k panel genes measured by both assays in every sample of the dataset (5975 genes).
+
+## Definitions
+1. **Composition**: fraction of each donor's cells per six-type label (all generated cells; all observed QC cells).
+   `comp_l1` = sum over the six labels of |generated - observed| (immicro_common.l1). `comp_l1_obs_loo_null` = L1 of the
+   donor's observed composition to the mean of the *other scored donors'* observed compositions. `composition_top1` =
+   fraction of donors whose generated composition is closest (L1) to their own observed composition.
+2. **Pseudobulk**: `pb_r` = Pearson r of CPM-log1p(summed generated_counts) vs CPM-log1p(observed pseudobulk) on the scoring
+   genes (evaluate_realbulk.cpm_log1p / paired_r). `pb_top1` = the donor's generated pseudobulk is closest (Pearson) to its own
+   observed pseudobulk among the 16 scored donors (immicro_common.retrieval = evaluate_realbulk.top1 arithmetic).
+3. **Precision / coverage / density** (stage_b_embed.py functions): per-cell CP10K over the scoring genes + log1p; PCA-50
+   (randomized, random_state 20260925) and k = 10 real-real k-NN balls fitted on **R1 of the scored donors only**.
+   Reference rule (independent of fold and arm): per donor, cells in canonical order (library_id+barcode), one
+   permutation from `default_rng([20260925, uint64(sha256(donor)[:8])])`; R1 = first n, R2 = next n,
+   n = min(150, n_cells // 2) (2400 R1 cells here). Generated query: per donor n cells (the donor's R1 size),
+   without replacement, `default_rng([20260926, same key])`. Precision = fraction of generated cells inside any R1 ball
+   (Kynkaanniemi 2019); coverage = fraction of R1 cells whose ball holds a generated cell; density = Naeem 2020.
+   Per-donor precision = that donor's generated cells; per-donor coverage = that donor's R1 cells (the pooled values are
+   their n-weighted means; plain means when every donor has 150 reference cells). R2 (held-out real cells of the same donors) is the
+   perfect-generator control: precision 0.9975, coverage 1.0000.
+   **Depth matched**: stage_b_embed depth matching run pairwise (R1 vs this arm): per donor and within-donor rank of detected
+   genes, both cells are thinned by UMI sampling without replacement to the smaller detected-gene count; PCA kept from
+   un-thinned R1, radii recomputed on thinned R1; fresh rng(20260929) per arm. R2 control: precision
+   0.9958, coverage 1.0000.
+4. **Monocyte %** = 100 x Monocyte fraction (all cells) per donor; **median UMI per cell** = median over cells of the
+   Poisson-read (generated) / observed counts summed over the scoring genes (per donor, and pooled).
+5. **Group contrast** (T, B, Monocyte, NK): case groups ['adult_sepsis', 'ped_sepsis'] (n = 12) vs control groups ['adult_HC', 'adult_ICU_ctrl', 'ped_HC', 'ped_ICU_ctrl'] (n = 4); rule: explicit --case-groups / --control-groups; groups from features groups.
+   Genes per type = scoring genes detected (count > 0) in >= 1 % of the scored donors' observed
+   cells of that type. Primary statistic `mean_cp10k`: per-cell CP10K over the scoring genes; per donor the mean over its
+   cells of the type (donors with >= 1 such cell); group mean = mean over the group's donors (donor-balanced);
+   logFC = log2(case mean + 1) - log2(control mean + 1). Secondary `pseudobulk_log1p_cpm`: the SLE benchmark evaluator's
+   statistic (donor x type summed counts, log1p CPM over the scoring genes, mean case - mean control;
+   evaluate_generated_vs_truth.group_log_fold_change). Both are computed on the generated cells (their frozen labels) and on
+   the observed cells; `r_generated` = Pearson r of the two logFC vectors. Broadcast baseline `r_broadcast`: the same
+   observed vector against the logFC of the input bulk (condition NPZ `bulk_counts` of the arm, scoring genes, per-donor
+   CP10K resp. log1p CPM, group means). `partial_r_generated_given_broadcast`: r of the OLS residuals of observed and
+   generated logFC on the bulk logFC (residual_specificity.py).
+6. **Top-50 DEG overlap** (contrast.csv rows `statistic = welch_top50_deg`; summary `deg_top50`): the SLE benchmark's score (`total_top50_overlap` of
+   `bulk2sc-paper/scripts/evaluate_generated_vs_truth.py`, Welch backend: the "top-50 DEG 81 / 84" of the final model), run
+   with the evaluator's own functions on the same donor x type pseudobulks as the contrast. Per cell type, observed and
+   generated separately: donor x type summed counts on all 5975 scoring genes (donors of the two groups with
+   >= 1 cell of the type) -> `welch_bh_deg` (log1p CPM, two-sided Welch t-test case vs control, Benjamini-Hochberg over the
+   genes with a defined p-value; needs >= 2 donors per group) -> `top_deg_genes` (FDR < 0.05, descending |logFC|,
+   first 50; shorter when fewer genes pass). `top50_overlap` = |observed list & generated list|; `top50_jaccard`
+   the evaluator's Jaccard. Summary `sum` = total over the scored types (the quantity reported as 81), `max` = the sum of
+   the observed list lengths (the "of 262"), `sum_4type` = T + B + Monocyte + NK. A type is scored when its observed test
+   is defined; an arm whose generated test is undefined for a scored type gets overlap 0. Types: T cell, B cell, Monocyte, NK cell, DC
+   (aggregates: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/deg_reference.npz).
+   **Broadcast analogue** (`top50_overlap_broadcast`, as `sle_broadcast.py` `broadcast_input_bulk`; not part of the
+   published benchmark): the same test on the arm's condition bulk_counts (scoring genes, the 12 + 4
+   donors), one whole-PBMC top-50 list, compared with every type's observed list.
+
+## Files
+- `per_donor.csv`: one row per arm x donor (all per-donor quantities above).
+- `summary.json`: per arm the donor means / pooled values, provenance (NPZ sha, receipt check, generation metadata,
+  Poisson nnz of the whole NPZ, annotation contract), and the reference block (PCA, R2 control).
+- `contrast.csv`: per arm x cell type x statistic (`mean_cp10k`, `pseudobulk_log1p_cpm`: logFC r; `welch_top50_deg`:
+  top-50 DEG overlap, T / B / Monocyte / NK / DC; columns of the other kind are empty).
+- `arms/<arm>/labels/` (predict_scanvi.py labels + manifest), `arms/<arm>/generated_cells.parquet` (NPZ row, donor, label,
+  UMI on the scoring genes, P/C subsample flag), `arms/<arm>/deg_top50_lists.csv` (observed, generated and broadcast
+  top-50 gene lists, Ensembl ids, rank order).

paper-releases/pbisc-diffusion/updates/20261008T022410Z-deconv-scsemi-deck/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores/ye2026_sepsis/scsemi_budget_20261002/pooled/README.md

@@ -0,0 +1,73 @@
+# score_ft.py output: ye2026_sepsis, 82 scored donors
+
+Arms: scsemi_n5__raw, scsemi_n10__raw, scsemi_n20__raw, scsemi_n40__raw.  Every arm is scored on the same donors against the same observed reference.
+
+## Inputs and fixed contracts
+- Observed reference: `/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/reference.npz` (sha bac12da8a241), built from datadir:/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/data/ye2026_sepsis;
+  observed cells sum exactly to the features pseudobulk of every donor. Observed labels: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/observed_labels/labels/labels.parquet.
+- Generated cells: `generated_counts` read out once with Poisson(seed 20260916) over the **whole NPZ in row order**
+  (`poisson_readout.poisson_integerise`), then the scored donors' rows. These integer counts are used for labels, embedding,
+  depth and the contrast. Pseudobulk uses the summed `generated_counts` (rates), as `score_generated.py`.
+- Labels: frozen f01common scANVI (`predict_scanvi.py`, CPU, zero query epochs), six-type gate ['T cell', 'B cell', 'Monocyte', 'NK cell', 'DC', 'Other']; query genes =
+  the 1,006-gene annotation list minus genes the dataset's sc assay did not measure in every sample (1006 genes).
+- Scoring genes: 6k panel genes measured by both assays in every sample of the dataset (5975 genes).
+
+## Definitions
+1. **Composition**: fraction of each donor's cells per six-type label (all generated cells; all observed QC cells).
+   `comp_l1` = sum over the six labels of |generated - observed| (immicro_common.l1). `comp_l1_obs_loo_null` = L1 of the
+   donor's observed composition to the mean of the *other scored donors'* observed compositions. `composition_top1` =
+   fraction of donors whose generated composition is closest (L1) to their own observed composition.
+2. **Pseudobulk**: `pb_r` = Pearson r of CPM-log1p(summed generated_counts) vs CPM-log1p(observed pseudobulk) on the scoring
+   genes (evaluate_realbulk.cpm_log1p / paired_r). `pb_top1` = the donor's generated pseudobulk is closest (Pearson) to its own
+   observed pseudobulk among the 82 scored donors (immicro_common.retrieval = evaluate_realbulk.top1 arithmetic).
+3. **Precision / coverage / density** (stage_b_embed.py functions): per-cell CP10K over the scoring genes + log1p; PCA-50
+   (randomized, random_state 20260925) and k = 10 real-real k-NN balls fitted on **R1 of the scored donors only**.
+   Reference rule (independent of fold and arm): per donor, cells in canonical order (library_id+barcode), one
+   permutation from `default_rng([20260925, uint64(sha256(donor)[:8])])`; R1 = first n, R2 = next n,
+   n = min(150, n_cells // 2) (12300 R1 cells here). Generated query: per donor n cells (the donor's R1 size),
+   without replacement, `default_rng([20260926, same key])`. Precision = fraction of generated cells inside any R1 ball
+   (Kynkaanniemi 2019); coverage = fraction of R1 cells whose ball holds a generated cell; density = Naeem 2020.
+   Per-donor precision = that donor's generated cells; per-donor coverage = that donor's R1 cells (the pooled values are
+   their n-weighted means; plain means when every donor has 150 reference cells). R2 (held-out real cells of the same donors) is the
+   perfect-generator control: precision 0.9805, coverage 0.9997.
+   **Depth matched**: stage_b_embed depth matching run pairwise (R1 vs this arm): per donor and within-donor rank of detected
+   genes, both cells are thinned by UMI sampling without replacement to the smaller detected-gene count; PCA kept from
+   un-thinned R1, radii recomputed on thinned R1; fresh rng(20260929) per arm. R2 control: precision
+   0.9796, coverage 0.9998.
+4. **Monocyte %** = 100 x Monocyte fraction (all cells) per donor; **median UMI per cell** = median over cells of the
+   Poisson-read (generated) / observed counts summed over the scoring genes (per donor, and pooled).
+5. **Group contrast** (T, B, Monocyte, NK): case groups ['adult_sepsis', 'ped_sepsis'] (n = 61) vs control groups ['adult_HC', 'adult_ICU_ctrl', 'ped_HC', 'ped_ICU_ctrl'] (n = 21); rule: explicit --case-groups / --control-groups; groups from features groups.
+   Genes per type = scoring genes detected (count > 0) in >= 1 % of the scored donors' observed
+   cells of that type. Primary statistic `mean_cp10k`: per-cell CP10K over the scoring genes; per donor the mean over its
+   cells of the type (donors with >= 1 such cell); group mean = mean over the group's donors (donor-balanced);
+   logFC = log2(case mean + 1) - log2(control mean + 1). Secondary `pseudobulk_log1p_cpm`: the SLE benchmark evaluator's
+   statistic (donor x type summed counts, log1p CPM over the scoring genes, mean case - mean control;
+   evaluate_generated_vs_truth.group_log_fold_change). Both are computed on the generated cells (their frozen labels) and on
+   the observed cells; `r_generated` = Pearson r of the two logFC vectors. Broadcast baseline `r_broadcast`: the same
+   observed vector against the logFC of the input bulk (condition NPZ `bulk_counts` of the arm, scoring genes, per-donor
+   CP10K resp. log1p CPM, group means). `partial_r_generated_given_broadcast`: r of the OLS residuals of observed and
+   generated logFC on the bulk logFC (residual_specificity.py).
+6. **Top-50 DEG overlap** (contrast.csv rows `statistic = welch_top50_deg`; summary `deg_top50`): the SLE benchmark's score (`total_top50_overlap` of
+   `bulk2sc-paper/scripts/evaluate_generated_vs_truth.py`, Welch backend: the "top-50 DEG 81 / 84" of the final model), run
+   with the evaluator's own functions on the same donor x type pseudobulks as the contrast. Per cell type, observed and
+   generated separately: donor x type summed counts on all 5975 scoring genes (donors of the two groups with
+   >= 1 cell of the type) -> `welch_bh_deg` (log1p CPM, two-sided Welch t-test case vs control, Benjamini-Hochberg over the
+   genes with a defined p-value; needs >= 2 donors per group) -> `top_deg_genes` (FDR < 0.05, descending |logFC|,
+   first 50; shorter when fewer genes pass). `top50_overlap` = |observed list & generated list|; `top50_jaccard`
+   the evaluator's Jaccard. Summary `sum` = total over the scored types (the quantity reported as 81), `max` = the sum of
+   the observed list lengths (the "of 262"), `sum_4type` = T + B + Monocyte + NK. A type is scored when its observed test
+   is defined; an arm whose generated test is undefined for a scored type gets overlap 0. Types: T cell, B cell, Monocyte, NK cell, DC
+   (aggregates: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/ye2026_sepsis/deg_reference.npz).
+   **Broadcast analogue** (`top50_overlap_broadcast`, as `sle_broadcast.py` `broadcast_input_bulk`; not part of the
+   published benchmark): the same test on the arm's condition bulk_counts (scoring genes, the 61 + 21
+   donors), one whole-PBMC top-50 list, compared with every type's observed list.
+
+## Files
+- `per_donor.csv`: one row per arm x donor (all per-donor quantities above).
+- `summary.json`: per arm the donor means / pooled values, provenance (NPZ sha, receipt check, generation metadata,
+  Poisson nnz of the whole NPZ, annotation contract), and the reference block (PCA, R2 control).
+- `contrast.csv`: per arm x cell type x statistic (`mean_cp10k`, `pseudobulk_log1p_cpm`: logFC r; `welch_top50_deg`:
+  top-50 DEG overlap, T / B / Monocyte / NK / DC; columns of the other kind are empty).
+- `arms/<arm>/labels/` (predict_scanvi.py labels + manifest), `arms/<arm>/generated_cells.parquet` (NPZ row, donor, label,
+  UMI on the scoring genes, P/C subsample flag), `arms/<arm>/deg_top50_lists.csv` (observed, generated and broadcast
+  top-50 gene lists, Ensembl ids, rank order).

paper-releases/pbisc-diffusion/updates/20261008T022410Z-deconv-scsemi-deck/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores_contrast/ye2026_sepsis/scsemi_cv_20261002/pooled/cat.csv

@@ -0,0 +1,61 @@
+dataset,design,unit,arm,predictor,celltype,k,k_eff,cat_same_sign,cat_any_sign,chance_same_sign,chance_any_sign,G_both,pred_defined,n_tested_obs,n_tested_pred
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,T cell,10,10,0.0,0.0,0.0011496895838123706,0.002299379167624741,4349,True,5810,4349
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,T cell,20,20,0.05,0.05,0.002299379167624741,0.004598758335249482,4349,True,5810,4349
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,T cell,50,50,0.02,0.02,0.005748447919061853,0.011496895838123706,4349,True,5810,4349
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,T cell,100,100,0.01,0.01,0.011496895838123706,0.022993791676247412,4349,True,5810,4349
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,T cell,200,200,0.01,0.03,0.022993791676247412,0.045987583352494824,4349,True,5810,4349
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,T cell,500,500,0.056,0.102,0.05748447919061853,0.11496895838123707,4349,True,5810,4349
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,T cell,10,10,0.0,0.0,0.0008735150244584207,0.0017470300489168414,5724,True,5810,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,T cell,20,20,0.0,0.0,0.0017470300489168414,0.0034940600978336828,5724,True,5810,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,T cell,50,50,0.04,0.04,0.004367575122292104,0.008735150244584208,5724,True,5810,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,T cell,100,100,0.09,0.09,0.008735150244584208,0.017470300489168415,5724,True,5810,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,T cell,200,200,0.1,0.1,0.017470300489168415,0.03494060097833683,5724,True,5810,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,T cell,500,500,0.14,0.144,0.043675751222921035,0.08735150244584207,5724,True,5810,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,B cell,10,10,0.2,0.2,0.001937984496124031,0.003875968992248062,2580,True,5752,2582
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,B cell,20,20,0.15,0.15,0.003875968992248062,0.007751937984496124,2580,True,5752,2582
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,B cell,50,50,0.08,0.08,0.009689922480620155,0.01937984496124031,2580,True,5752,2582
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,B cell,100,100,0.08,0.08,0.01937984496124031,0.03875968992248062,2580,True,5752,2582
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,B cell,200,200,0.12,0.12,0.03875968992248062,0.07751937984496124,2580,True,5752,2582
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,B cell,500,500,0.146,0.176,0.09689922480620156,0.1937984496124031,2580,True,5752,2582
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,B cell,10,10,0.0,0.0,0.0008819897689186806,0.0017639795378373612,5669,True,5752,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,B cell,20,20,0.0,0.0,0.0017639795378373612,0.0035279590756747223,5669,True,5752,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,B cell,50,50,0.02,0.02,0.0044099488445934025,0.008819897689186805,5669,True,5752,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,B cell,100,100,0.04,0.04,0.008819897689186805,0.01763979537837361,5669,True,5752,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,B cell,200,200,0.06,0.06,0.01763979537837361,0.03527959075674722,5669,True,5752,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,B cell,500,500,0.122,0.128,0.044099488445934025,0.08819897689186805,5669,True,5752,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,Monocyte,10,10,0.0,0.0,0.0015060240963855422,0.0030120481927710845,3320,True,5754,3320
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,Monocyte,20,20,0.1,0.1,0.0030120481927710845,0.006024096385542169,3320,True,5754,3320
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,Monocyte,50,50,0.08,0.08,0.007530120481927711,0.015060240963855422,3320,True,5754,3320
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,Monocyte,100,100,0.08,0.08,0.015060240963855422,0.030120481927710843,3320,True,5754,3320
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,Monocyte,200,200,0.06,0.065,0.030120481927710843,0.060240963855421686,3320,True,5754,3320
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,Monocyte,500,500,0.106,0.14,0.07530120481927711,0.15060240963855423,3320,True,5754,3320
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,Monocyte,10,10,0.1,0.1,0.0008813678829543451,0.0017627357659086903,5673,True,5754,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,Monocyte,20,20,0.1,0.1,0.0017627357659086903,0.0035254715318173806,5673,True,5754,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,Monocyte,50,50,0.1,0.1,0.004406839414771726,0.008813678829543451,5673,True,5754,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,Monocyte,100,100,0.13,0.13,0.008813678829543451,0.017627357659086903,5673,True,5754,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,Monocyte,200,200,0.23,0.23,0.017627357659086903,0.035254715318173806,5673,True,5754,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,Monocyte,500,500,0.264,0.264,0.04406839414771726,0.08813678829543452,5673,True,5754,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,NK cell,10,10,0.0,0.0,0.001954652071931196,0.003909304143862392,2558,True,5623,2558
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,NK cell,20,20,0.0,0.0,0.003909304143862392,0.007818608287724784,2558,True,5623,2558
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,NK cell,50,50,0.0,0.0,0.009773260359655981,0.019546520719311962,2558,True,5623,2558
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,NK cell,100,100,0.01,0.01,0.019546520719311962,0.039093041438623924,2558,True,5623,2558
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,NK cell,200,200,0.035,0.05,0.039093041438623924,0.07818608287724785,2558,True,5623,2558
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,NK cell,500,500,0.074,0.14,0.09773260359655982,0.19546520719311963,2558,True,5623,2558
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,NK cell,10,10,0.0,0.0,0.0009007386056566384,0.0018014772113132769,5551,True,5623,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,NK cell,20,20,0.0,0.0,0.0018014772113132769,0.0036029544226265538,5551,True,5623,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,NK cell,50,50,0.0,0.0,0.004503693028283192,0.009007386056566384,5551,True,5623,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,NK cell,100,100,0.04,0.04,0.009007386056566384,0.01801477211313277,5551,True,5623,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,NK cell,200,200,0.075,0.08,0.01801477211313277,0.03602954422626554,5551,True,5623,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,NK cell,500,500,0.082,0.098,0.04503693028283192,0.09007386056566384,5551,True,5623,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,DC,10,10,0.0,0.0,0.005213764337851929,0.010427528675703858,959,True,5330,963
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,DC,20,20,0.0,0.0,0.010427528675703858,0.020855057351407715,959,True,5330,963
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,DC,50,50,0.02,0.02,0.026068821689259645,0.05213764337851929,959,True,5330,963
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,DC,100,100,0.02,0.02,0.05213764337851929,0.10427528675703858,959,True,5330,963
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,DC,200,200,0.03,0.03,0.10427528675703858,0.20855057351407716,959,True,5330,963
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,generated:scsemi__raw,DC,500,500,0.05,0.052,0.26068821689259647,0.5213764337851929,959,True,5330,963
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,DC,10,10,0.0,0.0,0.000946969696969697,0.001893939393939394,5280,True,5330,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,DC,20,20,0.0,0.0,0.001893939393939394,0.003787878787878788,5280,True,5330,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,DC,50,50,0.08,0.08,0.004734848484848485,0.00946969696969697,5280,True,5330,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,DC,100,100,0.06,0.06,0.00946969696969697,0.01893939393939394,5280,True,5330,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,DC,200,200,0.105,0.105,0.01893939393939394,0.03787878787878788,5280,True,5330,5858
+ye2026_sepsis,scored,scsemi_cv_20261002/pooled,scsemi__raw,broadcast:conditions_pooled_raw.npz@3b2d8ef2,DC,500,500,0.142,0.16,0.04734848484848485,0.0946969696969697,5280,True,5330,5858