Archive 4 checkpoints (1-4)

.gitattributes

@@ -259,3 +259,4 @@ paper-releases/pbisc-diffusion/updates/20261013T094815Z-all-celltypes-final-pale
 paper-releases/pbisc-diffusion/updates/20261013T094815Z-all-celltypes-final-palette-142/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/figures/final_palette_20261013/fig2_covid_hc_monocyte_arrows_depth1000.pdf filter=lfs diff=lfs merge=lfs -text
 paper-releases/pbisc-diffusion/updates/20261013T094815Z-all-celltypes-final-palette-142/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/figures/final_palette_20261013/native/gene_length_input.pdf filter=lfs diff=lfs merge=lfs -text
 paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261013_strengthened.pdf filter=lfs diff=lfs merge=lfs -text
+paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261013_strengthened.pptx filter=lfs diff=lfs merge=lfs -text

paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_expanded_master_20261013_strengthened.pptx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:627025a1a90172e6cccc81b905f064dd1229f761d4b6a10f462bad0a6dc765ca
+size 23031083

paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/README.md

@@ -0,0 +1,392 @@
+# PBISC-Diffusion expanded evidence master
+
+**Current working master:154 slides.** The142-page parent is preserved;12 panels add corpus detail, verified training/test splits, direct adaptation views, disease metrics and paper-grounded rare-population targets.
+
+[Current editable PPTX](PBISC-Diffusion_expanded_master_20261013_strengthened.pptx) · [Current PDF](PBISC-Diffusion_expanded_master_20261013_strengthened.pdf)
+
+- [Current changes and page guide](data/strengthening_20261013/README.md)
+- [Revised strengthening proposal](data/figure_assignment_20261013/STRENGTHENING_PROPOSAL.md)
+- [Page map](data/strengthening_20261013/page_map.csv) · [Verification](data/strengthening_20261013/qa/VERIFICATION.json)
+- Builder: `tools/build_strengthened_master_20261013.py`; CPU-only export/verification: `data/strengthening_20261013/render.py`.
+
+Approved method palette remains PBISC blue, scSemi mint, external comparators yellow. Biological category colours retain their meaning. Figure5 now contains original-paper targets and an observed marker diagnostic; exact rare-population recovery is still gated on validated identity. Figure6 remains TODO. No duplicate model experiments were launched.
+
+# Earlier142-slide master (preserved)
+
+**Previous142-slide master: all-cell-type pair budgets and approved method palette.**
+
+[Current editable PPTX](PBISC-Diffusion_expanded_master_20261013_celltypes_palette.pptx) · [Current PDF](PBISC-Diffusion_expanded_master_20261013_celltypes_palette.pdf)
+
+Pages **31–32** now show donor-fraction correlation and absolute error for T, B, Monocyte, NK, DC and Other. The historical global pair-budget summary remains on page **84**. All 140 parent pages are retained once. These curves use **n training + 7 validation pairs**; the genuine total-five-pair comparison remains separate on pages 33–34.
+
+The [approved palette](PBISC-Diffusion_color_palette_20261013.md) is applied throughout: PBISC blue, scSemiProfiler mint green, external comparators yellow; biological/cohort colours and quantitative heatmap scales keep their meaning.
+
+- [Current changes and interpretation](data/celltypes_palette_master_20261013/README.md)
+- [Current/parent/original page map](data/celltypes_palette_master_20261013/page_map.csv)
+- [Verification](data/celltypes_palette_master_20261013/qa/VERIFICATION.json)
+- Builder: `tools/build_celltypes_palette_master_20261013.py`; uses saved cell-type results and the validated semantic palette map.
+
+## Earlier 140-slide enriched master (preserved)
+
+The links and page ranges below describe the prior version.
+
+[Current editable PPTX](PBISC-Diffusion_expanded_master_20261013_enriched.pptx) · [Current PDF](PBISC-Diffusion_expanded_master_20261013_enriched.pdf)
+
+Figure 1 adds four recorded PCA denoising stages. Figure 2 now includes context/trunk and pred/oracle-hint diagrams, complete saved validation progressions, all-arm parameter/training-time comparisons, and a plain-language real-cell-repeat control. Figure 3 directly compares same-person bulk/pseudobulk expression values instead of donor-retrieval bars. The six author-defined scopes remain unchanged.
+
+- [Changes, page guide and scientific scope](data/figure_enrichment_20261013/README.md)
+- [Current/previous/original page map](data/figure_enrichment_20261013/page_map.csv)
+- [Verification](data/figure_enrichment_20261013/qa/VERIFICATION.json)
+- Builder: `tools/build_enriched_master_20261013.py`; consumes saved enrichment assets and the preserved 131-slide ordered spec. No training/inference launch is part of rebuilding this deck.
+
+## Earlier 131-slide ordered master (preserved)
+
+The following links and page ranges describe the previous ordered version.
+
+[Corrected editable PPTX](PBISC-Diffusion_expanded_master_20261013_reordered.pptx) · [Corrected PDF](PBISC-Diffusion_expanded_master_20261013_reordered.pdf)
+
+The actual PPTX/PDF now follow the [author-defined scope](data/figure_assignment_20261013/AUTHOR_SCOPE.md): Figure 1 overview/corpus (pages 5–9), Figure 2 SLE architecture/gene-context ablations (10–14), Figure 3 healthy paired mismatch → full FT → rank → paired budget (15–23), Figure 4 disease integrity (24–29), Figure 5 paired rare qualitative plan (30–31), Figure 6 published-finding reproduction plan (32–34). Supplementary begins at page 35. There is no separate main figure devoted only to healthy PBISC-versus-scSemi performance.
+
+- [Current order and correction report](data/figure_order_fixed_20261013/README.md)
+- [Previous 131-page IDs → current pages, with original 76-page IDs](data/figure_order_fixed_20261013/page_map.csv)
+- [Verification](data/figure_order_fixed_20261013/qa/VERIFICATION.json)
+- Current builder: `tools/reorder_master_figures_20261013.py`; consumes the verified earlier master spec and saved figures directly. Do not run the old expanded builder to update this order.
+
+All 131 source pages are accounted for. Original figure data are retained, obsolete planning cards are replaced, and the former availability-guide slot now shows a healthy-only extraction of the saved full-FT table. The mixed-cohort source plot remains in Supplementary. The original 76-slide files and prior 131-slide master remain unchanged. The [earlier classification CSV](data/figure_assignment_20261013/slide_assignment.csv) uses **previous** page numbers; use the current page map above.
+
+## Earlier 131-slide expansion (preserved)
+
+[Earlier PPTX](PBISC-Diffusion_expanded_master_20261013.pptx) · [Earlier PDF](PBISC-Diffusion_expanded_master_20261013.pdf)
+
+This is the comprehensive manuscript-working deck requested by the user, not a shortened talk. Every original slide is retained in its original relative order; new evidence is inserted next to the relevant topic. Original protocol wording is clarified where necessary.
+
+- [Original 76 → expanded page mapping](data/expanded_master_20261013/original_76_coverage.csv)
+- [All slide titles and provenance](data/expanded_master_20261013/slide_map.csv)
+- [Expansion report and evidence boundaries](data/expanded_master_20261013/REPORT.md)
+- [Verification](data/expanded_master_20261013/qa/VERIFICATION.json)
+- [Sources and immutable input hashes](data/expanded_master_20261013/manifest.json)
+
+The 55 additions comprise 20 healthy/input-control views, 25 disease/rare-state/external or exploratory views, six reconstructed review figures and four evidence/protocol guides. Existing saved results and coordinates are reused; no new training, inference or statistical fitting was launched. Historical scoring, diagnostic-only reference access, completed strict subsets and Figure 6 awaiting new data are labelled explicitly.
+
+**Palette:** PBISC muted yellow/gold #C5AA4B; scSemiProfiler blue #2166AC. Method variants use related shades, while biological/cohort colors retain their own meaning. The complete policy is `data/expanded_master_20261013/palette.json`.
+
+Builder: `tools/build_expanded_master_20261013.py`. CPU export/verification entry: `data/expanded_master_20261013/render.py` (use the host's bounded-memory execution policy). Plot caches are reused only after module, input and output hashes match.
+
+The previous [34-slide review](PBISC-Diffusion_results_review_20261013.pptx) is a historical subset; it is not the working master. The original 76-slide source deck and frozen archives remain unchanged.
+
+## Original 2026-09-30 r00 deck and historical change log
+
+76-slide English deck, ordered by the six main figures of the Nature-style manuscript plan (user request 2026-10-06,
+following `paper-release-20260928/reference-papers/Nature Style Manuscript Writing.pdf`):
+- Title and outline: 6 figures, Supplementary, Appendix, Discussion.
+- "Manuscript plan: six main figures": a table of each figure's claim and how many panels are ready.
+- For each figure:
+  - a divider;
+  - a plan slide, a table of panel / content / evidence slide / status (Ready, Adapt, Draft, Build, Run, Plan);
+  - the existing slides that serve its panels, each labelled "Fig. 2c" etc. in the top-right corner.
+- Supplementary: a plan slide plus the from-scratch results. Appendix: the modified scSemiProfiler run. Discussion: the
+  other-tissues outlook.
+
+User decisions of 2026-10-06:
+- No PBISC-VAE: scSemiProfiler, a VAE-based method, is the comparison (Fig. 3).
+- Fig. 2 shows LoRA first on the healthy cohort (ImMicro) with the realism metrics (precision, coverage, classifier
+  AUROC, MMD, E_N); the disease-cohort comparison comes later.
+- CYBERSORTx and other bulk deconvolution methods go to the Supplementary only (PBISC generates single cells).
+- Figs 4 and 6 stay unplanned. Their existing slides are kept in the section as "Fig. 4 candidate" / "Fig. 6 candidate".
+
+Only order, section labels, the agenda and new plan or draft slides changed. The new slides are "Overview draft (Nature
+style)", the manuscript's candidate Fig. 1, and "Pretraining corpus", the Fig. 1d strip. No new result figures were
+made.
+- The plan lives in `data/figure_plan.json`.
+- `tools/figure_plan.py` draws the plan tables and the slide text, and checks that every slide is placed exactly once.
+- `tools/restructure_figure_order_20261006.py` rebuilds the order from the plan. It is rerun after every plan change and
+  keeps all slide fields.
+- The earlier section order (10 sections, 45 slides) is in `_archive/2026-10-06_before_figure_order/`.
+
+Short keyword lines only; the figures carry the results. Every slide that uses a published method, metric or
+baseline has a one-line "References:" footer; full citations are in the speaker notes.
+
+- `PBISC-Diffusion_paper_summary_20260930.pptx`: the deck
+- `PBISC-Diffusion_paper_summary_20260930.pdf`: LibreOffice PDF of the deck
+- `data/slides_spec.json`: slide order, titles, keywords, footnotes, references, speaker-note text
+- `generate_deck.py`: builds the pptx from `data/slides_spec.json` only (python-pptx)
+- `figures/`: exactly the figures the slides show (PNG + PDF), made by `tools/sec*.py`. Intermediate panels go to a
+  temporary folder; superseded figures are deleted, not kept (2026-10-02, user request; every deleted file was
+  checked to be in the paper archive by sha256 first).
+- `data/sec*.csv` and `data/*_slide_text.json`: every plotted or quoted number with its source path
+- `render/`: PDF and 70-dpi slide JPGs for checking
+
+## Slide order
+
+| section | slides |
+|---|---|
+| 1 Model overview | bulk in, cells out; denoiser |
+| 2 Backbone | SDD → PBISC-Diffusion; bulk conditioning |
+| 3 Architecture ablation | width & context (loss, benchmark); gene panel |
+| 4 Metrics + SLE benchmark | **how we score generated cells** (metric table with references); SLE benchmark; per cell type; bulk broadcast baseline |
+| 5 Real bulk vs pseudobulk | input shift; same shift in three cohorts; cells off-manifold; disease direction flips |
+| 6 Paired fine-tuning | method; evaluation design |
+| 7 LoRA ablations | full vs LoRA vs cond.-encoder-only; LoRA rank r = 4 / 16 / 64; K diversification |
+| 8 Pairs needed | composition / coverage / steps (with scSemiProfiler at the same n); donor and disease signal; pairs from other cohorts; longer training on them (with training minutes) |
+| 9 Paired validation | cohorts; 5 donors: cells, realism beyond P/C, P/C at k = 3 / 5 / 10, scSemiProfiler copies its representative, disease signal, CAT curves, time to apply; 5-fold CV (with scSemiProfiler on the same pairs); same training pairs: LoRA vs scSemiProfiler; 5-fold realism metrics |
+| 10 Next steps | paired-subset cohorts (planned); other tissues from a few dozen pairs (Discussion framing of the from-scratch results) |
+| Supplementary | no pretraining: same cohort, from scratch (pilot); no pretraining: other cohorts, from scratch (pilot); rare states: from scratch vs pretrained; deconvolution with the same pairs: composition; deconvolution with the same pairs: disease signal |
+| Appendix | scSemiProfiler without its output threshold (sensitivity; modified read-out, not the official output) |
+
+## Rebuild
+
+Everything is CPU only; no GPU or checkpoint is touched.
+
+```bash
+# prerequisites (plan folder plans/paired_finetune_intradataset_20260927/, CPU):
+#   scsemi/tools/representative_dependence.py      -> scsemi/analysis/representative_dependence_20261002/
+#   tools/contrast_extra.py (+ tools/contrast_extra_tables.py) -> scores_contrast/   (Spearman, ceiling, limma, CAT, RBO)
+#   tools/build_tpm_zeroshot.py + tools/score_tpm_zeroshot.sh -> scores/<ds>/zeroshot_tpm_20261002/pooled/
+#     (the length-corrected zero-shot arm needs one GPU sampling step: tools/sample_lockwait.sh, coordinator only)
+cd /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00
+# shared helpers, not run directly: style.py, sec03_common.py, scsemi_results.py, fold_sizes.py
+for f in sec01_overview sec02_backbone sec03_width_loss sec03_width_bench sec03_panel \
+         sec04_metrics_overview sec04_sle_benchmark sec05_condition_space sec05_cells_offmanifold \
+         sec05_behcet_direction sec05_shift_cohorts sec06_finetune_method sec07_ft_modes \
+         sec07_k_aug sec07_rank_sweep sec08_pairs_needed sec08_cross_cohort sec09_paired_validation sec09_extra_metrics \
+         sec09_time_to_apply sec09_scsemi_representative sec09_cat_curves sec10_paired_subset_cohorts \
+         secA_scsemi_readout sec07_no_pretraining sec08_cross_scratch sec09_same_pairs \
+         secS_scratch_rare_states secS_deconvolution fig3_time_days fig4_disease fig23_healthy fig5_rare_states sec10_other_tissues figure_plan; do   # keep this order (sec10 reads secS; figure_plan last)
+  python3 "tools/$f.py" || break
+done
+python3 tools/restructure_figure_order_20261006.py   # order the slides by data/figure_plan.json
+python3 tools/sync_sec09_text.py      # copy every generated slide text (sec09_index / text_source) into the spec
+python3 generate_deck.py              # -> PBISC-Diffusion_paper_summary_20260930.pptx
+soffice --headless --convert-to pdf --outdir render PBISC-Diffusion_paper_summary_20260930.pptx
+pdftoppm -jpeg -r 70 render/PBISC-Diffusion_paper_summary_20260930.pdf render/slide
+cp render/PBISC-Diffusion_paper_summary_20260930.pdf .
+```
+
+`generate_deck.py` stops with an error if a figure is missing, a title does not fit on one line, a footnote needs
+more than 2 lines (1 line when the slide has references), the reference line needs more than 1 line, or any text
+contains Korean or placeholder words. `tools/restructure_20261002.py` is the one-off script that produced the current
+slide order and the references (it refuses to run twice).
+
+## Notes
+
+- **No "all" label (2026-10-02).** The 5-fold arm trains each fold model on the other four folds minus 7–8
+  inner-validation donors; its test fold is never trained on (RUN.json `test_rows_never_loaded_for_training`). Figures
+  label it "5-fold CV (other 4 folds)" and learning-curve ticks show the training-pair count per fold (78–79 ImMicro,
+  58–59 sepsis), read by `tools/fold_sizes.py`.
+- **Learning curve x = 0 uses no pairs (2026-10-02).** It shows the zero-shot model fed the real bulk as measured
+  and after exon-length correction (counts / Ensembl union-exon kb, then the same fixed-depth arithmetic;
+  `plans/.../tools/build_tpm_zeroshot.py`, scored in `scores/<ds>/zeroshot_tpm_20261002/pooled`). The old x = 0
+  calibration point, whose alpha used every training pair of the fold, is gone; calibration appears only fitted on the
+  same n pairs as the LoRA point (squares; `cal_budget_20261001`).
+- **Disease metrics (2026-10-02).** Next to Pearson logFC r and the Welch top-50 DEG overlap: Spearman r, the
+  split-half ceiling (observed vs observed, Spearman–Brown), limma-voom top-50 DEG, each arm's chance overlap, and
+  CAT curves with the k / 2G chance line (`plans/.../scores_contrast/`, tools/contrast_extra.py; every Pearson r and
+  Welch count reproduced exactly). Repeated samples of one patient (Imperial, Giroux, Ji) count as separate samples in
+  both DE tests, as in the scorer (footnote).
+- Section 9 text is generated: `tools/sec09_paired_validation.py`, `sec09_extra_metrics.py`, `sec07_rank_sweep.py`,
+  `sec09_time_to_apply.py`, `sec09_scsemi_representative.py`, `sec09_cat_curves.py` write slide text JSON; sync copies
+  it. Hand-written: the keywords of the two "Pairs needed" slides (section 8) and of every slide in sections 1–7 and 10.
+- Giroux's disease metrics exclude 196 sex-linked genes (severity is confounded with sex), from
+  `plans/.../scores_nosex/`. The calibration comparator is "Zero-shot + calibration" (user decision 2026-10-01).
+- Story rule (user decision 2026-09-30): no PBISC-VAE and no reference calibration (refcal). The exon-length-corrected
+  input is used only as the no-pair x = 0 point of the learning curve (user decision 2026-10-02).
+  `tools/sec04_immicro.py` (refcal vs scSemiProfiler on ImMicro) is kept but is not part of the deck or the rebuild.
+- scSemiProfiler 5-fold and equal-budget results are picked up automatically when they appear (`tools/scsemi_results.py`;
+  arm `scsemi__raw` in `scores/<cohort>/scsemi_cv_20261002/pooled*/`, `scsemi_n{5,10,20,40}__raw` in
+  `scores/<cohort>/scsemi_budget_20261002/pooled*/`; same donors and reference cells as the main pooled call only).
+  Then rerun sections 8 and 9, sync, recheck the section 8 keywords and rebuild.
+- Earlier versions: `_archive/2026-09-30_before_no_vae_refcal/`, `_archive/2026-10-01_before_extra_metrics/`,
+  `_archive/2026-10-02_before_time_to_apply/`, `_archive/2026-10-02_before_restructure/` (pptx, PDF, spec; their
+  figures are in the paper archive). `DRAFT_1150_*`: the copy shown at the 2026-09-30 12:00 meeting.
+
+- Added 2026-10-03 (user decision: the cross-cohort experiment is reported at K = 1000 only and is the figure that explains
+  why PBISC fits LoRA on a few pairs of each cohort): slides 25-26 in section 8, `tools/sec08_cross_cohort.py`.
+  - Slide 25 "Pairs from other cohorts do not transfer": per held-out cohort, composition L1 / coverage / disease logFC r of
+    zero-shot, calibration fitted on the other cohorts, LoRA r16 / r64 / full fine-tune trained on the other cohorts
+    (+ Liu 2021) at step saturation, and LoRA on the cohort's own pairs (5-fold pooled).
+  - Slide 26 "Longer training on other cohorts" (": gap stays" since 2026-10-06): the same metrics at every 20k-step
+    milestone.
+  - Source plan: `plans/paired_crosscohort_20261002/` (README; same-reference and same-sample checks in the script).
+    Agenda item 8 now reads "How many pairs, and do other cohorts' pairs transfer". The deck has 37 slides.
+    Pre-update copy: `_archive/2026-10-03_before_cross_cohort/`.
+
+## Sources
+
+Final model: `runs/f01_6k_1x_seed20260627` (epoch 16, EMA). Scores and receipts come from
+`plans/utility_studies_final6k_20260925/`, `plans/paired_finetune_intradataset_20260927/`,
+`plans/immicro_scsemi_comparison_20260927/`, `plans/floor01_campaign_20260923/`,
+`analysis/scsemi_benchmark_diffusion_20260913/` and the run folders named in each `data/sec*.csv`. Reference PDFs:
+`/home/sj_server_1/PBISC/paper-release-20260928/reference-papers/` (INDEX.md). Manuscript:
+`/home/sj_server_1/PBISC/paper-release-20260928/manuscript/`.
+- 2026-10-04 appendix (user decision: supplementary, appendix only, because it changes scSemiProfiler's official
+  output). Slide 38 is the divider "Appendix" and slide 39 is "Appendix: scSemiProfiler without output threshold"
+  (`tools/secA_scsemi_readout.py`, `data/secA_scsemi_readout{.csv,_slide_text.json}`).
+  - Inputs: `plans/paired_finetune_intradataset_20260927/scsemi/analysis/output_variants_20261003/`. The script checks
+    that its LoRA-5 and official scSemiProfiler values equal the section 9 slide data draw by draw.
+  - Result: without upstream's output threshold, scSemiProfiler's coverage rises in 5/5 cohorts, and 23/23 LoRA-5
+    comparisons (coverage, precision, composition, logFC r, DEG) keep their winner.
+  - The outline keeps its 10 sections (the builder requires 10). The pre-change copy is in
+    `_archive/2026-10-04_before_appendix/`. The deck has 39 slides.
+- 2026-10-05 update (user request: put the results so far into the deck). Pre-change copy: `_archive/2026-10-05_before_scratch_cv/`.
+  - New slide 23, "No pretraining: same cohort, from scratch" (`tools/sec07_no_pretraining.py`). Pilot on sepsis fold 0
+    (17 donors). Reads `fold0_scratch_20261005/` and the runs' RUN.json. The donor composition r is recomputed from
+    per_donor.csv and reproduces ablation_extra_donor.csv.
+  - New slide 28, "No pretraining: other cohorts, from scratch" (`tools/sec08_cross_scratch.py`). Held-out sepsis pilot.
+    Every arm is read from `paired_crosscohort_20261002/scores_masked/`: genes no training cohort measured are zeroed
+    in all arms, because the unmasked scratch arm puts 91–94 % of its UMI there (read from the generated_masked logs
+    since 2026-10-06).
+  - New slide 38, "Same training pairs: LoRA vs scSemiProfiler" (`tools/sec09_same_pairs.py`). scSemiProfiler 5-fold
+    CV with LoRA's own training pairs, plus GPU hours per fold from the receipts on either host. Cohorts that are still
+    running show "in progress"; rerun the script as they finish.
+  - These two slides now draw the scSemiProfiler CV / budget runs automatically:
+    - Slide 24, "Pairs needed" (diamonds; ImMicro so far): keyword 2 reworded, footnote extended.
+    - Slide 37, "5-fold CV" (yellow bar; ImMicro and Ji so far).
+  - `tools/sec09_paired_validation.py` reads the scSemiProfiler CV arm's Spearman / limma values from
+    `scores_contrast/<cohort>/scsemi_cv_20261002/pooled/` (contrast_extra.py --design scored). Pearson r and the Welch
+    DEG must match the scorer; the values stay blank until computed. Cohorts not yet run say "in progress".
+  - Agenda item 7 now reads "LoRA: compute, rank, K; training from scratch". The deck has 42 slides.
+  - New helper: `tools/grouped_bars.py`.
+- 2026-10-06 update (user request: check that the cross-cohort LoRA / full FT, cross-cohort from-scratch and
+  same-cohort from-scratch (+ recipes) results, with benchmarks and training times, and the scratch-vs-pretrained rare
+  states are in the deck; user decision: the from-scratch results go to the paper's Supplementary and the Discussion
+  presents them as a route to other tissues). Pre-change copy: `_archive/2026-10-06_before_supplementary/`.
+  - **Supplementary block (new divider, slides 40-43).**
+    - "No pretraining: same cohort, from scratch" and "No pretraining: other cohorts, from scratch" moved here from
+      sections 7 and 8. Their content is unchanged; slide 42's notes now name the cross-cohort slides by title.
+    - New "Rare states: from scratch vs pretrained" (`tools/secS_scratch_rare_states.py`).
+      - It reads the rare-state screen (`plans/rare_states_screen_20261004/out/sepsis_fold0_*` and
+        `sepsis_loco_masked_*`).
+      - The cross-cohort arms use the masked scoring (46 never-measured genes zeroed in every arm).
+      - The donor bootstrap reproduces the verified analysis in `out/scratch_vs_lora_20261005/REPORT.md`.
+  - **Section 10 "Next steps".**
+    - New "Next: other tissues from a few dozen pairs" (`tools/sec10_other_tissues.py`): a comparison table, pretrained
+      + LoRA vs from scratch.
+    - Every value is read from the deck's own source tables. The say text is the user's Discussion framing.
+  - **Slide 26 "Longer training on other cohorts".** The first keyword is now training minutes to saturation (RUN.json
+    wall seconds), with per-cohort minutes and steps in the notes (`tools/sec08_cross_cohort.py`).
+  - **Slide 34 "5 donors: time to apply".**
+    - The pretraining hours are read from `data/sec06_ft_method.csv`, no longer hard-coded.
+    - The note names the 'Fine-tuning with paired bulk + sc' slide by title. Its old "slide 15" pointed at the wrong
+      slide.
+  - Agenda items 7 and 10 updated. Spec change: `tools/restructure_20261006.py` (refuses to run twice). The deck has 45
+    slides; no Korean text.
+  - **Review pass (same day).** An independent 5-agent review of the changed slides confirmed 39 issues; all are
+    fixed. Pre-review copies of the touched scripts and the spec: `_archive/2026-10-06_before_supplementary/*.pre_review`.
+    - Slide 39: pretraining now reads 33.1 M train cells (`train_split_cells`; the 41.4 M corpus includes val/test).
+      The rare-state row follows the bootstrap verdicts on both sides (closer on ISG-high T / on DN2 B; labels and
+      states both arms miss by > 4-fold are not counted). "No corpus" became "come close": the scratch runs still took
+      the panel, value scaler and z-score constants from the PBMC checkpoint (RUN.json `kept_from_checkpoint`, now in
+      the notes). The say text carries the one-fold pilot qualifier; the footnote names both pilots.
+    - Slide 43: "no significant difference on average" (not "no difference"). Keyword 2 is the cycling T/NK count
+      (8 of 8 comparisons). The notes add the ISG-high T split by panel, the scratch-closer states (DN2 B, MS1), the
+      52-call tally, the fold-0 caveat (fold 0 is the own-pair LoRA's best fold for C1Q-high and cycling T/NK,
+      computed from `folds.csv`), label states unthinned, and "8 of 13 states shown".
+    - Slide 26: title "...: gap stays" and "drifts in Sepsis and Ji" are computed from the milestones (gap = no milestone
+      reaches the own-cohort coverage; drift = full fine-tune L1 up by >= 0.05). Minutes are in
+      `data/sec08_cross_cohort.csv` (wall_seconds, minutes, run_receipt) and in the numbers list.
+    - Slide 42: UMI share (94 %, 91–94 % across milestones), the observed share (0.014 %, from `cells_panel6k.npz`)
+      and both arms' stopping settings are read from the receipts; the "none" label sits inside the axes.
+    - Slide 34: keywords use the cohort means of the figure (17–59×, 57 min–4.1 h); the per-draw range 11–63× is in the
+      notes. The right panel shares the bars' rows and has a title.
+    - Talk text and pointers: slides 23, 25, 26, 45 (say), 22 (recipes description), 32 and 36 (slide pointers by
+      title), slide 1 subtitle "updated 2026-10-06".
+- 2026-10-08 update (user request: put the BayesPrism, CIBERSORTx and scSemiProfiler 5-fold results into the deck).
+  Pre-change copy: `_archive/2026-10-08_before_deconvolution/`.
+  - **Two new Supplementary slides** (`tools/secS_deconvolution.py`, `data/secS_deconvolution{.csv,_slide_text.json}`):
+    "Deconvolution with the same pairs: composition" and "...: disease signal".
+    - Source: `plans/bayesprism_paired_20261006/` (README, `results/RESULTS_full.md`, `RESULTS_cibersortx.md`).
+    - BayesPrism (recommended genes) and official CIBERSORTx (Fractions S-mode + HiRes) use the same paired donors
+      as the reference and the test samples' real bulk as the mixture, with the benchmark's own truth and scorer.
+    - Keywords and notes are computed from the tables.
+  - **Supplementary plan.** The planned "Bulk deconvolution methods" row is replaced by the two Ready slides.
+    `tools/figure_plan.py` now takes an optional per-slide note (`supplementary.notes`) and drops "Planned:" when
+    nothing is planned.
+  - **New scSemiProfiler results picked up (server_1):**
+    - Sepsis and Giroux 5-fold CV. Giroux was finalized on server_1 from sha-checked copies of its fold NPZs, and
+      rescored without sex-linked genes in `scores_nosex/giroux2022/scsemi_cv_20261002/pooled`.
+    - Sepsis equal-budget n = 5/10/20/40.
+    - Imperial 5-fold: finished on server_2 at 13:45 and was finalized on server_1 from sha-checked fold copies.
+      scSemiProfiler 5-fold now covers all five cohorts.
+    - Spearman/limma extras for sepsis, Giroux and Imperial; realism extras for all five cohorts.
+  - **Text changes from the new results:**
+    - "Same training pairs": keywords are data-driven. LoRA leads on coverage and precision in 5/5 cohorts;
+      composition 4/5 and logFC r 3/4, where scSemiProfiler leads in Giroux. The GPU-time ratio is 19–62×.
+    - sec09_same_pairs.py now reads the sepsis fold-0 pilot queue tag (`sep_f0`). For Giroux it quotes DEG from
+      scores_nosex, and for older pooled LoRA calls from pooled_deg.
+    - "5-fold CV" keyword: logFC r 3/4.
+    - "Pairs needed" (hand-written): scSemiProfiler coverage 0.04 / 0.10. On sepsis disease logFC r,
+      scSemiProfiler at 5–40 pairs (0.52–0.58) is above LoRA (0.29–0.48), while its top-50 DEG stays at 2–6.
+- 2026-10-08, new figure split (user decision; supersedes the 2026-10-06 split; pre-change copy
+  `_archive/2026-10-08_before_new_figure_plan/`).
+  - Fig. 2: LoRA fine-tuning with no disease metrics.
+  - Fig. 3: PBISC vs scSemiProfiler on healthy donors, with time as a main result.
+  - Fig. 4: disease benchmarks on the paired cohorts (PBISC versions, scSemiProfiler, and BayesPrism / CIBERSORTx
+    labelled as cell-type deconvolution; the deconvolution slides moved here from the Supplementary).
+  - Fig. 5: rare-cell fractions and detection.
+  - Fig. 6: new findings from the bulk-only sepsis and COVID samples (K = 1,000 and 4,000).
+  - The cross-cohort transfer slides moved to the Supplementary; their placement is open.
+  - New slide "Time: days for scSemiProfiler, minutes for LoRA" (`tools/fig3_time_days.py`, Fig. 3d).
+    - Inputs: GPU hours per 5-fold experiment from the receipts (`data/sec09_same_pairs.csv`, checked against
+      the scSemiProfiler event logs on both hosts), the 5-donor draws (`data/sec09_time_to_apply.csv`) and the
+      one-time pretraining (`data/sec06_ft_method.csv`).
+    - Results: ImMicro 24 h vs 1.2 h, sepsis 38 h vs 37 min. All runs: scSemiProfiler 184 h (7.7 days) vs PBISC
+      22 h including pretraining.
+    - The equal-budget scSemiProfiler runs are excluded: they reuse the 5-fold run's representative models.
+  - Panel status: Build / Run marks new slides or analyses still to do; Adapt marks slides to re-scope (healthy-only
+    Fig. 2-3, disease columns into Fig. 4).
+- 2026-10-08, comparison rules (user decision; memory feedback_paper_comparison_rules):
+  - **Time.** Only per-cohort GPU work counts: no PBISC pretraining, no GPU queue waits. The time slide now shows the
+    5-donor draws (main) next to the 5-fold runs (separate). All runs: 7.7 GPU-days vs 5.7 h.
+  - **No bulk-broadcast baseline.** It is removed from the SLE benchmark, CAT, deconvolution and "Pairs needed"
+    figures.
+    - The "Bulk broadcast baseline" slide is gone. Its partial r panel is now on the SLE benchmark slide
+      (0.21 vs 0.17).
+    - Still to rebuild in their own phases: Fig. 2e (ImMicro only) and Fig. 6a.
+  - **Main comparison = 5 pairs** (LoRA-5 vs scSemiProfiler-5); 5-fold is reported separately. Partial r is
+    emphasised (bold panel titles).
+  - **New `tools/fig4_disease.py`**, which checks all 200 logFC r / DEG values against the section-9 tables:
+    - "Disease signal with 5 pairs: LoRA-5 vs others" (4b);
+    - "5-fold CV: disease signal" (4c);
+    - "Sepsis: fine-tuning variants" (details).
+  - The old "5 donors: LoRA vs scSemiProfiler — disease signal" slide is retired. `tools/sync_sec09_text.py` lists it
+    in RETIRED.
+  - The deconvolution slides are retitled "Cell-type deconvolution, same pairs: ...".
+- 2026-10-08, healthy-only Figs 2-3 (`tools/fig23_healthy.py`).
+  - Eight ImMicro-only slides, re-plotted from the deck's own data tables; nothing is re-scored.
+  - Fig. 2 (no disease metrics):
+    - cells, realism, full / LoRA / condition encoder, pairs needed, rank and K;
+    - 5 donors (main) shown next to 5-fold (separate).
+  - Fig. 3 (vs scSemiProfiler):
+    - LoRA-5 vs scSemiProfiler-5 on 6 cell metrics: LoRA ahead on 6 of 6;
+    - the same pairs, 5 to all;
+    - scSemiProfiler copies its representative;
+    - time.
+  - The earlier all-cohort slides moved to the Supplementary (note "all five cohorts").
+  - "5-fold CV: train on the other 4 folds" is retired (it carried the broadcast; its content is now in Fig. 4c and
+    the healthy slides). It is listed in RETIRED of `tools/sync_sec09_text.py`.
+- 2026-10-08, Fig. 5 rare cells (`tools/fig5_rare_states.py`).
+  - Source: new plan `plans/rare_states_paired_20261008` (README). It covers 4 paired disease cohorts, the 5-donor
+    draws and 5-fold, and the arms zero-shot, calibration, scSemiProfiler and LoRA. Each NPZ is sha-checked and the
+    benchmark's Poisson read-out and labels are reproduced cell by cell.
+  - Three slides:
+    - fraction (5a);
+    - donors and disease direction (5b);
+    - 5-fold, separate (5c).
+  - Results over 56 state × cohort pairs, 5 donors:
+    - LoRA-5 closer to the observed fraction than scSemiProfiler-5 in 51/56;
+    - within 2-fold: 19 vs 5;
+    - better donor ranking: 43/56;
+    - disease direction kept: 73% vs 61%.
+  - Fig. 6a no longer mentions the broadcast. No slide text mentions it now.
+- 2026-10-10, last broadcast mention removed from a figure (found while rewriting the manuscript from this deck).
+  - Fig. 2d "How we score generated cells" (`tools/sec04_metrics_overview.py`) still listed "broadcast; split-half
+    ceiling" as the comparator of logFC r inside the table image. It now reads "zero-shot; partial r (vs bulk)".
+  - Old tool, CSV and figure: `_archive/2026-10-10_metrics_overview_broadcast/`. The deck was rebuilt (76 slides); a
+    text search of every slide, the notes and the PDF twin of every figure a slide uses finds no "broadcast". Three
+    figure files of retired slides still contain it and are not used: `sec04_sle_broadcast`, `sec09_fewpair_disease`,
+    `sec09_fivefold`.
+  - The manuscript was rewritten from this deck on 2026-10-10 (`paper-release-20260928/manuscript/`,
+    `editorial/rewrite_20261010.md`); its figures are copies of this folder's figure PDFs.
+- 2026-10-10, LoRA rank figure key (`tools/sec07_rank_sweep.py`): "LoRA rank r = 16 (used in the deck)" now reads
+  "(used throughout)", because the same panel is Supplementary Fig. 12a of the manuscript. Old files:
+  `_archive/2026-10-10_rank_sweep_label/`. Deck rebuilt (76 slides).

paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/figure_assignment_20261013/STRENGTHENING_PROPOSAL.md

@@ -0,0 +1,154 @@
+# Figure 1–6 / Supplementary: 근거 보강 계획과 반영 결과
+
+2026-10-13 사용자 수정 지시를 반영했다. Figure 번호는 [AUTHOR_SCOPE.md](AUTHOR_SCOPE.md)를 따른다. 현재 덱의 쪽수는 [새 페이지 대응표](../strengthening_20261013/page_map.csv), 변경·검수 결과는 [작업 보고서](../strengthening_20261013/README.md)를 기준으로 한다. 이전 논의용 제안서는 [수정 전 사본](../strengthening_20261013/STRENGTHENING_PROPOSAL_before_revision.md)에 보존했다.
+
+이번 작업은 저장된 결과 재시각화, 분할·원 논문 검증, observed marker의 CPU 분석이다. 학습·생성·deconvolution을 새로 실행하지 않았다. 기존 campaign의 미완료 단위와 새로운 분석을 구분한다. 과거 13:58 KST 상태를 현재 실행 상태로 인용하지 않는다.
+
+## 우선 결정
+
+| 구분 | 이번 반영 | 남은 핵심 작업 | 새 모델 학습 필요 |
+|---|---|---|---|
+| Figure 1 | Healthy / Disease·at-risk → 상세 condition, cell type; corpus split와 K 모식도 | 사용 경계·donor 교차표의 최종 캡션 확인 | 없음 |
+| Figure 2 | 기존 architecture/hint 도식, validation progression, 비용·parameter 비교 유지 | 동일 family 안에서 주 패널 선정; 미완료 run의 saturation 과장 금지 | 기본적으로 없음 |
+| Figure 3 | 같은 107명 observed/raw/calibration/full FT/LoRA 직접 비교; 실제 split 도식 | 기존 strict total-support campaign 완료; 공통 cell-space 정성 패널은 저장 출력에서 후속 조립 | 중복 실행 없음 |
+| Figure 4 | 누락 scSemi CV CAT/RBO 복원, 다중 지표 비교, strict Sepsis donor-aware DEG | 나머지 cohort의 기존 strict 분석 완료, 동일 donor를 보존하는 CI | 중복 실행 없음 |
+| Figure 5 | 원 논문 target 대조; Ye paired observed marker 진단 | 원 저자 cluster identity 연결 후 observed/PBISC/scSemi 비교 | 먼저 CPU 분석 |
+| Figure 6 | 기존 논문 결과 재현 계획 유지 | endpoint·cohort 고정 후 기존 bulk-only 출력 분석 | 기본적으로 없음 |
+
+지금 가장 먼저 보강할 일은 Figure 5의 정확한 population identity, Figure 4의 donor-aware 비교 완성이다. 그림 수나 유리한 지표의 수를 늘리는 것을 종료 기준으로 삼지 않는다.
+
+## Figure 1 — condition 비율과 corpus → recipe의 관계
+
+### 완료: 두 단계 condition 도표
+
+총 **41,395,184개 source cells**를 기준으로 Healthy **36,639,276개, 88.51%**, Disease / at-risk **4,755,908개, 11.49%**를 나눈다. 가운데 도넛에서 후자를 11개 원래 condition으로 세분하고 cell-type 도넛을 함께 둔다.
+
+- ACPA+ at-risk는 **1,792,129개**로, 확진 RA가 아니다. 이 때문에 상위 범주를 단순 Disease로 쓰지 않는다.
+- At-risk를 제외한 disease/history annotation은 **2,963,779개, 전체의 7.16%**다. Post-COVID와 Cytopenia도 원래 annotation 의미를 유지한다.
+- Healthy는 donor 상태다. In-vitro 자극, post-vaccination, CMV-seropositive healthy를 모두 resting untreated cell로 해석하지 않는다.
+- **88.5%는 donor 비율이 아니라 cell 비율**이다. Condition 라벨은 corpus 설명용이며 최종 base model의 입력 label이 아니다.
+
+### 완료: 41.4M와 33.1M 차이는 split
+
+| Source allocation | Cells | 전체 중 비율 |
+|---|---:|---:|
+| Train | 33,141,380 | 80.06% |
+| Validation | 4,361,032 | 10.54% |
+| Test | 3,892,772 | 9.40% |
+
+따라서 약 **33.4M가 아니라 33.1M**이며, 차이 8,253,804개는 validation+test다. K recipe 때문에 제외된 세포 수가 아니다. 실제 group 크기가 달라 nominal 80/10/10이 cell 비율로 정확히 일치하지 않는다.
+
+### 완료: K와 target exposure 모식도
+
+`train sample에서 서로 다른 K cells 선택 → counts 합으로 pseudobulk condition 생성 → 같은 condition으로 각 member cell을 target으로 학습`한다. Registered K는 250/500/1,000/2,000/4,000, schedule part당 8,192 recipes이며 16 parts를 사용한다.
+
+Regular grid로 31,145,450개 고유 train cells를 덮은 뒤, **3,658개 top-up recipes**로 남은 1,995,930개를 포함한다. Final contract는 **33,141,380개 train cells 전부의 coverage**를 확인한다. 총 target occurrences 195,639,930회는 반복 노출을 포함하므로 고유 세포 수와 다르다. Top-up의 실제 K는 1–4,000이다.
+
+`unique source cells`, `recipe 개수`, `K`, `target occurrences`, `N_generated`를 서로 바꿔 쓰지 않는다. Source condition 비율을 sampler exposure 비율로 해석하지 않는다.
+
+**보충자료:** 내부 corpus split에는 sample 중복은 없지만 biological donor 27명이 split을 가로지른 기록이 있다. 외부 SLE/paired test 누출을 입증하는 결과와 구분한다. AIFI reference의 pretraining 노출도 target donor 중복과 별개로 기록한다.
+
+**재사용/산출물:** [corpus bundle](../strengthening_20261013/corpus/result.json), `manuscript/source_data/corpus/`, final run `ALLCELL_DATA_CONTRACT.json`. 새 학습 없음.
+
+## Figure 2 — architecture / gene context에 집중
+
+기존에 추가한 context/trunk, pred/oracle hint 경로 도식과 full/late validation curves, parameter/training-time 패널을 유지한다. 이 Figure에 real-bulk mismatch·paired adaptation을 넣지 않는다.
+
+**주 그림 조합:** 간략한 모델 구조 → width/context family의 validation progression → SLE pseudobulk 성능 → gene-panel family → parameter/time trade-off. Saved corpus-validation loss는 SLE held-out benchmark metric과 다른 자료임을 표시한다.
+
+**실제로 남는 보강:** 같은 평가 조건의 arm만 비교하고, 제한된 학습 기록을 saturation으로 부르지 않는다. 16× run은 E2까지만 있어 수렴 여부가 미확인이다. Architecture 1% detection floor와 gene-panel 0.1% floor의 차이도 보존한다. 6k의 5 seeds는 sampling 반복이며 독립 pretraining 5회가 아니다.
+
+**추가 실험 조건:** 특정 architecture의 안정적 우위를 논문의 필수 주장으로 삼는 경우에만 가장 가까운 경쟁 checkpoint의 동일 sampling 반복을 검토한다. 현재 시각화 보강 때문에 architecture 전체를 재학습할 필요는 없다.
+
+## Figure 3 — 입력 mismatch와 생성 결과 회복을 연결
+
+### 완료: 같은 사람을 나란히 보여주는 직접 비교
+
+Observed paired scRNA의 cell-type mixture와 raw zero-shot / input calibration / full FT / LoRA r16을 **동일 107명, 동일 순서**로 나란히 그렸다. Donor 순서는 observed monocyte 비율만으로 정했고, 모든 donor와 6개 type을 남겼다. Full FT/LoRA와 raw zero-shot은 같은 raw measured-bulk 입력을 사용한다.
+
+같은 사람을 잇는 보조 패널은 composition L1과 **generated-vs-observed pseudobulk gene-expression Pearson r**를 함께 보여준다. 후자는 같은 5,508 genes의 log1p-CPM에 대한 값이며 cell별 일대일 일치도가 아니다.
+
+- Mean composition L1: raw **0.4549**, calibration **0.1985**, full FT **0.1572**, LoRA r16 **0.1617**.
+- Median expression r: raw **0.8921**, calibration **0.8802**, full FT **0.9141**, LoRA r16 **0.8994**.
+
+즉 output composition 회복은 분명하지만 모든 지표에서 LoRA가 가장 좋거나 모든 expression mismatch가 사라진다고 쓰지 않는다. Adaptation은 measured-bulk 값을 직접 바꾸는 연산이 아니다. 기존 input-calibration density/error 패널과 생성 output 패널의 역할을 구분한다.
+
+### 확인: LoRA all / LoRA20의 실제 test
+
+기존 pair-budget 그림은 **고정된 5-fold outer test**를 사용한다.
+
+| 표기 | Training | Validation | Test/fold | 사용하지 않는 나머지 |
+|---|---:|---:|---:|---:|
+| LoRA5 | 5 | 7 | 21–22 | 73–74 |
+| LoRA20 | 20 | 7 | 21–22 | 58–59 |
+| LoRA all | 78–79 | 7 | 21–22 | 0 |
+
+“All”은 **각 fold에서 사용할 수 있는 training pool 전체**다. 107명 전체를 학습해 같은 사람에게 test하지 않는다. 각 donor의 held-out prediction을 모아 pooled 107명 결과를 만든다. LoRA20의 나머지 87명이 모두 test가 되는 설계도 아니다. Test를 고정해야 support 양에 따른 차이를 같은 사람들에서 비교할 수 있다.
+
+Outer/inner 분할 seed는 **20260927**이다. Training subset은 fold별 **1000–1004** seed의 random permutation에서 prefix를 취한다. 따라서 `5 ⊂ 10 ⊂ 20 ⊂ 40 ⊂ all`인 **random nested selection**이다. 저장된 20개 subset과 outer/inner 분할을 replay해 정확히 일치함을 확인했다.
+
+**프로토콜 구분:** 기존 별도 total-five 5 draws는 5 support+0 validation, 나머지 102명 test/draw다. 새 strict healthy campaign은 5/10/20/40 total support와 고정 outer-test fold, 추가 validation 없음이다. 두 total-five 설계도 같은 실험으로 합치지 않는다.
+
+**남은 작업:** 기존 strict curve의 미완료 산출물을 마무리한다. 추가 정성 패널이 필요하면 저장된 observed/raw/LoRA cells를 observed-only fit의 동일 projection에 배치하고 동일 test·depth·N_gen을 유지한다. 이는 CPU figure 작업을 우선하며 새로운 학습 근거가 아니다. r4/r16/r64/full FT는 지표별 절충으로 보고 r16 보편 우위를 만들기 위한 합성점수는 사용하지 않는다.
+
+**근거:** [healthy bundle](../strengthening_20261013/healthy/result.json), [split audit](../strengthening_20261013/healthy/split_audit.csv).
+
+## Figure 4 — reference가 있는 여러 평가를 보강
+
+### 완료: 기존 결과에서 추가한 세 패널
+
+1. **Complete historical CV-CAT:** 누락됐던 네 cohort의 scSemi CV 결과를 포함한다. Raw/calibration/LoRA/scSemi/bulk-broadcast, 저장 결과가 있는 Sepsis full FT를 같은 조건에서 비교한다.
+2. **LogFC r / partial r / RBO:** observed paired scRNA를 reference로 사용한다. Same-patient cell-repeat r는 내부 cell resampling control이며 독립 환자 generalization ceiling이 아니다.
+3. **Strict five-support Sepsis:** 완료된 5 draws��� logFC r, partial r, CAT100, donor-aware DEG overlap을 함께 제시한다. 점은 overlapping support draws, 큰 점은 descriptive mean이다. Draw를 독립 환자로 간주한 CI를 붙이지 않는다.
+
+CAT는 [Irizarry et al., 2005](https://doi.org/10.1038/nmeth756), RBO는 [Webber et al., 2010](https://doi.org/10.1145/1852102.1852106)를 근거로 한다. Same-sign CAT는 현재 benchmark에서 방향 일치를 추가한 정의다. Donor-aware DEG의 필요성은 [Squair et al., 2021](https://www.nature.com/articles/s41467-021-25960-2)에 근거를 둔다. 구체적 구현·유전자 축·reference·출처는 [metric bundle](../strengthening_20261013/disease_metrics/result.json)에 묶었다.
+
+**주장 범위:** Strict Sepsis에서 LoRA/scSemi의 평균 logFC r는 **0.3421/0.4025**, CAT100은 **0.0768/0.0328**, donor-aware DEG overlap은 **17.6/11.4**다. 순위·DEG 회복에서의 장점과 전체 logFC correlation의 반례가 동시에 있다. ‘모든 질환·모든 지표에서 최고’라는 결론은 근거와 맞지 않는다.
+
+### 남은 필수 보강
+
+- 나머지 질환의 기존 strict outputs와 patient-aware statistics를 끝까지 연결한다. Sepsis 기존 deconvolution adoption을 다시 실행하지 않는다.
+- 같은 donor가 여러 visit/draw에 나타나는 구조를 보존해 **방법 간 차이의 donor CI**를 계산한다. 고정된 생성 결과의 CI는 재학습 불확실성까지 포함하지 않는다.
+- Primary endpoint와 보조 endpoint를 고정한다. LogFC r, partial r, CAT/RBO, DEG가 서로 다른 질문에 답한다는 점을 설명한다.
+- Cohort/cell type별 NA, 불리한 결과, support/HVG/test 접근 범위를 Supplementary에 보존한다. Giroux의 sex/severity confounding은 sex gene 제거만으로 없어지지 않는다.
+
+새 지표를 계속 찾거나 원하는 순위가 나올 때까지 실험을 반복하기보다, 이미 있는 reference-backed 지표들의 비교 조건과 불확실성을 완성한다.
+
+## Figure 5 — 각 paired 원 논문의 population을 먼저 검증
+
+원 논문에서 직접 보고한 population과 일반적인 marker rule을 구분한다. 네 cohort에 억지로 rare population 하나씩을 배정하지 않는다. 상세 원문/figure/marker/시점/파일 가용성은 [문헌·observed 분석 bundle](../strengthening_20261013/rare_literature/result.json)을 따른다.
+
+| Cohort | 원 논문 기반 우선 대상 | 다음 단계 |
+|---|---|---|
+| Ye sepsis | Fig.5 **Mono_c06-CD14-MKI67**, pediatric pulmonary sepsis; MKI67/STMN1/MPO/ELANE | 원 저자의 fine-cluster identity와 감염 부위·연령 대조를 연결 |
+| Imperial ESKD | Fig.5 **dexamethasone-associated monocytes**, wave2 | Steroid 투약 시점·wave2·세부 cluster를 매칭; wave1 음성 대조 |
+| Giroux COVID | IgG 전환 전 CD14 monocyte clusters 2/4(mild), 7(moderate), RNA/regulon states | Fine labels·serostatus 연결; IRF1/KLF3는 저장된 6k panel 밖이므로 exact rare identity 주장 보류 |
+| Ji CHILI | Cell-type DE/pathway 중심 | 기존 DE/pathway endpoint는 Figure4에 적합; rare target 근거가 있어야 Figure5에 채택 |
+
+### 이번 observed-truth 진단에서 배운 점
+
+Ye의 identity-passed **82명, frozen AIFI monocyte 111,269개**에서 CD14/MKI67/STMN1/MPO/ELANE만 CPU로 추출했다. MKI67+STMN1 native-depth 동시검출의 donor 평균은 소아 폐렴군 **2.33%**, 소아 복부감염군 **4.10%**, 성인 폐렴군 **0.39%**다.
+
+이 결과는 cycling-marker 신호가 실제 paired sc에 있다는 진단이지 **원 논문의 Mono_c06 정체성 또는 폐렴 특이성 재현 완료가 아니다**. 검출률은 depth와 dropout에도 영향을 받는다. 73-marker cache의 일반 rule로 원 논문의 cluster를 대체하면 잘못된 결론을 만들 수 있다.
+
+### 실행 순서와 완료 기준
+
+1. **로컬 재사용 조사는 완료:** Ye의 canonical H5AD obs schema 133개, raw archive member 665개, benchmark metadata를 조사했으나 저자 fine-cluster labels를 찾지 못했다. [조사 목록](../strengthening_20261013/rare_literature/ye_author_label_inventory.json)을 보존했다. 다음은 원 논문의 공개 annotation/분석 코드에서 author_cell_id와 연결되는 라벨을 찾는 일이다. 확보할 수 없을 때만 원 논문 marker 기반 reference-only classifier/rule을 정의한다.
+2. **Observed paired sc에서 먼저** 연령·감염 부위 또는 steroid timing에 따른 population identity/abundance를 재현한다. 이 단계가 안 맞으면 generated 결과를 성공 사례로 제시하지 않는다.
+3. 같은 held-out samples의 observed/PBISC/scSemi를 동일 projection·측정 gene·depth/N_gen 조건에서 비교한다. Original fine labels로 확인한 parent population과 rare cluster를 분리한다.
+4. 정의·projection에 사용하지 않은 marker/program으로 identity를 추가 검증한다. 모델 panel에 없는 marker를 zero로 채우지 않는다.
+5. 성공·과소생성·관측 음성 사례와 donor별 전체 결과를 함께 둔다. Observed zero를 진정한 부재로 단정하지 않는다.
+
+기존 paired K1000/K4000은 모두 **1,500 generated cells/sample**이다. K sensitivity와 generated-cell count sensitivity를 혼동하지 않는다. 지금 필요한 작업은 저장 자료의 정확한 annotation·CPU 재분석이며, 새 GPU 학습이 기본 선행 조건은 아니다.
+
+## Figure 6 — bulk-only 논문 결과 재현은 별도 TODO
+
+추천 후보는 [Stephenson et al., 2025](https://pmc.ncbi.nlm.nih.gov/articles/PMC12366655/) Fig.3C의 monocyte HLA-II 감소다. Sample-time severity와 peak severity를 구분해 endpoint를 고정한 뒤, 기존 Imperial bulk-only 출력에서 donor-aware 효과를 계산한다.
+
+Paired donor를 모두 제외한 후보 집합은 기존 audit상 97 samples/45 donors, wave1이다. 이 집합으로 wave2 steroid-associated population을 재현할 수는 없다. Figure5의 **paired wave2** 가능성과 Figure6의 **donor-disjoint bulk-only wave1**을 혼동하지 않는다. HLA-II 방향은 기존 workflow에서도 살펴본 후향적 재현이며 blind independent validation이 아니다. 이번 작업에서 Figure6 분석은 실행하지 않았다.
+
+## Supplementary와 추가 GPU의 경계
+
+Supplementary에는 전체 cohort/type/draw, 부정적 결과, support·validation·HVG 접근량, source/exposure 구분, rank/full FT 절충, 평가법·readout 감도, rare marker/annotation 근거와 음성 사례를 둔다. Main Figure2는 SLE pseudobulk, Figure3–5는 measured paired라는 범위를 유지한다.
+
+새 GPU 작업은 (a) 필수적인 architecture/training-seed 안정성 주장에 근거가 부족하거나, (b) adapted model의 conditioning 기전을 확인하는 matched-noise wrong-bulk intervention이 꼭 필요할 때에만 별도로 정의한다. 두 항목 모두 현재 기본 필수 과제는 아니다. 실행 전 reuse registry와 checkpoint/support/test/mask/seed/readout 계약을 대조하고 바뀐 연산만 수행한다.

paper-releases/pbisc-diffusion/updates/20261013T115045Z-figure-strengthening-154/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/strengthening_20261013/README.md

@@ -0,0 +1,36 @@
+# Figure strengthening — 2026-10-13
+
+Current evidence master: **154 slides**, preserving every parent142 page and every original76 page once. The older decks and their scientific figure bytes remain unchanged. Twelve new pages use saved data and observed-marker CPU analysis. No model training, generation, GPU scoring or deconvolution was launched.
+
+[Editable PPTX](../../PBISC-Diffusion_expanded_master_20261013_strengthened.pptx) · [PDF](../../PBISC-Diffusion_expanded_master_20261013_strengthened.pdf) · [Revised strengthening proposal](../figure_assignment_20261013/STRENGTHENING_PROPOSAL.md)
+
+| Pages | Scope | New material |
+|---|---|---|
+| 8–10 | Figure1 | Healthy/disease-at-risk detail and cell types; exact train/validation/test allocation; K recipe coverage and repeated exposure |
+| 13–25 | Figure2 | Existing structure/hint, validation, common-axis SLE and cost panels retained |
+| 31–32 | Figure3 | Same107 held-out donors: observed/raw/calibration/fullFT/LoRA mixtures and expression agreement |
+| 35 | Figure3 | Fixed test fold, random nested training support, validation and unused donor roles |
+| 36–37 | Figure3 | Previous all-six-cell-type pair-budget curves retained |
+| 46–48 | Figure4 | Complete CV CAT; correlation/partialr/RBO; strict Sepsis donor-aware DEG across5 support draws |
+| 51–53 | Figure5 | Four original-paper targets/boundaries and observed Ye marker diagnostic |
+
+Original coarse corpus overview moves from parent142 page8 to current page59; the revised three-donut panel is in main Figure1. Full section and parent/original mapping: [page_map.csv](page_map.csv).
+
+## Scientific interpretation
+
+- Total source41,395,184 = train33,141,380 +validation4,361,032 +test3,892,772. K is not the reason for the difference. The final contract covers every training cell; repeated target occurrences are not unique cells.
+- Healthy88.51% is a source-cell share, not a donor share. The11.49% remainder includes ACPA+ at-risk; it is not all confirmed disease.
+- Historical LoRAall trains78–79 and validates7 per fold, keeping21–22test. LoRA20 keeps the same test and validation;58–59 remaining pool donors are unused. Subsets are random and nested (seeds1000–1004). Outer/inner seed20260927 and all20 subset memberships replay exactly.
+- The legacy five-draw healthy total-five design tests102 remaining donors/draw. The newer strict healthy campaign keeps outer-fold test sets. These are distinct protocols.
+- Adaptation improves generated output composition. It does not alter measured bulk values, and LoRA is not best on every metric. FullFT has the lowest mean composition L1 here.
+- Disease metrics remain endpoint/cohort dependent. Strict Sepsis favours scSemi for mean logFCr but LoRA for CAT100 and donor-aware DEG overlap. Draw points are not independent biological replicates or CIs.
+- Ye observed111,269 monocytes/82donors contain cycling-marker signal. Simple MKI67+STMN1 detection does not reproduce pulmonary specificity (pediatric pulmonary2.33%, abdominal4.10%). All133 canonical obs schemas were inspected; fine author labels were not found. This is an observed diagnostic, not completed population recovery.
+- Figure5 generated rare-identity comparisons and Figure6 bulk-only endpoint reproduction remain explicitly pending. The next priority is source-defined identity/metadata alignment and existing donor-aware analyses, not duplicate training.
+
+## Reproducibility and verification
+
+Each family has `result.json`, source hashes, machine-readable plotted tables, and PNG/PDF/SVG assets. Builders are `tools/strengthen_*_20261013.py`; master builder is `tools/build_strengthened_master_20261013.py`. CPU rendering uses [render.py](render.py).
+
+[Verification record](qa/VERIFICATION.json) checks154 PPTX/PDF/spec pages, all142 parent and76 original IDs, source/protected hashes, exact embedded scientific pictures without cropping, Arial fonts and text/shape bounds. All13 contact sheets and the12 new panels receive visual review. Source CSV arithmetic, corpus totals, outer/inner and nested splits, method comparability and streamed marker extraction have separate family-level checks.
+
+The before-revision proposal is frozen as background provenance. The live revised proposal is the current plan; it is not substituted for old source evidence.