Archive 4 checkpoints (1-4)

.gitattributes

@@ -92,3 +92,5 @@ paper-releases/pbisc-diffusion/updates/20261005T175437Z-deck-supplementary-other
 paper-releases/pbisc-diffusion/updates/20261006T121425Z-fig1-architecture-plus-corpus/diffusion/dataset/server_1/manuscript/main.pdf filter=lfs diff=lfs merge=lfs -text
 paper-releases/pbisc-diffusion/updates/20261006T125143Z-supp-fig-s1-corpus/diffusion/dataset/server_1/manuscript/main.pdf filter=lfs diff=lfs merge=lfs -text
 paper-releases/pbisc-diffusion/updates/20261006T125143Z-supp-fig-s1-corpus/diffusion/dataset/server_1/manuscript/supplementary.pdf filter=lfs diff=lfs merge=lfs -text
+paper-releases/pbisc-diffusion/updates/20261006T145850Z-deck-six-figure-order/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_paper_summary_20260930.pdf filter=lfs diff=lfs merge=lfs -text
+paper-releases/pbisc-diffusion/updates/20261006T145850Z-deck-six-figure-order/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_paper_summary_20260930.pptx filter=lfs diff=lfs merge=lfs -text

paper-releases/pbisc-diffusion/updates/20261006T145850Z-deck-six-figure-order/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_paper_summary_20260930.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c990d1b7a4562b53e03591cd9ea52276789289a7e61cb5ae0a622f995eddb263
+size 8740044

paper-releases/pbisc-diffusion/updates/20261006T145850Z-deck-six-figure-order/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/PBISC-Diffusion_paper_summary_20260930.pptx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99ed7675ef69913f55b9bad7c0d78457fddd8b4b17ba505424df21a5bc6ea119
+size 9798501

paper-releases/pbisc-diffusion/updates/20261006T145850Z-deck-six-figure-order/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/README.md

@@ -0,0 +1,205 @@
+# PBISC-Diffusion paper summary deck (2026-09-30, r00; ordered by the six manuscript figures since 2026-10-06)
+
+60-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.
+- "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, Decide);
+  - the existing slides that serve its panels, each labelled "Fig. 2c" etc. in the top-right corner.
+- The Supplementary block (from-scratch results) and the Appendix block (modified scSemiProfiler run), unchanged.
+
+Only order, section labels, the agenda and two new content 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` made the order (it
+refuses to run twice). 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 |
+| 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 sec10_other_tissues figure_plan; do   # keep this order (sec10 reads secS; figure_plan last)
+  python3 "tools/$f.py" || break
+done
+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".

paper-releases/pbisc-diffusion/updates/20261006T145850Z-deck-six-figure-order/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/generate_deck.py

@@ -0,0 +1,361 @@
+#!/usr/bin/env python3
+"""Build the PBISC-Diffusion paper-summary deck.
+
+Reads data/slides_spec.json only and writes
+PBISC-Diffusion_paper_summary_20260930.pptx next to this script.
+Rerunnable; CPU only.
+
+Design rules (user request: plain, clean, no decoration):
+  16:9 13.333 x 7.5 in, white background, Arial only.
+  Text #1A1A1A, footnotes / labels #6E6E6E. Colour only inside figures.
+  Title 26 pt bold at (0.45, 0.30); section label 11 pt grey, top right.
+  Figures fill x 0.45-12.9, y 1.05-6.75, aspect kept, never cropped.
+  Keywords 16 pt, max 3 lines of content, beside or below the figure.
+  Footnote 9 pt grey from y 7.0, at most 2 lines; slide number bottom right.
+  With 'refs' (short citations) the footer is one footnote line plus one 'References:' line;
+  'refs_full' (full citations with DOI) goes to the speaker notes.
+
+The script measures text with the Arial TrueType metrics and fails loudly
+if a title, keyword block or footnote would not fit.
+"""
+from __future__ import annotations
+
+import json
+import re
+import sys
+from pathlib import Path
+
+from PIL import Image, ImageFont
+from pptx import Presentation
+from pptx.dml.color import RGBColor
+from pptx.enum.text import MSO_ANCHOR, PP_ALIGN
+from pptx.util import Emu, Inches, Pt
+
+DECK = Path(__file__).resolve().parent
+SPEC = DECK / "data" / "slides_spec.json"
+
+# ---------------------------------------------------------------- geometry
+SLIDE_W, SLIDE_H = 13.333, 7.5
+X0, X1 = 0.45, 12.9            # content left / right
+Y0, YFIG1 = 1.05, 6.75         # figure top / bottom limit
+KW_BOTTOM = 6.85               # keywords below a figure must end here
+FOOT_Y = 7.0                   # footnote top
+FOOT_W = 11.6                  # leaves room for the slide number
+GAP_BELOW = 0.20               # figure -> keywords (below)
+GAP_RIGHT = 0.35               # figure -> keyword column (right)
+MIN_COL = 2.9                  # narrowest keyword column
+FIG_GAP = 0.25                 # between two figures
+
+INK = RGBColor(0x1A, 0x1A, 0x1A)
+GREY = RGBColor(0x6E, 0x6E, 0x6E)
+WHITE = RGBColor(0xFF, 0xFF, 0xFF)
+FONT = "Arial"
+
+TITLE_PT, LABEL_PT, KW_PT, FOOT_PT, NUM_PT = 26, 11, 16, 9, 9
+AGENDA_PT, COVER_PT, COVER_SUB_PT = 18, 36, 18
+LINE = 1.2                     # line height / font size used for layout
+KW_PARA_GAP_PT = 6
+
+_FONT_FILES = {
+    False: "/usr/share/fonts/truetype/msttcorefonts/Arial.ttf",
+    True: "/usr/share/fonts/truetype/msttcorefonts/arialbd.ttf",
+}
+_font_cache: dict = {}
+
+HANGUL = re.compile("[ᄀ-ᇿ㄰-㆏가-힣]")
+
+
+def text_w_in(text: str, pt: float, bold: bool = False) -> float:
+    """Rendered width of one line in inches (Arial metrics)."""
+    key = bold
+    if key not in _font_cache:
+        _font_cache[key] = ImageFont.truetype(_FONT_FILES[bold], 1000)
+    return _font_cache[key].getlength(text) * pt / 1000.0 / 72.0
+
+
+def wrap(text: str, pt: float, width_in: float, bold: bool = False) -> list[str]:
+    words, lines, cur = text.split(" "), [], ""
+    for w in words:
+        trial = w if not cur else cur + " " + w
+        if text_w_in(trial, pt, bold) <= width_in:
+            cur = trial
+        else:
+            if cur:
+                lines.append(cur)
+            cur = w
+    if cur:
+        lines.append(cur)
+    return lines
+
+
+def line_h(pt: float) -> float:
+    return pt * LINE / 72.0
+
+
+def kw_block_h(keywords: list[str], width_in: float) -> float:
+    n_lines = sum(len(wrap(k, KW_PT, width_in)) for k in keywords)
+    return n_lines * line_h(KW_PT) + (len(keywords) - 1) * KW_PARA_GAP_PT / 72.0
+
+
+# ---------------------------------------------------------------- helpers
+def check_text(s: str, where: str) -> None:
+    if HANGUL.search(s):
+        sys.exit(f"Korean text in {where}: {s!r}")
+    for bad in ("TODO", "TBD", "lorem", "xxxx", "PLACEHOLDER"):
+        if bad.lower() in s.lower():
+            sys.exit(f"placeholder text {bad!r} in {where}: {s!r}")
+
+
+def white_bg(slide) -> None:
+    fill = slide.background.fill
+    fill.solid()
+    fill.fore_color.rgb = WHITE
+
+
+def add_text(slide, x, y, w, h, paras, pt, color=INK, bold=False,
+             align=PP_ALIGN.LEFT, anchor=MSO_ANCHOR.TOP, wrap_text=True,
+             space_before_pt=0):
+    """paras: list of strings, one paragraph each."""
+    tb = slide.shapes.add_textbox(Inches(x), Inches(y), Inches(w), Inches(h))
+    tf = tb.text_frame
+    tf.word_wrap = wrap_text
+    tf.auto_size = None
+    tf.margin_left = tf.margin_right = tf.margin_top = tf.margin_bottom = Emu(0)
+    tf.vertical_anchor = anchor
+    for i, text in enumerate(paras):
+        p = tf.paragraphs[0] if i == 0 else tf.add_paragraph()
+        p.alignment = align
+        p.line_spacing = 1.0
+        if i > 0 and space_before_pt:
+            p.space_before = Pt(space_before_pt)
+        r = p.add_run()
+        r.text = text
+        f = r.font
+        f.name = FONT
+        f.size = Pt(pt)
+        f.bold = bold
+        f.color.rgb = color
+    return tb
+
+
+def add_title(slide, title: str, label: str | None) -> None:
+    label_w = text_w_in(label, LABEL_PT) if label else 0.0
+    room = (X1 - X0) - (label_w + 0.3 if label else 0.0)
+    tw = text_w_in(title, TITLE_PT, bold=True)
+    if tw > room:
+        sys.exit(f"title too long ({tw:.2f} > {room:.2f} in): {title!r}")
+    # title box stops short of the section label so the two never overlap
+    add_text(slide, X0, 0.30, min(12.4, room), 0.5, [title], TITLE_PT,
+             bold=True, wrap_text=False)
+    if label:
+        # baseline of the label aligned with the title baseline
+        asc = 0.905
+        y = 0.30 + asc * TITLE_PT / 72.0 - asc * LABEL_PT / 72.0
+        add_text(slide, X1 - label_w - 0.05, y, label_w + 0.05, 0.25, [label],
+                 LABEL_PT,
+                 color=GREY, align=PP_ALIGN.RIGHT, wrap_text=False)
+
+
+def refs_line(refs: list[str] | None) -> str | None:
+    return ("References: " + "; ".join(refs)) if refs else None
+
+
+def add_footer(slide, footnote: str | None, number: int, refs: list[str] | None = None) -> None:
+    ref = refs_line(refs)
+    if ref:
+        # one footnote line + one reference line (2026-10-02: every slide cites the papers behind its method or metric)
+        if footnote and len(wrap(footnote, FOOT_PT, FOOT_W)) > 1:
+            sys.exit(f"footnote must fit one line when the slide has references: {footnote!r}")
+        if len(wrap(ref, FOOT_PT, FOOT_W)) > 1:
+            sys.exit(f"reference line needs more than one line: {ref!r}")
+        check_text(ref, "references")
+        y = FOOT_Y
+        if footnote:
+            add_text(slide, X0, y, FOOT_W, line_h(FOOT_PT) + 0.02, [footnote], FOOT_PT, color=GREY)
+            y += line_h(FOOT_PT) + 0.02
+        add_text(slide, X0, y, FOOT_W, line_h(FOOT_PT) + 0.02, [ref], FOOT_PT, color=GREY)
+    elif footnote:
+        lines = wrap(footnote, FOOT_PT, FOOT_W)
+        if len(lines) > 2:
+            sys.exit(f"footnote needs {len(lines)} lines (max 2): {footnote!r}")
+        add_text(slide, X0, FOOT_Y, FOOT_W, 2 * line_h(FOOT_PT) + 0.02,
+                 [footnote], FOOT_PT, color=GREY)
+    add_text(slide, X1 - 0.6, FOOT_Y, 0.6, line_h(NUM_PT), [str(number)],
+             NUM_PT, color=GREY, align=PP_ALIGN.RIGHT, wrap_text=False)
+
+
+def add_notes(slide, s: dict) -> None:
+    parts = []
+    if s.get("say"):
+        parts.append(s["say"])
+    full = s.get("footnote_full")
+    if full and full != s.get("footnote"):
+        parts.append("Full footnote: " + full)
+    if s.get("refs_full"):
+        parts.append("References: " + " | ".join(s["refs_full"]))
+    if s.get("sources"):
+        parts.append("Sources: " + s["sources"])
+    if parts:
+        slide.notes_slide.notes_text_frame.text = "\n\n".join(parts)
+
+
+# ---------------------------------------------------------------- layout
+def aspect(path: Path) -> float:
+    with Image.open(path) as im:
+        w, h = im.size
+    return w / h
+
+
+def fit(total_aspect: float, extra_w: float, wmax: float, hmax: float):
+    """Largest (w, h) with w = h * total_aspect + extra_w inside the box."""
+    h = min(hmax, (wmax - extra_w) / total_aspect)
+    return h * total_aspect + extra_w, h
+
+
+def plan(aspects: list[float], keywords: list[str], layout: str):
+    """Return dict with figure height/width and keyword placement."""
+    n = len(aspects)
+    tot = sum(aspects)
+    extra = FIG_GAP * (n - 1)
+    full_w, full_h = X1 - X0, YFIG1 - Y0
+    options = []
+    if not keywords:
+        w, h = fit(tot, extra, full_w, full_h)
+        return {"mode": "none", "w": w, "h": h}
+    # keywords below
+    kh = kw_block_h(keywords, full_w)
+    hmax = min(full_h, KW_BOTTOM - GAP_BELOW - kh - Y0)
+    w, h = fit(tot, extra, full_w, hmax)
+    kh = kw_block_h(keywords, full_w)   # keywords sit at the text margin X0
+    if Y0 + h + GAP_BELOW + kh <= KW_BOTTOM + 1e-6:
+        options.append({"mode": "below", "w": w, "h": h, "area": w * h,
+                        "kw_w": full_w, "kw_h": kh})
+    # keywords right
+    w, h = fit(tot, extra, full_w - GAP_RIGHT - MIN_COL, full_h)
+    col = X1 - (X0 + w) - GAP_RIGHT
+    kh = kw_block_h(keywords, col)
+    if kh <= full_h:
+        options.append({"mode": "right", "w": w, "h": h, "area": w * h,
+                        "kw_w": col, "kw_h": kh})
+    if not options:
+        sys.exit(f"no layout fits keywords {keywords!r}")
+    best = max(options, key=lambda o: o["area"])
+    if layout == "figure_keywords":
+        right = [o for o in options if o["mode"] == "right"]
+        if right and right[0]["area"] * 1.15 >= best["area"]:
+            best = right[0]
+    return best
+
+
+def build_content(prs, s: dict, number: int) -> None:
+    slide = prs.slides.add_slide(prs.slide_layouts[6])
+    white_bg(slide)
+    for fld in ("title", "footnote", "section_label"):
+        if s.get(fld):
+            check_text(s[fld], f"slide {number} {fld}")
+    kws = s.get("keywords", [])
+    if len(kws) > 3:
+        sys.exit(f"slide {number}: more than 3 keywords")
+    for k in kws:
+        check_text(k, f"slide {number} keyword")
+    add_title(slide, s["title"], s.get("section_label"))
+
+    paths = [DECK / f for f in s["figures"]]
+    for p in paths:
+        if not p.is_file():
+            sys.exit(f"slide {number}: missing figure {p}")
+    asp = [aspect(p) for p in paths]
+    L = plan(asp, kws, s.get("layout", "full"))
+    w, h = L["w"], L["h"]
+    if L["mode"] == "below":
+        fx = X0 + ((X1 - X0) - w) / 2.0      # centred; keywords at the text margin
+    else:
+        fx = X0
+    fy = Y0
+    x = fx
+    for p, a in zip(paths, asp):
+        pw = h * a
+        slide.shapes.add_picture(str(p), Inches(x), Inches(fy),
+                                 Inches(pw), Inches(h))
+        x += pw + FIG_GAP
+    assert fy + h <= YFIG1 + 1e-6 and fx + w <= X1 + 1e-6, s["title"]
+
+    if L["mode"] == "below":
+        add_text(slide, X0, fy + h + GAP_BELOW, L["kw_w"], L["kw_h"] + 0.05,
+                 kws, KW_PT, space_before_pt=KW_PARA_GAP_PT)
+    elif L["mode"] == "right":
+        kx = fx + w + GAP_RIGHT
+        add_text(slide, kx, fy + 0.15, L["kw_w"], L["kw_h"] + 0.05, kws,
+                 KW_PT, space_before_pt=KW_PARA_GAP_PT)
+    add_footer(slide, s.get("footnote"), number, s.get("refs"))
+    add_notes(slide, s)
+    print(f"{number:>2}  {L['mode']:<5} fig {w:5.2f}x{h:4.2f} in  {s['title']}")
+
+
+def build_title(prs, s: dict) -> None:
+    slide = prs.slides.add_slide(prs.slide_layouts[6])
+    white_bg(slide)
+    check_text(s["title"] + s.get("subtitle", ""), "title slide")
+    if text_w_in(s["title"], COVER_PT, bold=True) > X1 - X0:
+        sys.exit("cover title too long")
+    add_text(slide, X0, 2.95, X1 - X0, 0.7, [s["title"]], COVER_PT, bold=True,
+             wrap_text=False)
+    if s.get("subtitle"):
+        add_text(slide, X0, 3.8, X1 - X0, 0.4, [s["subtitle"]], COVER_SUB_PT,
+                 color=GREY, wrap_text=False)
+
+
+def build_agenda(prs, s: dict, number: int) -> None:
+    slide = prs.slides.add_slide(prs.slide_layouts[6])
+    white_bg(slide)
+    add_title(slide, s["title"], None)
+    items = s["items"]
+    if not 6 <= len(items) <= 10:  # 10 lines fit the slide; the figure-order deck (2026-10-06) lists 8
+        sys.exit("agenda must list 6 to 10 sections")
+    step = 0.52
+    for i, item in enumerate(items):
+        check_text(item, "agenda")
+        y = 1.3 + i * step
+        add_text(slide, X0, y, 0.45, 0.35, [str(i + 1)], AGENDA_PT, color=GREY,
+                 align=PP_ALIGN.RIGHT, wrap_text=False)
+        add_text(slide, X0 + 0.65, y, 10.5, 0.35, [item], AGENDA_PT,
+                 wrap_text=False)
+    add_footer(slide, None, number)
+
+
+def build_divider(prs, s: dict, number: int) -> None:
+    slide = prs.slides.add_slide(prs.slide_layouts[6])
+    white_bg(slide)
+    check_text(s["title"], "divider")
+    add_text(slide, X0, 2.95, X1 - X0, 0.7, [s["title"]], COVER_PT, bold=True,
+             wrap_text=False)
+    add_footer(slide, None, number)
+
+
+def main() -> None:
+    spec = json.loads(SPEC.read_text())
+    out = DECK / spec["deck"]["output_pptx"]
+    prs = Presentation()
+    prs.slide_width = Inches(SLIDE_W)
+    prs.slide_height = Inches(SLIDE_H)
+    for i, s in enumerate(spec["slides"], start=1):
+        kind = s["kind"]
+        if kind == "title":
+            build_title(prs, s)
+            print(f"{i:>2}  title  {s['title']}")
+        elif kind == "agenda":
+            build_agenda(prs, s, i)
+            print(f"{i:>2}  agenda {s['title']}")
+        elif kind == "divider":
+            build_divider(prs, s, i)
+            print(f"{i:>2}  divider {s['title']}")
+        elif kind == "content":
+            build_content(prs, s, i)
+        else:
+            sys.exit(f"unknown slide kind {kind!r}")
+    prs.save(out)
+    print(f"wrote {out} ({len(spec['slides'])} slides)")
+
+
+if __name__ == "__main__":
+    main()