Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261013T034842Z-expanded-131-slide-master/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/expanded_master_20261013/slide_map.csv

@@ -0,0 +1,132 @@
+slide,original_slide,addition_key,title,section,footnote
+1,1,,PBISC-Diffusion: expanded evidence deck,,
+2,2,,Full evidence map,,
+3,,guide:scope,How to read this expanded evidence deck,Evidence scope,Original figures retain their data and evidence lineage; sources and limitations are in the speaker notes.
+4,3,,Manuscript plan: six main figures,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+5,,guide:availability,Availability is not the same as a new experiment,Evidence map,"Planning cards from the original deck are retained as dated snapshots, not a fresh list of experiments to rerun."
+6,4,,Figure 1. Design and performance of PBISC,,
+7,5,,Figure 1 plan: Design and performance of PBISC,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+8,6,,Backbone: SDD → PBISC-Diffusion,Fig. 1a,"SDD: Ostheimer et al., arXiv 2502.02448. scDiffusion: Luo et al., Bioinformatics 2024."
+9,7,,Overview draft (Nature style),Fig. 1b,Schematic. Conditioning recipe size and the number of generated cells are distinct; no cell-type quotas.
+10,8,,Bulk conditioning,Fig. 1c,Final 6k model. Recipe size K conditions the bulk; the number of generated cells is chosen separately.
+11,9,,Pretraining corpus,Fig. 1d,Corpus labels describe the data; they are not model inputs. Dataset composition is not sampler exposure.
+12,10,,Width & context ablation: loss saturation,Fig. 1e,Old 6k panel (1% floor); one validation set. 16x: memory-floor stop after part 2.
+13,11,,Width & context ablation: benchmark,Fig. 1e,"SLE, 42 donors, 2,980 genes; old 6k panel. ctx x4: better L1 (0.161 vs 0.197)."
+14,12,,Gene context: 2k / 3k / 6k / 15k,Fig. 1f,Existing panel benchmark. 6k: five sampling seeds; other panels: one seed. Different axis from the architecture sweep.
+15,,review:6,Native-pseudobulk model comparisons,Fig. 2a,Completed held-out rescoring. Architecture: 1% detection floor; gene-panel comparison: 0.1% floor.
+16,13,,"PBISC-Diffusion: bulk in, cells out",Fig. 1 details,"Fig. 1b redraw. Final model corpus: 33.1M train cells; K = 1–4,000 cells per pseudobulk."
+17,14,,Denoiser: joint value + support diffusion,Fig. 1 details,"Fig. 1a redraw. Final model f01 6k 1x, epoch 16, EMA; 50 DDIM steps."
+18,,healthy_sle_conditioning_interventions,Does the model follow its bulk conditioning?,Control evidence,"Existing SLE pseudobulk controls: same noise, 42 donors, 2,279 genes. This is a conditioning intervention, not real-bulk validation."
+19,15,,Figure 2. Fine-tuning PBISC with LoRA,,
+20,16,,Figure 2 plan: Fine-tuning PBISC with LoRA,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+21,17,,Real bulk ≠ pseudobulk (model input),Fig. 2a,Existing paired-input diagnostic; mean-shift removal uses paired pseudobulks and is not a deployment method.
+22,18,,Same shift in three paired cohorts,Fig. 2a,"Model inputs only, no generation. PCA fitted per cohort on sc pseudobulks."
+23,19,,Consequence: cells off-manifold,Fig. 2b,Existing same-noise input diagnostic; not the new LoRA/scSemi OOF embedding. UMAP overlap is not a quantitative score.
+24,,healthy_stored_real_reference_umap,Input corrections on one real-reference UMAP,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+25,,healthy_saved_density_difference,Where generated-cell density differs from real cells,Control evidence,Original high-resolution density map; zoom for panel labels. Quantitative colors preserved. Historical native-depth controls.
+26,,healthy_input_control_composition,Input calibration changes cell-type composition,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+27,,healthy_stored_matched_trajectories,"Same noise, different input: matched-cell shifts",Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+28,,healthy_matched_similarity,Matched noise separates cell state from cell identity,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+29,,healthy_label_transitions,Input changes can switch the generated cell label,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+30,,healthy_nearest_real_by_type,Nearest real cells: distance and label agreement,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+31,,healthy_within_type_centroids,Composition agreement does not ensure state agreement,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+32,,healthy_gene_length_input,Gene-length effects in the bulk input,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+33,,healthy_gene_length_stable,Gene-length effects within label-stable cells,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+34,,healthy_raw_tpm_top_genes,Raw and TPM shifts persist within cell labels,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+35,,healthy_refcal_top_genes,Reference calibration leaves specific gene-state shifts,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+36,,healthy_donor_axes,Donor trait axes and donor identification differ,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+37,,healthy_reference_choice,Reference choice changes different quality endpoints,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+38,20,,Fine-tuning with paired bulk + sc,Fig. 2c,Base: final model f01 6k 1x E16. Times from 40 LoRA runs on this host.
+39,21,,Evaluation design,Fig. 2c,Bars: Ye 2026 sepsis. QC-failed pairs excluded.
+40,,guide:protocol,Why the historical and strict protocols differ,Protocol boundary,Support draws overlap in donors. Displayed draw variation is not independent population replication.
+41,22,,How we score generated cells,Fig. 2d,"Definitions follow the cited papers; scorers: tools/score_ft.py, extra_metrics.py, contrast_extra.py."
+42,,healthy_immicro_replicate_controls,Technical repeats and follow-up are different controls,Control evidence,Observed ImMicro controls. Technical remeasurement and follow-up: 16 donors each. Follow-up is not a technical replicate.
+43,,healthy_annotation_agreement,Annotation agreement and difficult categories,Control evidence,Observed cells only. Overall agreement uses the supported six-type gate; author mixed/unlabelled cells are separate. Other is difficult.
+44,23,,Healthy cohort: LoRA cells cover the real ones,Fig. 2e,"ImMicro, 107 healthy donors. 5 donors: mean of 5 draws (dots); 5-fold CV shown separately."
+45,24,,Healthy cohort: realism beyond precision/coverage,Fig. 2f,"ImMicro, 107 healthy donors. Dashed = real vs real cells of the same donors."
+46,,healthy_depth_matched_quality,Depth matching changes precision and coverage,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+47,,healthy_detection_and_library,Detection and library size are separate diagnostics,Control evidence,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+48,25,,Healthy cohort: full vs LoRA vs condition encoder,Fig. 2g,"ImMicro, 107 healthy donors; cond.-encoder-only ran on fold 0 (22 donors) only."
+49,26,,Healthy cohort: how many pairs,Fig. 2h,Historical n + 7 validation design. The fixed-3000-step strict curve is a separate protocol update.
+50,27,,Healthy cohort: LoRA rank and K,Fig. 2i,"ImMicro, 107 donors, 5-fold CV pooled. Open symbols: K choices with every training pair."
+51,,review:8,Healthy bulk: full fine-tuning and LoRA rank,Fig. 2c,Original donor-level five-fold comparison. LoRA rank 16 is a quality/parameter compromise.
+52,28,,Figure 3. PBISC vs scSemiProfiler on healthy donors,,
+53,29,,Figure 3 plan: PBISC vs scSemiProfiler on healthy donors,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+54,30,,"Healthy donors, 5 pairs: LoRA-5 vs scSemiProfiler-5",Fig. 3a,"ImMicro: exactly 5 pairs, no extra validation, 3000-step LoRA; five draws, remaining 102 donors per draw."
+55,,review:13,Composition: error and donor-level wins,Fig. 3c,"Exactly five pairs per draw. Absolute errors averaged within each held-out donor; 107 donors, equal weight. Descriptive."
+56,31,,Healthy paired-data learning curves,Fig. 3b,"LoRA: n training + 7 validation. scSemi: n + 7 representatives, with CV-HVG reuse. Not the strict total-n protocol."
+57,32,,Representative-dependent composition,Fig. 3c,"Existing healthy five-draw diagnostic. This evaluates composition dependence, not literal copying of every cell."
+58,33,,"Time: days for scSemiProfiler, minutes for LoRA",Fig. 3d,"One GB10 GPU, job alone on it. GPU time from receipts; waits for the GPU excluded."
+59,34,,5 donors: time to apply,Fig. 3d,"Existing measurements: adaptation + target generation; pretraining, GPU wait and scoring excluded."
+60,35,,Figure 4. Disease signal on paired cohorts,,
+61,36,,Figure 4 plan: Disease signal on paired cohorts,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+62,37,,Five paired cohorts,Fig. 4a,"Existing paired-cohort map. Five donors means 5 pairs in Sepsis, 12-15 Imperial, 12-14 Giroux, 7-9 Ji."
+63,38,,Disease signals with five training donors,Fig. 4b,Existing five-donor protocol; sample counts differ by cohort. Original sample-level estimator; corrected update separate.
+64,,disease_strict_sepsis_quality,Sepsis: exact-five reconstruction across five draws,Fig. 4 update,Completed exact-five Sepsis draws. Bars = means; dots = support draws. Five matched methods; no new generation.
+65,,disease_strict_sepsis_realism,Sepsis: realism beyond precision and coverage,Fig. 4 update,"Completed exact-five outputs; raw readout, fixed R1 reference. Real/real control shown. Dots are five support draws."
+66,,disease_strict_imperial_quality,Imperial: first completed exact-five comparison,Fig. 4 update,One completed support draw (draw 0). Exact five paired samples; all five methods scored. Not the final five-draw comparison.
+67,39,,5-fold CV: disease signal,Fig. 4c,Original donor-level split and original signal estimator. These results are distinct from the exact-five update.
+68,40,,Disease genes at every list length (CAT),Fig. 4d,Existing CAT curves; ranking agreement and chance line. Patient-aware reanalysis reuses the same predictions.
+69,,disease_strict_ye2026_sepsis_deg,Sepsis: patient-aware disease-gene agreement,Fig. 4 update,Completed corrected count-based DEG fits; mean of five support draws.
+70,,disease_strict_imperial_eskd_deg,Imperial: patient-aware disease-gene agreement,Fig. 4 update,Single completed exact-five draw 0; patient-aware DEG fits. Other draws are outside this frozen snapshot.
+71,41,,"Cell-type deconvolution, same pairs: disease signal",Fig. 4e,"Completed BayesPrism/CIBERSORTx comparisons; same five-donor support, method-supported expression metrics."
+72,42,,SLE benchmark: PBISC vs scSemiProfiler,Fig. 4f,"Completed pseudobulk benchmark; separate from external paired real-bulk validation. PBISC five seeds, scSemi one run."
+73,43,,SLE benchmark: per cell type,Fig. 4f,"Same 42 donors, 1,006 genes. PBISC: 5 seeds, mean ± SD; scSemiProfiler: 1 run. Other: top-50 of 12."
+74,,healthy_sle_disjoint_truth_control,Disjoint-cell controls test the bulk-broadcast baseline,Control evidence,SLE pseudobulk benchmark; five existing half-cell splits. Error bars are saved split SDs. No new cell splitting or scoring.
+75,,disease_sle_broadcast,SLE: what is added beyond the bulk broadcast?,Development benchmark,Existing SLE pseudobulk development benchmark; bulk broadcast is an explicit comparator. Separate from paired real-bulk validation.
+76,,disease_influenza_response,Influenza: response order versus response genes,External diagnostic,Historical zero-shot influenza test. Response-strength ordering is recovered; gene-wise agreement remains below the matched bulk broadcast.
+77,,disease_covid_map,COVID: calibration on a shared real-cell map,External diagnostic,"Historical 7 COVID / 5 HC samples, common depth. High-resolution reference overview; following detail panels provide readable quantitative results."
+78,,disease_covid_direction,COVID monocytes: direction after depth matching,External diagnostic,"Historical 7 COVID / 5 HC samples, common depth. High-resolution reference overview; following panels separate direction, magnitude and relabeling null."
+79,,disease_covid_monocyte_direction_detail,"COVID monocytes: direction, magnitude and null",External diagnostic,Readable detail of the retained COVID overview. Existing 7 COVID / 5 HC paired samples; native and depth-matched results stay separate.
+80,,disease_covid_cd14_direction_detail,"COVID CD14 monocytes: direction, magnitude and null",External diagnostic,Readable detail of the retained COVID overview. Existing 7 COVID / 5 HC paired samples; native and depth-matched results stay separate.
+81,,disease_covid_endpoints,COVID: biological endpoints give a mixed verdict,External diagnostic,Historical 7 COVID / 5 HC samples. Saved effect ratios and donor ranks; no new fits. Reference calibration is not uniformly best.
+82,44,,Sepsis: fine-tuning variants,Fig. 4 details,"Sepsis, 82 donors, pooled 5-fold. LoRA r = 16 is the model used throughout."
+83,45,,Pairs needed: donor and disease signal,Fig. 4 details,Sepsis: 61 case vs 21 control. logFC r = mean over T / B / Mono / NK. 0 = no pairs.
+84,46,,"Cell-type deconvolution, same pairs: composition",Fig. 4 details,Completed deconvolution outputs already exist. Verified identical Sepsis supports use adoption instead of method reruns.
+85,,disease_deconv_expression_coverage,Deconvolution: expression coverage limits the comparison,Deconvolution detail,Completed historical CIBERSORTx five-donor runs. Missing disease correlations remain unavailable; deconvolution does not generate individual cells.
+86,47,,Figure 5. Rare cell states in disease cohorts,,
+87,48,,Figure 5 plan: Rare cell states in disease cohorts,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+88,49,,Rare-state fractions: original analysis,Historical scorer,Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate.
+89,,review:22,Rare-state frequencies in observed cells,Fig. 5a,Original five-donor outputs. Median across draws; state share within its parent population. Controls included.
+90,50,,Rare-state ranking: original analysis,Historical scorer,Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate.
+91,,disease_rare_misses_1,Rare-state misses: rules 1–8,Fig. 5 detail,"Saved zero-aware diagnostic; all 15 predefined rules are shown across two slides. NA means not evaluable, not zero error."
+92,,disease_rare_misses_2,Rare-state misses: rules 9–15,Fig. 5 detail,"Saved zero-aware diagnostic; all 15 predefined rules are shown across two slides. NA means not evaluable, not zero error."
+93,,review:23,Rare-state fidelity: zero-aware reanalysis,Fig. 5b,"Original five-donor outputs; zero draws retained and actual donor IDs. 59 state/cohort combinations, including controls."
+94,51,,Rare states: original five-fold analysis,Historical scorer,Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate.
+95,,disease_sepsis_rare_support,Sepsis rare states across paired-data budgets,Fig. 5 detail,"Historical cross-fitted learning curve: n training pairs plus validation, not strict total n. All 15 rules; arm-specific depth matching."
+96,,disease_sepsis_rare_direction,Sepsis rare-state direction across support budgets,Fig. 5 detail,"Saved sign-agreement counts across all 15 rules, including controls. Historical n + validation; directions are not effect-size accuracy."
+97,,disease_external_rare_screen,"Rare-state screen in DRESS, Kawasaki and SLE",Rare-state appendix,Historical one-seed zero-shot screen conditioned on each sample's own sc pseudobulk. DRESS was in training; real-bulk validation is separate.
+98,52,,Consequence: disease direction flips,Fig. 5d,Historical Behcet diagnostic; reference is unmatched and includes training donors. This is not independent patient-level subtype validation.
+99,,disease_c1q_cross_disease,C1Q: single-gene detection versus co-detection,Rare-state appendix,"Existing pseudobulk-conditioned comparison at matched depth. Multi-gene C1Q rates remain compressed, including in-corpus DRESS."
+100,,disease_behcet_k4000,"Behçet: K=4,000 changes direction more than identity",Behçet diagnostic,"Two existing sampling seeds; 6 BD / 10 HC held-out bulks. K is condition size; each bulk still generates 5,000 cells. Descriptive."
+101,,review:24,Behcet: one-gene and multi-gene C1Q rules,Fig. 5c,Descriptive: unmatched reference includes training donors. Same three pairs; two identity-pass and one uncertain.
+102,53,,Figure 6. Discovery from public bulk-only data,,
+103,54,,Figure 6 plan: Discovery from public bulk-only data,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+104,55,,Bulk-only cohort plan: original snapshot,Historical plan,Historical cohort counts. Existing exploratory outputs are reviewed separately; reproduction with the user-prepared new dataset remains pending.
+105,,disease_bulkonly_verification_history,Historical bulk-only search: no confirmed new finding,Historical exploration,Historical exploratory analysis of existing bulk-only generations. Review dispositions are not experimental validation. Figure 6 reproduction still awaits new data.
+106,,disease_bulkonly_imperial_wave_shift,Bulk-only exploration: calibration populations differ,Historical exploration,Historical Imperial sample composition. Paired and bulk-only sets differ in wave; repeated samples are not independent donors.
+107,,guide:figure6,What remains to establish Figure 6,Figure 6,Historical candidate analyses do not substitute for independent reproduction in the new cohort.
+108,56,,Supplementary,,
+109,57,,Supplementary plan,Original roadmap,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+110,58,,5 donors: LoRA vs scSemiProfiler — cells,Supplementary,Completed five-donor comparisons. Multi-visit cohorts used more than five paired samples.
+111,59,,5 donors: realism beyond precision/coverage,Supplementary,"E_N: scSemiProfiler's own error metric. Misses only vs calibration, e.g. Imperial E_N 1.28 vs 1.22."
+112,60,,"5 donors: precision/coverage at k = 3, 5, 10",Supplementary,"Completed sensitivity at k=3,5,10; no new sensitivity campaign required."
+113,61,,5-fold: realism beyond precision/coverage,Supplementary,"Donor-level 5-fold CV, one pooled call per cohort. E_N: scSemiProfiler's own error metric."
+114,62,,Same training pairs: LoRA vs scSemiProfiler,Supplementary,Completed original five-fold LoRA/scSemi comparison; not the new total-n curve.
+115,63,,scSemiProfiler and the representative-copy baseline,Supplementary,Copy = composition of the one training sample scSemiProfiler assigns to each test sample.
+116,64,,Full vs LoRA vs cond.-encoder-only,Supplementary,"Fold 0, cond.-enc. vs ZS + cal.: coverage 0.73 vs 0.73 / 0.74 vs 0.63, L1 p 0.06 / 0.93 (ImMicro / sepsis)."
+117,65,,"Pairs needed: composition, coverage, training steps",Supplementary,"ImMicro / sepsis, 5-fold CV (107 / 82 donors). 0 = no pairs; squares = calibration fitted on the same n pairs."
+118,66,,"LoRA rank: r = 4, 16, 64",Supplementary,"Sepsis, 58–59 pairs/fold: r64 logFC r 0.53 → 0.55, DEG 23 → 26; r4 0.48, DEG 18. Pooled 5-fold scores."
+119,67,,K diversification: no gain,Supplementary,"LoRA r16, 5-fold pooled. p = paired Wilcoxon on donor L1, uncorrected."
+120,,disease_k4000_quality,"K=4,000 recipe: paired reconstruction quality",Recipe sensitivity,"Existing paired benchmark, five support draws per cohort and 1,500 cells/test sample. Bars = means; dots = draws. No new runs."
+121,,disease_k4000_disease,"K=4,000 recipe: disease signal does not improve uniformly",Recipe sensitivity,"Existing paired benchmark, five support draws per cohort and 1,500 cells/test sample. Bars = means; dots = draws. No new runs."
+122,,disease_k4000_calibration,"K=4,000 recipe: calibration gains are selective",Recipe sensitivity,"Saved paired-test gates: 50 features × 5 draws per cohort. Descriptive counts, not 250 independent biological replicates."
+123,68,,Pairs from other cohorts do not transfer,Supplementary,Completed leave-one-cohort-out test at K=1000; target-cohort sc was excluded from training.
+124,69,,Longer training on other cohorts: gap stays,Supplementary,Existing saturation experiment; no additional saturation run introduced here.
+125,70,,"No pretraining: same cohort, from scratch",Supplementary,"Completed Sepsis fold0 pilot, not all folds. Pretraining supports speed/few-pair adaptation, not universal superiority."
+126,71,,"No pretraining: other cohorts, from scratch",Supplementary,"Pilot: held-out sepsis, 82 samples. Genes no training cohort measured are zeroed in every arm."
+127,72,,Rare states: from scratch vs pretrained,Supplementary,"Pilot, one seed: sepsis fold 0 (17 donors), held-out sepsis (82, masked). 1 = observed rate; grey = within 2-fold."
+128,73,,Appendix,,
+129,74,,Appendix: scSemiProfiler without output threshold,Appendix,Completed modified-readout check; main results use official output. Models were not retrained for the sensitivity.
+130,75,,Discussion,,
+131,76,,Next: other tissues from a few dozen pairs,Discussion,Pilots: sepsis fold 0 (17 donors); last row held-out sepsis (82). Folds 1–4 and a second cohort pending.

paper-releases/pbisc-diffusion/updates/20261013T034842Z-expanded-131-slide-master/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/expanded_master_20261013/task_plan.md

@@ -0,0 +1,23 @@
+# Expand the complete PBISC evidence deck
+
+## Goal
+Expand the original 76-slide PPTX into a comprehensive manuscript-working master. Preserve every original slide's content and add distinct available results, methods and diagnostics. Keep PBISC muted yellow/gold and scSemiProfiler blue. No new experiments.
+
+## Phases
+- [x] Confirm scope: expanded master, not a shortened talk.
+- [x] Inventory additional evidence and classify completed / historical / exploratory / pending.
+- [x] Extend semantic method palette to all original and added figures.
+- [x] Build original 76 plus additions with per-slide provenance and coverage map.
+- [x] Render and verify all pages, numerical fidelity, original coverage and scientific labels.
+- [ ] Publish exact archive delta and deliver expanded PPTX/PDF.
+
+## Constraints
+- All 76 original slide IDs must map to the expanded deck. Correct misleading protocol labels while retaining provenance.
+- Use original Arial renderer and readable figure scale; do not pad with duplicate plots.
+- Retain detailed backbone/denoiser, LoRA design, evaluation/metrics, condition-encoder controls, per-type benchmarks, rare CV and scratch/K ablations.
+- Preserve known caveats: historical n+7 validation and CV-HVG reuse; five donors are not five paired samples in multi-visit cohorts; rare zero/donor corrections; Behcet descriptive limitations; bulk-only reproduction awaits user data.
+- Completed saved outputs may be plotted on bounded CPU. No new training/inference/deconvolution or paid APIs.
+- Current 34-slide review is a derived subset, not the master. Do not overwrite original 76 or frozen releases.
+
+## Status
+131-slide expanded master completed and visually verified. All original76slides retained;55additions. Exactarchivepublication is the remaining step.