Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261013T044744Z-figure1to6-131page-classification/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/figure_assignment_20261013/ASSIGNMENT.md

@@ -0,0 +1,145 @@
+# 131장 자료의 새 Figure 1–6 / Supplementary 분류
+
+사용자가 2026-10-13에 다시 정의한 여섯 Figure를 기준으로 분류했다. **아래 쪽수는 131장 확장 덱의 쪽수**다. 덱 안의 기존 Figure 번호와 계획표는 이전 기획이다. 원본 PPTX/PDF는 수정하거나 삭제하지 않았다.
+
+- [131장 전체 분류표](slide_assignment.csv): 페이지별 주배정, 데이터 종류, 사용 조건, 필요한 수정, 원본 76장 번호.
+- [패널 구성안](panel_plan.json): Figure별 패널과 재사용 범위.
+- [사용자 정의](AUTHOR_SCOPE.md): 앞으로 적용할 Figure 1–6의 범위.
+
+## 본문 Figure 구성안
+
+| Figure | 본문에 넣을 자료 | 현재 페이지 | 상태 / 경계 |
+|---|---|---|---|
+| **1** | 전체 실험 흐름, corpus 통계, dataset별 규모와 K 분포 | **9, 11, 16 하단** | 기존 자료 재사용. 전체 corpus와 training subset의 분모를 구분 |
+| **2** | architecture, gene context/panel ablation, 공통 유전자 축의 비교 | **13, 14, 15** | **SLE pseudobulk**만 사용. 12의 corpus validation loss는 보충 |
+| **3** | healthy mismatch → full FT 개선 → LoRA rank 비교 → paired 수 ablation → exact-five 성능 | **21, 23, 116의 ImMicro pooled, 51, 49, 54**; **55** 선택 | 모두 healthy paired. 49의 추가 validation과 54의 exact-five를 구분 |
+| **4** | 질환 cohort, LoRA-five/5-fold의 logFC·DEG·CAT, deconvolution 비교 | **62 질환 행, 63, 67, 68, 71** | 핵심 정량 결과. 비교 단위와 cohort별 예외를 유지 |
+| **5** | paired sample의 real/LoRA/scSemi rare population, marker, donor별 사례 | **직접 충족하는 완성 슬라이드는 현재 없음** | 기존 생성 결과로 정성 패널 조립. 89, 91–96은 정량 보조 자료 |
+| **6** | paired adaptation → bulk-only inference → 기존 논문의 특정 결과 재현 | **TODO**; 설계 참고 **104, 106, 107** | 재현 결과 패널은 아직 없음. 105는 과거 탐색 기록 |
+
+## Figure 1 — 전체 실험 개요와 corpus
+
+- **1a / 9쪽:** pretraining → paired adaptation → single-cell generation의 전체 실험 흐름.
+- **1b / 11쪽:** corpus 규모 및 assay, condition, cell-type 분포. cell·donor·dataset의 분모를 명시한다.
+- **1c / 16쪽 하단:** dataset별 corpus 규모와 recipe-size 분포. 상단 workflow는 9쪽과 중복되므로 Methods/S01로 이동한다.
+
+8, 10, 17, 38–41쪽의 세부 방법은 Methods/보충으로 배치한다. 필요하면 62쪽의 cohort 요약을 작은 evaluation inset으로 재사용한다. Corpus의 자연적 구성과 학습 sampler의 노출 비율은 별도 통계다.
+
+## Figure 2 — SLE pseudobulk에서 architecture와 gene context
+
+- **2a / 13쪽:** architecture와 conditioning width에 따른 SLE pseudobulk 성능.
+- **2b / 14쪽:** 2k/3k/6k/15k gene context 비교. 6k가 모든 지표에서 최고라는 표현 대신 성능과 biological gene coverage를 함께 보여준다.
+- **2c / 15쪽:** 공통 1,006-gene 축의 precision/coverage 및 composition 비교. Architecture의 1% floor와 panel의 0.1% floor 실험군은 나누어 해석한다.
+
+**12쪽**은 corpus pretraining validation loss이므로 SLE benchmark 성능으로 부르지 않는다. **18, 72–75쪽**의 conditioning intervention, per-cell-type SLE, broadcast 및 split-half controls는 S02로 분류한다. 특히 72–75쪽은 이전 덱의 위치와 관계없이 새 Figure 4에 들어가지 않는다.
+
+## Figure 3 — healthy paired adaptation의 필요성과 효율
+
+패널 흐름은 **21 → 23 → 116의 healthy 부분 → 51 → 49 → 54**로 잡는다. 55쪽의 donor별 composition 오차는 선택 패널이다.
+
+**Full FT의 개선은 raw zero-shot과 비교하면 지지된다.** ImMicro pooled 결과의 precision/coverage/L1은 raw에서 **0.6804/0.1918/0.4549**, full FT에서 **0.8895/0.6954/0.1572**다. Calibrated zero-shot의 precision은 0.9050으로 full FT보다 높으므로, 모든 baseline의 모든 지표가 개선됐다고 표현하지 않는다.
+
+| ImMicro pooled CV | Precision ↑ | Coverage ↑ | Composition L1 ↓ | Trainable parameters |
+|---|---:|---:|---:|---:|
+| Full FT | 0.8895 | 0.6954 | **0.1572** | 54.185M |
+| LoRA r4 | **0.9333** | 0.7123 | 0.1858 | **0.324M** |
+| LoRA r16 | 0.9241 | 0.7843 | 0.1617 | 1.296M |
+| LoRA r64 | 0.9044 | **0.8159** | 0.1610 | 5.185M |
+
+**현재 결과로 r16을 모든 지표에서 가장 좋은 rank라고 단정할 수는 없다.** r4는 precision, r64는 coverage와 LoRA 중 L1에서 앞선다. r16은 r64보다 학습 파라미터가 4배 적고 precision이 높으며 관측 L1 차이가 작은 절충안이다. 이는 통계적 동등성을 입증한 결과가 아니다. 5 training + 7 validation 조건에서도 r4는 L1, r64는 coverage, r16은 precision에서 가장 좋아 순위가 나뉜다.
+
+권장 설명: **“성능과 학습 파라미터 수의 균형을 고려해 rank 16을 선택했다.”** 이 선택 이유는 제안이며, 사전에 정한 종합 선택 기준이 ��었는지는 별도로 확인해야 한다.
+
+- **116쪽:** `dataset=immicro, call=pooled`만 추출해 raw/calibrated/full FT/LoRA를 비교한다. Sepsis나 22명 fold0의 condition-encoder pilot을 107명 pooled 결과와 합치지 않는다.
+- **51쪽:** healthy full FT/rank/parameter 비교의 주패널. 50쪽의 rank별 곡선은 S06에서 보완한다.
+- **49쪽:** n training + 7 validation 및 early stopping 곡선이다. n=5를 “총 5 paired만 사용”으로 읽을 수 없다.
+- **54쪽:** 5 pairs, 추가 validation 없음, fixed 3,000 steps의 exact-five 비교다. **“5 paired로도 성능을 확보한다”의 핵심 근거는 이 패널**이다. 5 support draws와 각 draw의 나머지 102 donors를 구분한다.
+- **56쪽:** historical scSemi budget curve에는 CV-HVG 재사용이 있어 완전히 동일한 정보 접근량의 비교로 본문에 쓰지 않는다.
+- **23쪽:** mismatch 진단의 native-depth scorer가 FT benchmark와 다르다. 그림 사이의 precision을 동일 기준의 before/after 수치로 직접 연결하지 않는다.
+
+## Figure 4 — 질환 신호 보존의 정량 비교
+
+**62 → 63 → 67 → 68 → 71쪽**을 중심으로 배치한다. 62에서는 질환 행을 사용한다. 63/67의 logFC correlation, bulk-controlled partial correlation, DEG를 나누고, five-donor와 five-fold 설계를 같은 축으로 표시하되 합쳐 평균하지 않는다. 다중 방문 cohort에서 five donors는 five samples와 다르다.
+
+**현재 수치는 “모든 질환·지표에서 최고”를 지지하지 않는다.** 예를 들어 five-donor Sepsis의 logFC r은 scSemi 0.4025 > LoRA 0.3501, five-fold Giroux도 scSemi 0.5373 > LoRA 0.4610이다. Sepsis의 deconvolution DEG overlap은 BayesPrism 27.0/CIBERSORTx 25.4 > LoRA 16.4다. 이 예외를 본문 그림에서 제거하지 않는다.
+
+기존 20개 five-donor draw에서는 LoRA가 scSemi보다 logFC r **13/20**, partial r **14/20**에서 높고, DEG는 **12승 + 8동률**이다. 단, support draws는 독립적인 환자 반복이 아니다. 현재 지지되는 결론은 **여러 질환 cohort에서 cell-type별 질병 신호를 보존하며 많은 비교에서 우수하지만, cohort와 endpoint별 예외가 있다**는 범위다.
+
+**68쪽에 빠진 scSemi five-fold CAT는 새 실험이 필요 없다.** 저장된 `scsemi_cv_20261002/pooled/cat.csv`의 `predictor=generated` 행에서 가져올 수 있다. CAT100의 LoRA/scSemi 값은 Sepsis 0.1625/0.0450, Imperial 0.2700/0.2250, Giroux 0.0675/0.0675, Ji 0.1525/0.0525다. Bulk-broadcast 행을 섞지 않는다.
+
+- **69–70쪽:** patient-aware corrected DEG의 보조 근거. 현재 덱은 Sepsis 5 draws/Imperial 1 draw의 고정 snapshot이며 모든 cohort의 완료판은 아니다.
+- **64–66, 110–115쪽:** cell realism/composition 등 생성 품질. Disease integrity의 직접 증거와 구분해 S07에 둔다.
+- **82–85, 120–122쪽:** rank, pair count, K=4,000, deconvolution coverage의 감도 분석은 S08.
+- **71쪽:** deconvolution이 지원하는 cell-type expression endpoint에서 비교한다. 공통 gene axis, 가용 draw 수와 NA를 표시한다.
+
+## Figure 5 — paired rare population의 정성적 재현
+
+현재 131장에는 **같은 held-out paired sample의 real/LoRA/scSemi rare-population identity를 직접 대조하는 완성 패널이 없다.** 89, 91–96쪽은 빈도·miss·상관·방향의 정량 자료다. 97/99는 pseudobulk 조건, 98/100/101은 주로 unmatched Behcet reference라서 이 목적을 대신하지 못한다. 77–80은 paired COVID 자료지만 zero-shot/calibration의 parent-population 진단이다.
+
+기존 출력으로 다음 패널을 조립할 수 있다. 아래 사례는 후보이며, 성공 여부를 보고 사례를 고르는 대신 관측 자료 기준의 선택 규칙을 먼저 고정한다.
+
+1. **5a:** 같은 held-out sample의 observed/LoRA/scSemi를 공통 observed-reference 공간에 나란히 표시하고, 사전에 정의한 rare state를 강조한다.
+2. **5b:** 동일 marker의 feature plot과 co-detection으로 population identity를 보여준다.
+3. **5c:** case/control donor별 small multiples로 반복성을 확인한다. Donor/visit 단위를 유지한다.
+4. **5d:** Sepsis MS1-like와 cycling monocyte 등 서로 다른 재현 양상을 함께 제시하고, Imperial cycling T/NK 등 독립 cohort로 확장한다.
+
+**이 공백이 곧 새 학습/생성이 필요하다는 뜻은 아니다.** 저장된 marker cache와 관측/생성 세포를 먼저 이용한다. Marker-space PCA/UMAP는 whole-transcriptome embedding과 다르며, 정의에 쓴 marker의 분리만으로 독립적인 identity 검증이 되지는 않는다. Bulk–sc pairing도 real/generated 개별 cell의 일대일 pairing을 의미하지 않는다. 여기서 “발견”은 우선 관측된 rare state의 회복/식별로 표현한다.
+
+기존 재료 위치(모두 Model 8 아래):
+
+- `plans/rare_states_paired_20261008/work/<cohort>/drawN/{lora5,scsemi5,zs_raw,zs_cal}.npz`
+- `plans/paired_finetune_intradataset_20260927/data/<cohort>/{cells_panel6k.npz,cells_obs.parquet,samples.csv,folds.csv}`
+- `plans/paired_finetune_intradataset_20260927/scoring_cache/<cohort>/observed_labels/labels/labels.parquet`
+- `plans/rare_states_screen_20261004/tools/states.json`
+- `plans/paper_figures_strictpairs_20261012/tools/{rare_cache_builder.py,export_rare_visuals.py}`
+
+확인 시점에 strict visual cache, coordinate CSV, 최종 qualitative PDF는 없었다. 전체 campaign exporter의 20-unit 완료 조건을 우회하지 않는다. 필요하면 검증된 subset의 범위를 명시한 별도 illustration으로 만든다.
+
+## Figure 6 — bulk-only 자료에서 기존 논문 결과 재현
+
+**TODO를 유지한다.** 재현할 논문의 특정 결과, adaptation pairs, bulk-only targets, 재현 endpoint와 독립적인 보조 근거를 먼저 고정한다. 104/106/107쪽은 설계 참고 자료이고 105쪽은 과거 탐색 기록이다. 기존 탐색의 “0 confirmed”는 특정 논문 결과 재현의 성공/실패를 뜻하지 않는다.
+
+Bulk-only sample에 sc partner가 없더라도 같은 donor의 다른 paired visit가 학습에 쓰였을 수 있으므로 자동으로 donor-held-out이라고 표기하지 않는다.
+
+## Supplementary 후보군
+
+다음S번호는**주제별분류용임시번호**다. 각그룹이그대로한장의SupplementaryFigure가된다는뜻은아니다.
+
+- **S01 — 모델·conditioning·학습·평가 설계 상세:** 8, 10, 17, 38, 39, 41쪽. Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분
+- **S02 — SLE pseudobulk 모델 검증과 대조군:** 12, 18, 72, 73, 74, 75쪽. 12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음
+- **S03 — paired 입력 mismatch와 calibration 상세:** 22, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37쪽. sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음
+- **S04 — 평가 단위·annotation·depth/readout 검증:** 40, 42, 43, 46, 47쪽. technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분
+- **S05 — healthy paired 재구성·FT mode·대표 샘플 대조:** 44, 45, 48, 57쪽. 48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence
+- **S06 — rank·pair budget·K·시간 상세:** 50, 56, 58, 59, 117, 118, 119쪽. Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지
+- **S07 — 질환/다중 cohort 생성 품질과 비교 안정성:** 64, 65, 66, 110, 111, 112, 113, 114, 115쪽. 질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음
+- **S08 — 환자 단위 DEG·질환 sensitivity·deconvolution/K4000:** 69, 70, 82, 83, 84, 85, 120, 121, 122쪽. 69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음
+- **S09 — 외부 질환 real-bulk diagnostic:** 76, 77, 78, 79, 80, 81쪽. LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리
+- **S10 — paired rare-state 정량 검증과 case 선택 근거:** 89, 91, 92, 93, 95, 96쪽. Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음
+- **S11 — pseudobulk rare screen·unmatched Behcet 한계:** 97, 98, 99, 100, 101쪽. Fig5의 독립 paired qualitative validation으로 사용하지 않음
+- **S12 — pretraining·cross-cohort·scratch 대조:** 123, 124, 125, 126, 127, 131쪽. 131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조
+- **S13 — scSemi output-threshold sensitivity:** 129쪽. 본문은 official output. 수정 readout은 별도 sensitivity로 표기
+- **S14 — bulk-only target cohort 차이:** 106쪽. Fig6 설계용 shift 점검; published finding의 재현 결과가 아님
+
+## 전체 페이지 처리
+
+| 분류 | source pages |
+|---|---:|
+| 본문 후보: 부분 추출·조건부 사용 포함 | 18 |
+| Supplementary 후보 | 86 |
+| Figure 6 설계/TODO: 104, 107쪽 | 2 |
+| 논문 증거로 사용 보류: 과거 rare scorer의 88, 90, 94쪽 | 3 |
+| 과거 workflow 기록: 105쪽 | 1 |
+| 표지·구분·이전 계획·읽기 안내 | 21 |
+| **합계** | **131** |
+
+분류는 131장 모두에 하나의 주배정을 부여하며 부분 재사용은 별도 열에 적었다. 원본 덱의 축소·삭제·재정렬은 하지 않았다. 새 학습, inference, scoring 또는 통계 검정도 실행하지 않았다.
+
+## 주요 근거 파일
+
+`data/`는 이 덱 폴더 아래, `plans/`는 Model 8 아래 경로다.
+
+- `data/expanded_master_20261013/spec.json`: 현재 131장 내용과 protocol.
+- `data/sec07_rank_sweep.csv`, `data/sec07_rank_sweep_runs.csv`, `data/sec07_ft_modes_accuracy.csv`: healthy paired rank/FT.
+- `plans/paper_figures_strictpairs_20261012/source_data/fig2_historical_rank_full.csv`: full FT/rank summary.
+- `data/fig4_disease.csv`, `data/sec09_cat_curves.csv`, `data/secS_deconvolution.csv`: 질환 비교.
+- `plans/paired_finetune_intradataset_20260927/scores_contrast/<cohort>/scsemi_cv_20261002/pooled/cat.csv`: 빠진 CV-CAT를 채울 기존 결과.

paper-releases/pbisc-diffusion/updates/20261013T044744Z-figure1to6-131page-classification/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/figure_assignment_20261013/AUTHOR_SCOPE.md

@@ -0,0 +1,17 @@
+# Author-defined Figure 1–6 scope — 2026-10-13
+
+The user's new definitions supersede the historical section labels and planning cards inside the 131-slide evidence master. Classification does not delete or shorten that master.
+
+1. **Figure 1**: complete experimental overview and statistical charts describing the corpus.
+2. **Figure 2**: architecture and gene-context/panel ablations, using SLE pseudobulk benchmark data. This is not a measured bulk–sc paired benchmark.
+3. **Figure 3**: measured-bulk versus pseudobulk mismatch; healthy paired-data full fine-tuning improvements; comparison with LoRA ranks 4/16/64; paired-support ablation starting from five. Every performance result here must use measured bulk–sc paired data. The author wants to motivate rank16; the audit must distinguish that goal from what the measured rankings establish.
+4. **Figure 4**: non-healthy disease bulk–sc paired cohorts; disease-integrity benchmarks such as logFC correlation, partial correlation and DEG agreement; compare five-fold and LoRA-five designs with existing methods. This is the central quantitative contribution.
+5. **Figure 5**: qualitative identification/reconstruction of rare populations on bulk–sc paired datasets, complementing Figure4's quantitative evidence. Aggregate frequency/rank tables alone do not establish population identity.
+6. **Figure 6**: adaptation on paired data followed by inference on bulk-only samples, reproducing a prespecified finding from an existing paper. **TODO**.
+
+## Binding evidence constraints
+
+- These are the author-approved scopes. `ASSIGNMENT.md` and the per-slide CSV are proposed placements within those scopes.
+- Do not call rank16 universally best or disease integrity universally superior unless the corresponding data support it. Existing counterexamples must remain visible.
+- Separate genuine exact-five support from n training plus extra validation, and five donors from five samples in multi-visit cohorts.
+- New visualization of existing generated cells is distinct from new model inference. No new experiments are authorized by this classification task.

paper-releases/pbisc-diffusion/updates/20261013T044744Z-figure1to6-131page-classification/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/figure_assignment_20261013/slide_assignment.csv

@@ -0,0 +1,132 @@
+page_131_deck,original_76_page,title,target,use,role,data_regime,required_action,secondary_target,figure_files,current_protocol_note
+1,1,PBISC-Diffusion: expanded evidence deck,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+2,2,Full evidence map,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+3,,How to read this expanded evidence deck,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,Original figures retain their data and evidence lineage; sources and limitations are in the speaker notes.
+4,3,Manuscript plan: six main figures,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_overview.png,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+5,,Availability is not the same as a new experiment,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,"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,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+7,5,Figure 1 plan: Design and performance of PBISC,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_fig1.png,"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,Supp S01,SUPPLEMENT,모델·conditioning·학습·평가 설계 상세,개념/방법 설명,Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분,,figures/sec02_backbone_compare.png,"SDD: Ostheimer et al., arXiv 2502.02448. scDiffusion: Luo et al., Bioinformatics 2024."
+9,7,Overview draft (Nature style),Figure 1,MAIN,전체 pretraining → adaptation → generation 개요,개념도,사용자 정의에 맞게 전체 실험 흐름의 중심 패널로 사용,,figures/expanded_master_20261013/palette/fig1b_overview_nature.png,Schematic. Conditioning recipe size and the number of generated cells are distinct; no cell-type quotas.
+10,8,Bulk conditioning,Supp S01,SUPPLEMENT,모델·conditioning·학습·평가 설계 상세,개념/방법 설명,Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분,,figures/sec02_bulk_conditioning.png,Final 6k model. Recipe size K conditions the bulk; the number of generated cells is chosen separately.
+11,9,Pretraining corpus,Figure 1,MAIN,corpus 규모·assay·condition·cell-type 통계,pretraining corpus 통계,cell·donor·dataset 분모와 학습 노출량을 구분,,figures/fig1d_corpus.png,Corpus labels describe the data; they are not model inputs. Dataset composition is not sampler exposure.
+12,10,Width & context ablation: loss saturation,Supp S02,SUPPLEMENT,SLE pseudobulk 모델 검증과 대조군,pretraining validation 또는 SLE pseudobulk,"12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",,figures/expanded_master_20261013/palette/sec03_width_loss.png,Old 6k panel (1% floor); one validation set. 16x: memory-floor stop after part 2.
+13,11,Width & context ablation: benchmark,Figure 2,MAIN,architecture·gene-panel ablation의 SLE pseudobulk 성능,SLE pseudobulk development benchmark,architecture 1% floor와 panel 0.1% floor를 분리; paired measured-bulk 결과로 표기하지 않음,,figures/expanded_master_20261013/palette/sec03_width_bench.png,"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,Figure 2,MAIN,architecture·gene-panel ablation의 SLE pseudobulk 성능,SLE pseudobulk development benchmark,architecture 1% floor와 panel 0.1% floor를 분리; paired measured-bulk 결과로 표기하지 않음,,figures/expanded_master_20261013/palette/sec03_panel.png,Existing panel benchmark. 6k: five sampling seeds; other panels: one seed. Different axis from the architecture sweep.
+15,,Native-pseudobulk model comparisons,Figure 2,MAIN,architecture·gene-panel ablation의 SLE pseudobulk 성능,SLE pseudobulk development benchmark,architecture 1% floor와 panel 0.1% floor를 분리; paired measured-bulk 결과로 표기하지 않음,,figures/expanded_master_20261013/palette/review_style_20261013_ablation.png,Completed held-out rescoring. Architecture: 1% detection floor; gene-panel comparison: 0.1% floor.
+16,13,"PBISC-Diffusion: bulk in, cells out",Figure 1,MAIN_EXTRACT,corpus별 규모와 recipe-size 분포,개념도 + corpus 통계,하단 corpus/K 통계만 추출; 상단 workflow는 9쪽과 중복되면 보충으로 이동,Supp S01,figures/sec01_slide_a.png,"Fig. 1b redraw. Final model corpus: 33.1M train cells; K = 1–4,000 cells per pseudobulk."
+17,14,Denoiser: joint value + support diffusion,Supp S01,SUPPLEMENT,모델·conditioning·학습·평가 설계 상세,개념/방법 설명,Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분,,figures/sec01_fig1a_denoiser.png,"Fig. 1a redraw. Final model f01 6k 1x, epoch 16, EMA; 50 DDIM steps."
+18,,Does the model follow its bulk conditioning?,Supp S02,SUPPLEMENT,SLE pseudobulk 모델 검증과 대조군,pretraining validation 또는 SLE pseudobulk,"12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",,figures/expanded_master_20261013/healthy/sle_conditioning_interventions.png,"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,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+20,16,Figure 2 plan: Fine-tuning PBISC with LoRA,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_fig2.png,"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),Figure 3,MAIN,healthy real-bulk / pseudobulk 입력 mismatch,ImMicro measured bulk–sc paired,shift removal은 paired pseudobulk를 쓴 진단임을 명시,,figures/sec05_condition_space.png,Existing paired-input diagnostic; mean-shift removal uses paired pseudobulks and is not a deployment method.
+22,18,Same shift in three paired cohorts,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/sec05_shift_cohorts.png,"Model inputs only, no generation. PCA fitted per cohort on sc pseudobulks."
+23,19,Consequence: cells off-manifold,Figure 3,MAIN_ILLUSTRATION,mismatch가 생성 세포 공간에 미치는 영향,ImMicro paired 입력의 native-depth zero-shot 진단,FT 이후 그림으로 오해하지 않게 표시; 다른 scorer의 before/after 수치를 직접 연결하지 않음,,figures/sec05_cells_umap.png | figures/expanded_master_20261013/palette/sec05_cells_metrics.png,Existing same-noise input diagnostic; not the new LoRA/scSemi OOF embedding. UMAP overlap is not a quantitative score.
+24,,Input corrections on one real-reference UMAP,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/stored_real_reference_umap.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+25,,Where generated-cell density differs from real cells,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/saved_density_difference.png,Original high-resolution density map; zoom for panel labels. Quantitative colors preserved. Historical native-depth controls.
+26,,Input calibration changes cell-type composition,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/input_control_composition.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+27,,"Same noise, different input: matched-cell shifts",Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/stored_matched_trajectories.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+28,,Matched noise separates cell state from cell identity,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/matched_similarity.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+29,,Input changes can switch the generated cell label,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/label_transitions.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+30,,Nearest real cells: distance and label agreement,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/nearest_real_by_type.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+31,,Composition agreement does not ensure state agreement,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/within_type_centroids.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+32,,Gene-length effects in the bulk input,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/gene_length_input.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+33,,Gene-length effects within label-stable cells,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/gene_length_stable.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+34,,Raw and TPM shifts persist within cell labels,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/raw_tpm_top_genes.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+35,,Reference calibration leaves specific gene-state shifts,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/refcal_top_genes.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+36,,Donor trait axes and donor identification differ,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/donor_axes.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+37,,Reference choice changes different quality endpoints,Supp S03,SUPPLEMENT,paired 입력 mismatch와 calibration 상세,paired input diagnostics; 다수 native-depth 조건,"sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",,figures/expanded_master_20261013/healthy/reference_choice.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+38,20,Fine-tuning with paired bulk + sc,Supp S01,SUPPLEMENT,모델·conditioning·학습·평가 설계 상세,개념/방법 설명,Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분,,figures/expanded_master_20261013/palette/sec06_ft_method.png,Base: final model f01 6k 1x E16. Times from 40 LoRA runs on this host.
+39,21,Evaluation design,Supp S01,SUPPLEMENT,모델·conditioning·학습·평가 설계 상세,개념/방법 설명,Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분,,figures/sec06_eval_design.png,Bars: Ye 2026 sepsis. QC-failed pairs excluded.
+40,,Why the historical and strict protocols differ,Supp S04,SUPPLEMENT,평가 단위·annotation·depth/readout 검증,프로토콜/observed controls/native-depth 진단,"technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",,,Support draws overlap in donors. Displayed draw variation is not independent population replication.
+41,22,How we score generated cells,Supp S01,SUPPLEMENT,모델·conditioning·학습·평가 설계 상세,개념/방법 설명,Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분,,figures/sec04_metrics_overview.png,"Definitions follow the cited papers; scorers: tools/score_ft.py, extra_metrics.py, contrast_extra.py."
+42,,Technical repeats and follow-up are different controls,Supp S04,SUPPLEMENT,평가 단위·annotation·depth/readout 검증,프로토콜/observed controls/native-depth 진단,"technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",,figures/expanded_master_20261013/healthy/immicro_replicate_controls.png,Observed ImMicro controls. Technical remeasurement and follow-up: 16 donors each. Follow-up is not a technical replicate.
+43,,Annotation agreement and difficult categories,Supp S04,SUPPLEMENT,평가 단위·annotation·depth/readout 검증,프로토콜/observed controls/native-depth 진단,"technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",,figures/expanded_master_20261013/healthy/annotation_agreement.png,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,Supp S05,SUPPLEMENT,healthy paired 재구성·FT mode·대표 샘플 대조,ImMicro measured bulk–sc paired,48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence,,figures/expanded_master_20261013/palette/fig2_healthy_cells.png,"ImMicro, 107 healthy donors. 5 donors: mean of 5 draws (dots); 5-fold CV shown separately."
+45,24,Healthy cohort: realism beyond precision/coverage,Supp S05,SUPPLEMENT,healthy paired 재구성·FT mode·대표 샘플 대조,ImMicro measured bulk–sc paired,48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence,,figures/expanded_master_20261013/palette/fig2_healthy_realism.png,"ImMicro, 107 healthy donors. Dashed = real vs real cells of the same donors."
+46,,Depth matching changes precision and coverage,Supp S04,SUPPLEMENT,평가 단위·annotation·depth/readout 검증,프로토콜/observed controls/native-depth 진단,"technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",,figures/expanded_master_20261013/healthy/depth_matched_quality.png,Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed.
+47,,Detection and library size are separate diagnostics,Supp S04,SUPPLEMENT,평가 단위·annotation·depth/readout 검증,프로토콜/observed controls/native-depth 진단,"technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",,figures/expanded_master_20261013/healthy/detection_and_library.png,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,Supp S05,SUPPLEMENT,healthy paired 재구성·FT mode·대표 샘플 대조,ImMicro measured bulk–sc paired,48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence,,figures/expanded_master_20261013/palette/fig2_healthy_modes.png,"ImMicro, 107 healthy donors; cond.-encoder-only ran on fold 0 (22 donors) only."
+49,26,Healthy cohort: how many pairs,Figure 3,MAIN_CONDITIONAL,LoRA training-pair 수에 따른 학습곡선,ImMicro paired; n training + 7 validation,학습 n과 총 paired 접근량을 구분. strict total-n 주장에는 그대로 사용 불가; 정확히5쌍은 54쪽으로 입증,,figures/expanded_master_20261013/palette/fig2_healthy_pairs.png,Historical n + 7 validation design. The fixed-3000-step strict curve is a separate protocol update.
+50,27,Healthy cohort: LoRA rank and K,Supp S06,SUPPLEMENT,rank·pair budget·K·시간 상세,healthy 또는 mixed paired cohorts,Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지,,figures/expanded_master_20261013/palette/fig2_healthy_rank.png,"ImMicro, 107 donors, 5-fold CV pooled. Open symbols: K choices with every training pair."
+51,,Healthy bulk: full fine-tuning and LoRA rank,Figure 3,MAIN,full FT와 LoRA rank의 품질·파라미터 절충,ImMicro paired; historical pooled five-fold,r16을 모든 지표 최적으로 쓰지 않음. r4/r16/r64와 full FT를 함께 표시,Supp S06,figures/expanded_master_20261013/palette/review_style_20261013_rank.png,Original donor-level five-fold comparison. LoRA rank 16 is a quality/parameter compromise.
+52,28,Figure 3. PBISC vs scSemiProfiler on healthy donors,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+53,29,Figure 3 plan: PBISC vs scSemiProfiler on healthy donors,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_fig3.png,"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",Figure 3,MAIN,실제 exact-five adaptation 성능과 scSemi 비교,"ImMicro measured bulk–sc paired; 5 pairs, no extra validation, fixed3000",5개 support draw를 표시; 남은 102 donors/draw. 반복 draw를 독립 환자 반복으로 세지 않음,,figures/expanded_master_20261013/palette/fig3_healthy_fewpair.png,"ImMicro: exactly 5 pairs, no extra validation, 3000-step LoRA; five draws, remaining 102 donors per draw."
+55,,Composition: error and donor-level wins,Figure 3,MAIN_OPTIONAL,exact-five의 cell-type composition 오차·donor별 우위,ImMicro measured bulk–sc paired,54쪽 보완 패널 또는 Supp S05로 배치. per-draw 절대오차 후 donor 평균 정의 유지,Supp S05,figures/expanded_master_20261013/palette/review_style_20261013_healthy_composition.png,"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,Supp S06,SUPPLEMENT,rank·pair budget·K·시간 상세,healthy 또는 mixed paired cohorts,Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지,,figures/expanded_master_20261013/palette/fig3_healthy_pairs.png,"LoRA: n training + 7 validation. scSemi: n + 7 representatives, with CV-HVG reuse. Not the strict total-n protocol."
+57,32,Representative-dependent composition,Supp S05,SUPPLEMENT,healthy paired 재구성·FT mode·대표 샘플 대조,ImMicro measured bulk–sc paired,48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence,,figures/expanded_master_20261013/palette/fig3_healthy_copy.png,"Existing healthy five-draw diagnostic. This evaluates composition dependence, not literal copying of every cell."
+58,33,"Time: days for scSemiProfiler, minutes for LoRA",Supp S06,SUPPLEMENT,rank·pair budget·K·시간 상세,healthy 또는 mixed paired cohorts,Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지,,figures/expanded_master_20261013/palette/fig3_time_days.png,"One GB10 GPU, job alone on it. GPU time from receipts; waits for the GPU excluded."
+59,34,5 donors: time to apply,Supp S06,SUPPLEMENT,rank·pair budget·K·시간 상세,healthy 또는 mixed paired cohorts,Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지,,figures/expanded_master_20261013/palette/sec09_time_to_apply.png,"Existing measurements: adaptation + target generation; pretraining, GPU wait and scoring excluded."
+60,35,Figure 4. Disease signal on paired cohorts,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+61,36,Figure 4 plan: Disease signal on paired cohorts,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_fig4.png,"Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+62,37,Five paired cohorts,Figure 4,MAIN_EXTRACT,질환 paired cohorts와 sample/donor 단위,ImMicro + 질환 paired cohort metadata,Figure4에서는 질환 행만 사용; 전체 cohort 지도는 Figure1 선택 inset 가능,Figure 1,figures/sec09_cohorts.png,"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,Figure 4,MAIN_REANALYSIS,five-donor LoRA와 기존 방법의 disease integrity,질환 measured bulk–sc paired; historical five-donor draws,multi-visit의 five donors ≠ five samples. cohort별 예외와 동일한 donor-aware estimator 경계를 표시,,figures/expanded_master_20261013/palette/fig4_disease_fewpair.png,Existing five-donor protocol; sample counts differ by cohort. Original sample-level estimator; corrected update separate.
+64,,Sepsis: exact-five reconstruction across five draws,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/disease/strict_sepsis_quality.png,Completed exact-five Sepsis draws. Bars = means; dots = support draws. Five matched methods; no new generation.
+65,,Sepsis: realism beyond precision and coverage,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/disease/strict_sepsis_realism.png,"Completed exact-five outputs; raw readout, fixed R1 reference. Real/real control shown. Dots are five support draws."
+66,,Imperial: first completed exact-five comparison,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/disease/strict_imperial_draw0_quality.png,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,Figure 4,MAIN_REANALYSIS,five-fold held-out disease integrity,질환 measured bulk–sc paired; pooled five-fold,63쪽과 design을 분리; 같은 평가축/estimators로 기존 방법을 비교,,figures/expanded_master_20261013/palette/fig4_disease_fivefold.png,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),Figure 4,MAIN_REPLOT,gene-list 길이에 따른 CAT 비교,질환 measured bulk–sc paired; five-donor + five-fold,이미 저장된 scSemi five-fold CAT를 추가. generated predictor 행만 사용하고 bulk-broadcast 행 제외,,figures/expanded_master_20261013/palette/sec09_cat_curves.png,Existing CAT curves; ranking agreement and chance line. Patient-aware reanalysis reuses the same predictions.
+69,,Sepsis: patient-aware disease-gene agreement,Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/disease/strict_ye2026_sepsis_deg.png,Completed corrected count-based DEG fits; mean of five support draws.
+70,,Imperial: patient-aware disease-gene agreement,Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/disease/strict_imperial_eskd_deg.png,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",Figure 4,MAIN_QUALIFIED,BayesPrism/CIBERSORTx 대비 cell-type disease expression,질환 measured bulk–sc paired; supported expression endpoints,동일 support/공통 gene axis/가용 draw 수를 명시. 미계산 상관은0으로 처리하지 않고 DEG의 비우위도 유지,,figures/expanded_master_20261013/palette/secS_deconv_disease.png,"Completed BayesPrism/CIBERSORTx comparisons; same five-donor support, method-supported expression metrics."
+72,42,SLE benchmark: PBISC vs scSemiProfiler,Supp S02,SUPPLEMENT,SLE pseudobulk 모델 검증과 대조군,pretraining validation 또는 SLE pseudobulk,"12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",,figures/expanded_master_20261013/palette/sec04_sle_benchmark.png,"Completed pseudobulk benchmark; separate from external paired real-bulk validation. PBISC five seeds, scSemi one run."
+73,43,SLE benchmark: per cell type,Supp S02,SUPPLEMENT,SLE pseudobulk 모델 검증과 대조군,pretraining validation 또는 SLE pseudobulk,"12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",,figures/expanded_master_20261013/palette/sec04_sle_per_celltype.png,"Same 42 donors, 1,006 genes. PBISC: 5 seeds, mean ± SD; scSemiProfiler: 1 run. Other: top-50 of 12."
+74,,Disjoint-cell controls test the bulk-broadcast baseline,Supp S02,SUPPLEMENT,SLE pseudobulk 모델 검증과 대조군,pretraining validation 또는 SLE pseudobulk,"12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",,figures/expanded_master_20261013/healthy/sle_disjoint_truth_control.png,SLE pseudobulk benchmark; five existing half-cell splits. Error bars are saved split SDs. No new cell splitting or scoring.
+75,,SLE: what is added beyond the bulk broadcast?,Supp S02,SUPPLEMENT,SLE pseudobulk 모델 검증과 대조군,pretraining validation 또는 SLE pseudobulk,"12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",,figures/expanded_master_20261013/palette/sec04_sle_broadcast.png,Existing SLE pseudobulk development benchmark; bulk broadcast is an explicit comparator. Separate from paired real-bulk validation.
+76,,Influenza: response order versus response genes,Supp S09,SUPPLEMENT,외부 질환 real-bulk diagnostic,influenza/COVID historical zero-shot/calibration,LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리,,figures/expanded_master_20261013/palette/fig_randolph_celltype_response.png,Historical zero-shot influenza test. Response-strength ordering is recovered; gene-wise agreement remains below the matched bulk broadcast.
+77,,COVID: calibration on a shared real-cell map,Supp S09,SUPPLEMENT,외부 질환 real-bulk diagnostic,influenza/COVID historical zero-shot/calibration,LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/refcal_scoring/arunachalam/placement/figures/fig1_umap_seven_panels_by_label_depth1000.png,"Historical 7 COVID / 5 HC samples, common depth. High-resolution reference overview; following detail panels provide readable quantitative results."
+78,,COVID monocytes: direction after depth matching,Supp S09,SUPPLEMENT,외부 질환 real-bulk diagnostic,influenza/COVID historical zero-shot/calibration,LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리,,/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/refcal_scoring/arunachalam/placement/figures/fig2_covid_hc_monocyte_arrows_depth1000.png,"Historical 7 COVID / 5 HC samples, common depth. High-resolution reference overview; following panels separate direction, magnitude and relabeling null."
+79,,"COVID monocytes: direction, magnitude and null",Supp S09,SUPPLEMENT,외부 질환 real-bulk diagnostic,influenza/COVID historical zero-shot/calibration,LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리,,figures/expanded_master_20261013/disease/covid_monocyte_direction_detail.png,Readable detail of the retained COVID overview. Existing 7 COVID / 5 HC paired samples; native and depth-matched results stay separate.
+80,,"COVID CD14 monocytes: direction, magnitude and null",Supp S09,SUPPLEMENT,외부 질환 real-bulk diagnostic,influenza/COVID historical zero-shot/calibration,LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리,,figures/expanded_master_20261013/disease/covid_cd14_direction_detail.png,Readable detail of the retained COVID overview. Existing 7 COVID / 5 HC paired samples; native and depth-matched results stay separate.
+81,,COVID: biological endpoints give a mixed verdict,Supp S09,SUPPLEMENT,외부 질환 real-bulk diagnostic,influenza/COVID historical zero-shot/calibration,LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리,,figures/expanded_master_20261013/disease/covid_endpoints.png,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,Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/palette/fig4_disease_variants.png,"Sepsis, 82 donors, pooled 5-fold. LoRA r = 16 is the model used throughout."
+83,45,Pairs needed: donor and disease signal,Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/palette/sec08_pairs_signal.png,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",Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/palette/secS_deconv_composition.png,Completed deconvolution outputs already exist. Verified identical Sepsis supports use adoption instead of method reruns.
+85,,Deconvolution: expression coverage limits the comparison,Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/disease/deconv_expression_coverage.png,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,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+87,48,Figure 5 plan: Rare cell states in disease cohorts,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_fig5.png,"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 reference,HOLD,기존 rare-state scoring 기록,historical paired rare analysis,zero-drop/sample-unit 문제가 있어 현재 manuscript 본문/정식 supplementary에서는 사용 보류; 수정된5donor결과는 S10,,figures/expanded_master_20261013/palette/fig5_rare_fraction.png,Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate.
+89,,Rare-state frequencies in observed cells,Supp S10,SUPPLEMENT,paired rare-state 정량 검증과 case 선택 근거,paired disease rare-state quantitative,Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음,,figures/review_style_20261013_rare_observed.png,Original five-donor outputs. Median across draws; state share within its parent population. Controls included.
+90,50,Rare-state ranking: original analysis,Historical reference,HOLD,기존 rare-state scoring 기록,historical paired rare analysis,zero-drop/sample-unit 문제가 있어 현재 manuscript 본문/정식 supplementary에서는 사용 보류; 수정된5donor결과는 S10,,figures/expanded_master_20261013/legacy_rare/fig5_rare_donors.png,Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate.
+91,,Rare-state misses: rules 1–8,Supp S10,SUPPLEMENT,paired rare-state 정량 검증과 case 선택 근거,paired disease rare-state quantitative,Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음,,figures/expanded_master_20261013/disease/rare_misses_1.png,"Saved zero-aware diagnostic; all 15 predefined rules are shown across two slides. NA means not evaluable, not zero error."
+92,,Rare-state misses: rules 9–15,Supp S10,SUPPLEMENT,paired rare-state 정량 검증과 case 선택 근거,paired disease rare-state quantitative,Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음,,figures/expanded_master_20261013/disease/rare_misses_2.png,"Saved zero-aware diagnostic; all 15 predefined rules are shown across two slides. NA means not evaluable, not zero error."
+93,,Rare-state fidelity: zero-aware reanalysis,Supp S10,SUPPLEMENT,paired rare-state 정량 검증과 case 선택 근거,paired disease rare-state quantitative,Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음,,figures/review_style_20261013_rare_corrected_comparison.png,"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 reference,HOLD,기존 rare-state scoring 기록,historical paired rare analysis,zero-drop/sample-unit 문제가 있어 현재 manuscript 본문/정식 supplementary에서는 사용 보류; 수정된5donor결과는 S10,,figures/expanded_master_20261013/palette/fig5_rare_fivefold.png,Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate.
+95,,Sepsis rare states across paired-data budgets,Supp S10,SUPPLEMENT,paired rare-state 정량 검증과 case 선택 근거,paired disease rare-state quantitative,Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음,,figures/expanded_master_20261013/disease/sepsis_rare_support.png,"Historical cross-fitted learning curve: n training pairs plus validation, not strict total n. All 15 rules; arm-specific depth matching."
+96,,Sepsis rare-state direction across support budgets,Supp S10,SUPPLEMENT,paired rare-state 정량 검증과 case 선택 근거,paired disease rare-state quantitative,Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음,,figures/expanded_master_20261013/disease/sepsis_rare_direction.png,"Saved sign-agreement counts across all 15 rules, including controls. Historical n + validation; directions are not effect-size accuracy."
+97,,"Rare-state screen in DRESS, Kawasaki and SLE",Supp S11,SUPPLEMENT,pseudobulk rare screen·unmatched Behcet 한계,own-sc pseudobulk 또는 unmatched observed reference,Fig5의 독립 paired qualitative validation으로 사용하지 않음,,figures/expanded_master_20261013/disease/external_rare_screen.png,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,Supp S11,SUPPLEMENT,pseudobulk rare screen·unmatched Behcet 한계,own-sc pseudobulk 또는 unmatched observed reference,Fig5의 독립 paired qualitative validation으로 사용하지 않음,,figures/expanded_master_20261013/palette/sec05_behcet_direction.png,Historical Behcet diagnostic; reference is unmatched and includes training donors. This is not independent patient-level subtype validation.
+99,,C1Q: single-gene detection versus co-detection,Supp S11,SUPPLEMENT,pseudobulk rare screen·unmatched Behcet 한계,own-sc pseudobulk 또는 unmatched observed reference,Fig5의 독립 paired qualitative validation으로 사용하지 않음,,figures/expanded_master_20261013/disease/c1q_cross_disease.png,"Existing pseudobulk-conditioned comparison at matched depth. Multi-gene C1Q rates remain compressed, including in-corpus DRESS."
+100,,"Behçet: K=4,000 changes direction more than identity",Supp S11,SUPPLEMENT,pseudobulk rare screen·unmatched Behcet 한계,own-sc pseudobulk 또는 unmatched observed reference,Fig5의 독립 paired qualitative validation으로 사용하지 않음,,figures/expanded_master_20261013/disease/behcet_k4000.png,"Two existing sampling seeds; 6 BD / 10 HC held-out bulks. K is condition size; each bulk still generates 5,000 cells. Descriptive."
+101,,Behcet: one-gene and multi-gene C1Q rules,Supp S11,SUPPLEMENT,pseudobulk rare screen·unmatched Behcet 한계,own-sc pseudobulk 또는 unmatched observed reference,Fig5의 독립 paired qualitative validation으로 사용하지 않음,,figures/expanded_master_20261013/palette/review_style_20261013_behcet.png,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,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+103,54,Figure 6 plan: Discovery from public bulk-only data,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_fig6.png,"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,Figure 6,TODO_SETUP,adaptation / bulk-only cohort 설계 후보,과거 cohort inventory,새 재현 대상 논문·cohort가 확정된 후 다시 작성; 현재 수치를 최종 재현 cohort로 간주하지 않음,,figures/sec10_paired_subset_cohorts.png,Historical cohort counts. Existing exploratory outputs are reviewed separately; reproduction with the user-prepared new dataset remains pending.
+105,,Historical bulk-only search: no confirmed new finding,Figure 6 reference,WORKFLOW_REFERENCE,과거 bulk-only 탐색·검토 기록,historical exploratory candidate audit,0 confirmed는 특정 기존 논문 재현의 성공/실패를 뜻하지 않음; main Figure6 결과로 사용 금지,,figures/expanded_master_20261013/disease/bulkonly_verification_history.png,Historical exploratory analysis of existing bulk-only generations. Review dispositions are not experimental validation. Figure 6 reproduction still awaits new data.
+106,,Bulk-only exploration: calibration populations differ,Supp S14,SUPPLEMENT,bulk-only target cohort 차이,historical paired vs bulk-only population metadata,Fig6 설계용 shift 점검; published finding의 재현 결과가 아님,,figures/expanded_master_20261013/disease/bulkonly_imperial_wave_shift.png,Historical Imperial sample composition. Paired and bulk-only sets differ in wave; repeated samples are not independent donors.
+107,,What remains to establish Figure 6,Figure 6,TODO_PLAN,기존 논문 결과 재현의 검증 요건,계획,실제 재현 결과 패널은 현재 없음,,,Historical candidate analyses do not substitute for independent reproduction in the new cohort.
+108,56,Supplementary,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+109,57,Supplementary plan,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,figures/figplan_supp.png,"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,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/palette/sec09_fewpair_cells.png,Completed five-donor comparisons. Multi-visit cohorts used more than five paired samples.
+111,59,5 donors: realism beyond precision/coverage,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/palette/sec09_extra_fewpair.png,"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",Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/palette/sec09_extra_ksens.png,"Completed sensitivity at k=3,5,10; no new sensitivity campaign required."
+113,61,5-fold: realism beyond precision/coverage,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/palette/sec09_extra_fivefold.png,"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,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/palette/sec09_same_pairs.png,Completed original five-fold LoRA/scSemi comparison; not the new total-n curve.
+115,63,scSemiProfiler and the representative-copy baseline,Supp S07,SUPPLEMENT,질환/다중 cohort 생성 품질과 비교 안정성,paired cell-level/realism/composition,질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음,,figures/expanded_master_20261013/palette/sec09_scsemi_representative.png,Copy = composition of the one training sample scSemiProfiler assigns to each test sample.
+116,64,Full vs LoRA vs cond.-encoder-only,Figure 3,MAIN_EXTRACT,raw/calibrated baseline 대비 full FT 개선,ImMicro + Sepsis paired; pooled와 fold0 pilot 혼재,"dataset=immicro, call=pooled만 추출. Sepsis와 condition-encoder fold0 pilot을 본문 healthy 비교에 혼합하지 않음",Supp S05,figures/expanded_master_20261013/palette/sec07_ft_modes.png,"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",Supp S06,SUPPLEMENT,rank·pair budget·K·시간 상세,healthy 또는 mixed paired cohorts,Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지,,figures/expanded_master_20261013/palette/sec08_pairs_realism.png,"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",Supp S06,SUPPLEMENT,rank·pair budget·K·시간 상세,healthy 또는 mixed paired cohorts,Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지,,figures/expanded_master_20261013/palette/sec07_rank_sweep.png,"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,Supp S06,SUPPLEMENT,rank·pair budget·K·시간 상세,healthy 또는 mixed paired cohorts,Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지,,figures/expanded_master_20261013/palette/sec07_k_aug.png,"LoRA r16, 5-fold pooled. p = paired Wilcoxon on donor L1, uncorrected."
+120,,"K=4,000 recipe: paired reconstruction quality",Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/disease/k4000_quality.png,"Existing paired benchmark, five support draws per cohort and 1,500 cells/test sample. Bars = means; dots = draws. No new runs."
+121,,"K=4,000 recipe: disease signal does not improve uniformly",Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/disease/k4000_disease.png,"Existing paired benchmark, five support draws per cohort and 1,500 cells/test sample. Bars = means; dots = draws. No new runs."
+122,,"K=4,000 recipe: calibration gains are selective",Supp S08,SUPPLEMENT,환자 단위 DEG·질환 sensitivity·deconvolution/K4000,disease paired; 일부 mixed/partial,"69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",,figures/expanded_master_20261013/disease/k4000_calibration.png,"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,Supp S12,SUPPLEMENT,pretraining·cross-cohort·scratch 대조,paired controls; 일부 fold0/one-seed pilot,131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조,,figures/expanded_master_20261013/palette/sec08_cross_cohort.png,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,Supp S12,SUPPLEMENT,pretraining·cross-cohort·scratch 대조,paired controls; 일부 fold0/one-seed pilot,131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조,,figures/expanded_master_20261013/palette/sec08_cross_cohort_steps.png,Existing saturation experiment; no additional saturation run introduced here.
+125,70,"No pretraining: same cohort, from scratch",Supp S12,SUPPLEMENT,pretraining·cross-cohort·scratch 대조,paired controls; 일부 fold0/one-seed pilot,131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조,,figures/expanded_master_20261013/palette/sec07_no_pretraining.png,"Completed Sepsis fold0 pilot, not all folds. Pretraining supports speed/few-pair adaptation, not universal superiority."
+126,71,"No pretraining: other cohorts, from scratch",Supp S12,SUPPLEMENT,pretraining·cross-cohort·scratch 대조,paired controls; 일부 fold0/one-seed pilot,131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조,,figures/expanded_master_20261013/palette/sec08_cross_scratch.png,"Pilot: held-out sepsis, 82 samples. Genes no training cohort measured are zeroed in every arm."
+127,72,Rare states: from scratch vs pretrained,Supp S12,SUPPLEMENT,pretraining·cross-cohort·scratch 대조,paired controls; 일부 fold0/one-seed pilot,131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조,,figures/expanded_master_20261013/palette/secS_scratch_rare_states.png,"Pilot, one seed: sepsis fold 0 (17 donors), held-out sepsis (82, masked). 1 = observed rate; grey = within 2-fold."
+128,73,Appendix,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+129,74,Appendix: scSemiProfiler without output threshold,Supp S13,SUPPLEMENT,scSemi output-threshold sensitivity,paired comparative readout sensitivity,본문은 official output. 수정 readout은 별도 sensitivity로 표기,,figures/expanded_master_20261013/palette/secA_scsemi_readout.png,Completed modified-readout check; main results use official output. Models were not retrained for the sensitivity.
+130,75,Discussion,Deck / planning,ADMIN,표지·목차·구분·이전 계획·읽기 안내,non-result material,원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체,,,
+131,76,Next: other tissues from a few dozen pairs,Supp S12,SUPPLEMENT,pretraining·cross-cohort·scratch 대조,paired controls; 일부 fold0/one-seed pilot,131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조,,figures/expanded_master_20261013/palette/sec10_other_tissues.png,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/20261013T044744Z-figure1to6-131page-classification/diffusion/dataset/server_1/model8/presentations/2026-09-30--paper-summary--r00/data/figure_assignment_20261013/slide_assignment.json

@@ -0,0 +1,1713 @@
+{
+  "schema": "pbisc.figure-assignment.v1",
+  "date": "2026-10-13",
+  "source_spec": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/presentations/2026-09-30--paper-summary--r00/data/expanded_master_20261013/spec.json",
+  "source_spec_sha256": "60648e349b359bbd591609068699896e57f04b530114d7f623b6ae557f0b26e0",
+  "author_scope": "AUTHOR_SCOPE.md",
+  "page_count": 131,
+  "assignments": [
+    {
+      "page_131_deck": 1,
+      "original_76_page": 1,
+      "title": "PBISC-Diffusion: expanded evidence deck",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 2,
+      "original_76_page": 2,
+      "title": "Full evidence map",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 3,
+      "original_76_page": "",
+      "title": "How to read this expanded evidence deck",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": "Original figures retain their data and evidence lineage; sources and limitations are in the speaker notes."
+    },
+    {
+      "page_131_deck": 4,
+      "original_76_page": 3,
+      "title": "Manuscript plan: six main figures",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_overview.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 5,
+      "original_76_page": "",
+      "title": "Availability is not the same as a new experiment",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": "Planning cards from the original deck are retained as dated snapshots, not a fresh list of experiments to rerun."
+    },
+    {
+      "page_131_deck": 6,
+      "original_76_page": 4,
+      "title": "Figure 1. Design and performance of PBISC",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 7,
+      "original_76_page": 5,
+      "title": "Figure 1 plan: Design and performance of PBISC",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_fig1.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 8,
+      "original_76_page": 6,
+      "title": "Backbone: SDD → PBISC-Diffusion",
+      "target": "Supp S01",
+      "use": "SUPPLEMENT",
+      "role": "모델·conditioning·학습·평가 설계 상세",
+      "data_regime": "개념/방법 설명",
+      "required_action": "Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분",
+      "secondary_target": "",
+      "figure_files": "figures/sec02_backbone_compare.png",
+      "current_protocol_note": "SDD: Ostheimer et al., arXiv 2502.02448. scDiffusion: Luo et al., Bioinformatics 2024."
+    },
+    {
+      "page_131_deck": 9,
+      "original_76_page": 7,
+      "title": "Overview draft (Nature style)",
+      "target": "Figure 1",
+      "use": "MAIN",
+      "role": "전체 pretraining → adaptation → generation 개요",
+      "data_regime": "개념도",
+      "required_action": "사용자 정의에 맞게 전체 실험 흐름의 중심 패널로 사용",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig1b_overview_nature.png",
+      "current_protocol_note": "Schematic. Conditioning recipe size and the number of generated cells are distinct; no cell-type quotas."
+    },
+    {
+      "page_131_deck": 10,
+      "original_76_page": 8,
+      "title": "Bulk conditioning",
+      "target": "Supp S01",
+      "use": "SUPPLEMENT",
+      "role": "모델·conditioning·학습·평가 설계 상세",
+      "data_regime": "개념/방법 설명",
+      "required_action": "Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분",
+      "secondary_target": "",
+      "figure_files": "figures/sec02_bulk_conditioning.png",
+      "current_protocol_note": "Final 6k model. Recipe size K conditions the bulk; the number of generated cells is chosen separately."
+    },
+    {
+      "page_131_deck": 11,
+      "original_76_page": 9,
+      "title": "Pretraining corpus",
+      "target": "Figure 1",
+      "use": "MAIN",
+      "role": "corpus 규모·assay·condition·cell-type 통계",
+      "data_regime": "pretraining corpus 통계",
+      "required_action": "cell·donor·dataset 분모와 학습 노출량을 구분",
+      "secondary_target": "",
+      "figure_files": "figures/fig1d_corpus.png",
+      "current_protocol_note": "Corpus labels describe the data; they are not model inputs. Dataset composition is not sampler exposure."
+    },
+    {
+      "page_131_deck": 12,
+      "original_76_page": 10,
+      "title": "Width & context ablation: loss saturation",
+      "target": "Supp S02",
+      "use": "SUPPLEMENT",
+      "role": "SLE pseudobulk 모델 검증과 대조군",
+      "data_regime": "pretraining validation 또는 SLE pseudobulk",
+      "required_action": "12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec03_width_loss.png",
+      "current_protocol_note": "Old 6k panel (1% floor); one validation set. 16x: memory-floor stop after part 2."
+    },
+    {
+      "page_131_deck": 13,
+      "original_76_page": 11,
+      "title": "Width & context ablation: benchmark",
+      "target": "Figure 2",
+      "use": "MAIN",
+      "role": "architecture·gene-panel ablation의 SLE pseudobulk 성능",
+      "data_regime": "SLE pseudobulk development benchmark",
+      "required_action": "architecture 1% floor와 panel 0.1% floor를 분리; paired measured-bulk 결과로 표기하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec03_width_bench.png",
+      "current_protocol_note": "SLE, 42 donors, 2,980 genes; old 6k panel. ctx x4: better L1 (0.161 vs 0.197)."
+    },
+    {
+      "page_131_deck": 14,
+      "original_76_page": 12,
+      "title": "Gene context: 2k / 3k / 6k / 15k",
+      "target": "Figure 2",
+      "use": "MAIN",
+      "role": "architecture·gene-panel ablation의 SLE pseudobulk 성능",
+      "data_regime": "SLE pseudobulk development benchmark",
+      "required_action": "architecture 1% floor와 panel 0.1% floor를 분리; paired measured-bulk 결과로 표기하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec03_panel.png",
+      "current_protocol_note": "Existing panel benchmark. 6k: five sampling seeds; other panels: one seed. Different axis from the architecture sweep."
+    },
+    {
+      "page_131_deck": 15,
+      "original_76_page": "",
+      "title": "Native-pseudobulk model comparisons",
+      "target": "Figure 2",
+      "use": "MAIN",
+      "role": "architecture·gene-panel ablation의 SLE pseudobulk 성능",
+      "data_regime": "SLE pseudobulk development benchmark",
+      "required_action": "architecture 1% floor와 panel 0.1% floor를 분리; paired measured-bulk 결과로 표기하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/review_style_20261013_ablation.png",
+      "current_protocol_note": "Completed held-out rescoring. Architecture: 1% detection floor; gene-panel comparison: 0.1% floor."
+    },
+    {
+      "page_131_deck": 16,
+      "original_76_page": 13,
+      "title": "PBISC-Diffusion: bulk in, cells out",
+      "target": "Figure 1",
+      "use": "MAIN_EXTRACT",
+      "role": "corpus별 규모와 recipe-size 분포",
+      "data_regime": "개념도 + corpus 통계",
+      "required_action": "하단 corpus/K 통계만 추출; 상단 workflow는 9쪽과 중복되면 보충으로 이동",
+      "secondary_target": "Supp S01",
+      "figure_files": "figures/sec01_slide_a.png",
+      "current_protocol_note": "Fig. 1b redraw. Final model corpus: 33.1M train cells; K = 1–4,000 cells per pseudobulk."
+    },
+    {
+      "page_131_deck": 17,
+      "original_76_page": 14,
+      "title": "Denoiser: joint value + support diffusion",
+      "target": "Supp S01",
+      "use": "SUPPLEMENT",
+      "role": "모델·conditioning·학습·평가 설계 상세",
+      "data_regime": "개념/방법 설명",
+      "required_action": "Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분",
+      "secondary_target": "",
+      "figure_files": "figures/sec01_fig1a_denoiser.png",
+      "current_protocol_note": "Fig. 1a redraw. Final model f01 6k 1x, epoch 16, EMA; 50 DDIM steps."
+    },
+    {
+      "page_131_deck": 18,
+      "original_76_page": "",
+      "title": "Does the model follow its bulk conditioning?",
+      "target": "Supp S02",
+      "use": "SUPPLEMENT",
+      "role": "SLE pseudobulk 모델 검증과 대조군",
+      "data_regime": "pretraining validation 또는 SLE pseudobulk",
+      "required_action": "12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/sle_conditioning_interventions.png",
+      "current_protocol_note": "Existing SLE pseudobulk controls: same noise, 42 donors, 2,279 genes. This is a conditioning intervention, not real-bulk validation."
+    },
+    {
+      "page_131_deck": 19,
+      "original_76_page": 15,
+      "title": "Figure 2. Fine-tuning PBISC with LoRA",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 20,
+      "original_76_page": 16,
+      "title": "Figure 2 plan: Fine-tuning PBISC with LoRA",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_fig2.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 21,
+      "original_76_page": 17,
+      "title": "Real bulk ≠ pseudobulk (model input)",
+      "target": "Figure 3",
+      "use": "MAIN",
+      "role": "healthy real-bulk / pseudobulk 입력 mismatch",
+      "data_regime": "ImMicro measured bulk–sc paired",
+      "required_action": "shift removal은 paired pseudobulk를 쓴 진단임을 명시",
+      "secondary_target": "",
+      "figure_files": "figures/sec05_condition_space.png",
+      "current_protocol_note": "Existing paired-input diagnostic; mean-shift removal uses paired pseudobulks and is not a deployment method."
+    },
+    {
+      "page_131_deck": 22,
+      "original_76_page": 18,
+      "title": "Same shift in three paired cohorts",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/sec05_shift_cohorts.png",
+      "current_protocol_note": "Model inputs only, no generation. PCA fitted per cohort on sc pseudobulks."
+    },
+    {
+      "page_131_deck": 23,
+      "original_76_page": 19,
+      "title": "Consequence: cells off-manifold",
+      "target": "Figure 3",
+      "use": "MAIN_ILLUSTRATION",
+      "role": "mismatch가 생성 세포 공간에 미치는 영향",
+      "data_regime": "ImMicro paired 입력의 native-depth zero-shot 진단",
+      "required_action": "FT 이후 그림으로 오해하지 않게 표시; 다른 scorer의 before/after 수치를 직접 연결하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/sec05_cells_umap.png | figures/expanded_master_20261013/palette/sec05_cells_metrics.png",
+      "current_protocol_note": "Existing same-noise input diagnostic; not the new LoRA/scSemi OOF embedding. UMAP overlap is not a quantitative score."
+    },
+    {
+      "page_131_deck": 24,
+      "original_76_page": "",
+      "title": "Input corrections on one real-reference UMAP",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/stored_real_reference_umap.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 25,
+      "original_76_page": "",
+      "title": "Where generated-cell density differs from real cells",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/saved_density_difference.png",
+      "current_protocol_note": "Original high-resolution density map; zoom for panel labels. Quantitative colors preserved. Historical native-depth controls."
+    },
+    {
+      "page_131_deck": 26,
+      "original_76_page": "",
+      "title": "Input calibration changes cell-type composition",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/input_control_composition.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 27,
+      "original_76_page": "",
+      "title": "Same noise, different input: matched-cell shifts",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/stored_matched_trajectories.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 28,
+      "original_76_page": "",
+      "title": "Matched noise separates cell state from cell identity",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/matched_similarity.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 29,
+      "original_76_page": "",
+      "title": "Input changes can switch the generated cell label",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/label_transitions.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 30,
+      "original_76_page": "",
+      "title": "Nearest real cells: distance and label agreement",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/nearest_real_by_type.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 31,
+      "original_76_page": "",
+      "title": "Composition agreement does not ensure state agreement",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/within_type_centroids.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 32,
+      "original_76_page": "",
+      "title": "Gene-length effects in the bulk input",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/gene_length_input.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 33,
+      "original_76_page": "",
+      "title": "Gene-length effects within label-stable cells",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/gene_length_stable.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 34,
+      "original_76_page": "",
+      "title": "Raw and TPM shifts persist within cell labels",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/raw_tpm_top_genes.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 35,
+      "original_76_page": "",
+      "title": "Reference calibration leaves specific gene-state shifts",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/refcal_top_genes.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 36,
+      "original_76_page": "",
+      "title": "Donor trait axes and donor identification differ",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/donor_axes.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 37,
+      "original_76_page": "",
+      "title": "Reference choice changes different quality endpoints",
+      "target": "Supp S03",
+      "use": "SUPPLEMENT",
+      "role": "paired 입력 mismatch와 calibration 상세",
+      "data_regime": "paired input diagnostics; 다수 native-depth 조건",
+      "required_action": "sc-derived depth/masks,10-pair calibration,AIFI/pretraining overlap을 명시. strict-five 결과와 합치지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/reference_choice.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 38,
+      "original_76_page": 20,
+      "title": "Fine-tuning with paired bulk + sc",
+      "target": "Supp S01",
+      "use": "SUPPLEMENT",
+      "role": "모델·conditioning·학습·평가 설계 상세",
+      "data_regime": "개념/방법 설명",
+      "required_action": "Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec06_ft_method.png",
+      "current_protocol_note": "Base: final model f01 6k 1x E16. Times from 40 LoRA runs on this host."
+    },
+    {
+      "page_131_deck": 39,
+      "original_76_page": 21,
+      "title": "Evaluation design",
+      "target": "Supp S01",
+      "use": "SUPPLEMENT",
+      "role": "모델·conditioning·학습·평가 설계 상세",
+      "data_regime": "개념/방법 설명",
+      "required_action": "Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분",
+      "secondary_target": "",
+      "figure_files": "figures/sec06_eval_design.png",
+      "current_protocol_note": "Bars: Ye 2026 sepsis. QC-failed pairs excluded."
+    },
+    {
+      "page_131_deck": 40,
+      "original_76_page": "",
+      "title": "Why the historical and strict protocols differ",
+      "target": "Supp S04",
+      "use": "SUPPLEMENT",
+      "role": "평가 단위·annotation·depth/readout 검증",
+      "data_regime": "프로토콜/observed controls/native-depth 진단",
+      "required_action": "technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": "Support draws overlap in donors. Displayed draw variation is not independent population replication."
+    },
+    {
+      "page_131_deck": 41,
+      "original_76_page": 22,
+      "title": "How we score generated cells",
+      "target": "Supp S01",
+      "use": "SUPPLEMENT",
+      "role": "모델·conditioning·학습·평가 설계 상세",
+      "data_regime": "개념/방법 설명",
+      "required_action": "Figure1/2/3를 보완; 경험적 SLE/paired benchmark와 구분",
+      "secondary_target": "",
+      "figure_files": "figures/sec04_metrics_overview.png",
+      "current_protocol_note": "Definitions follow the cited papers; scorers: tools/score_ft.py, extra_metrics.py, contrast_extra.py."
+    },
+    {
+      "page_131_deck": 42,
+      "original_76_page": "",
+      "title": "Technical repeats and follow-up are different controls",
+      "target": "Supp S04",
+      "use": "SUPPLEMENT",
+      "role": "평가 단위·annotation·depth/readout 검증",
+      "data_regime": "프로토콜/observed controls/native-depth 진단",
+      "required_action": "technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/immicro_replicate_controls.png",
+      "current_protocol_note": "Observed ImMicro controls. Technical remeasurement and follow-up: 16 donors each. Follow-up is not a technical replicate."
+    },
+    {
+      "page_131_deck": 43,
+      "original_76_page": "",
+      "title": "Annotation agreement and difficult categories",
+      "target": "Supp S04",
+      "use": "SUPPLEMENT",
+      "role": "평가 단위·annotation·depth/readout 검증",
+      "data_regime": "프로토콜/observed controls/native-depth 진단",
+      "required_action": "technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/annotation_agreement.png",
+      "current_protocol_note": "Observed cells only. Overall agreement uses the supported six-type gate; author mixed/unlabelled cells are separate. Other is difficult."
+    },
+    {
+      "page_131_deck": 44,
+      "original_76_page": 23,
+      "title": "Healthy cohort: LoRA cells cover the real ones",
+      "target": "Supp S05",
+      "use": "SUPPLEMENT",
+      "role": "healthy paired 재구성·FT mode·대표 샘플 대조",
+      "data_regime": "ImMicro measured bulk–sc paired",
+      "required_action": "48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig2_healthy_cells.png",
+      "current_protocol_note": "ImMicro, 107 healthy donors. 5 donors: mean of 5 draws (dots); 5-fold CV shown separately."
+    },
+    {
+      "page_131_deck": 45,
+      "original_76_page": 24,
+      "title": "Healthy cohort: realism beyond precision/coverage",
+      "target": "Supp S05",
+      "use": "SUPPLEMENT",
+      "role": "healthy paired 재구성·FT mode·대표 샘플 대조",
+      "data_regime": "ImMicro measured bulk–sc paired",
+      "required_action": "48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig2_healthy_realism.png",
+      "current_protocol_note": "ImMicro, 107 healthy donors. Dashed = real vs real cells of the same donors."
+    },
+    {
+      "page_131_deck": 46,
+      "original_76_page": "",
+      "title": "Depth matching changes precision and coverage",
+      "target": "Supp S04",
+      "use": "SUPPLEMENT",
+      "role": "평가 단위·annotation·depth/readout 검증",
+      "data_regime": "프로토콜/observed controls/native-depth 진단",
+      "required_action": "technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/depth_matched_quality.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 47,
+      "original_76_page": "",
+      "title": "Detection and library size are separate diagnostics",
+      "target": "Supp S04",
+      "use": "SUPPLEMENT",
+      "role": "평가 단위·annotation·depth/readout 검증",
+      "data_regime": "프로토콜/observed controls/native-depth 진단",
+      "required_action": "technical repeat와 follow-up, annotation taxonomy, sequencing depth를 각각 구분",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/detection_and_library.png",
+      "current_protocol_note": "Historical native-depth diagnostic: sc-derived depth/masks; 10-pair calibration; AIFI overlaps pretraining. One seed."
+    },
+    {
+      "page_131_deck": 48,
+      "original_76_page": 25,
+      "title": "Healthy cohort: full vs LoRA vs condition encoder",
+      "target": "Supp S05",
+      "use": "SUPPLEMENT",
+      "role": "healthy paired 재구성·FT mode·대표 샘플 대조",
+      "data_regime": "ImMicro measured bulk–sc paired",
+      "required_action": "48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig2_healthy_modes.png",
+      "current_protocol_note": "ImMicro, 107 healthy donors; cond.-encoder-only ran on fold 0 (22 donors) only."
+    },
+    {
+      "page_131_deck": 49,
+      "original_76_page": 26,
+      "title": "Healthy cohort: how many pairs",
+      "target": "Figure 3",
+      "use": "MAIN_CONDITIONAL",
+      "role": "LoRA training-pair 수에 따른 학습곡선",
+      "data_regime": "ImMicro paired; n training + 7 validation",
+      "required_action": "학습 n과 총 paired 접근량을 구분. strict total-n 주장에는 그대로 사용 불가; 정확히5쌍은 54쪽으로 입증",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig2_healthy_pairs.png",
+      "current_protocol_note": "Historical n + 7 validation design. The fixed-3000-step strict curve is a separate protocol update."
+    },
+    {
+      "page_131_deck": 50,
+      "original_76_page": 27,
+      "title": "Healthy cohort: LoRA rank and K",
+      "target": "Supp S06",
+      "use": "SUPPLEMENT",
+      "role": "rank·pair budget·K·시간 상세",
+      "data_regime": "healthy 또는 mixed paired cohorts",
+      "required_action": "Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig2_healthy_rank.png",
+      "current_protocol_note": "ImMicro, 107 donors, 5-fold CV pooled. Open symbols: K choices with every training pair."
+    },
+    {
+      "page_131_deck": 51,
+      "original_76_page": "",
+      "title": "Healthy bulk: full fine-tuning and LoRA rank",
+      "target": "Figure 3",
+      "use": "MAIN",
+      "role": "full FT와 LoRA rank의 품질·파라미터 절충",
+      "data_regime": "ImMicro paired; historical pooled five-fold",
+      "required_action": "r16을 모든 지표 최적으로 쓰지 않음. r4/r16/r64와 full FT를 함께 표시",
+      "secondary_target": "Supp S06",
+      "figure_files": "figures/expanded_master_20261013/palette/review_style_20261013_rank.png",
+      "current_protocol_note": "Original donor-level five-fold comparison. LoRA rank 16 is a quality/parameter compromise."
+    },
+    {
+      "page_131_deck": 52,
+      "original_76_page": 28,
+      "title": "Figure 3. PBISC vs scSemiProfiler on healthy donors",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 ���획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 53,
+      "original_76_page": 29,
+      "title": "Figure 3 plan: PBISC vs scSemiProfiler on healthy donors",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_fig3.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 54,
+      "original_76_page": 30,
+      "title": "Healthy donors, 5 pairs: LoRA-5 vs scSemiProfiler-5",
+      "target": "Figure 3",
+      "use": "MAIN",
+      "role": "실제 exact-five adaptation 성능과 scSemi 비교",
+      "data_regime": "ImMicro measured bulk–sc paired; 5 pairs, no extra validation, fixed3000",
+      "required_action": "5개 support draw를 표시; 남은 102 donors/draw. 반복 draw를 독립 환자 반복으로 세지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig3_healthy_fewpair.png",
+      "current_protocol_note": "ImMicro: exactly 5 pairs, no extra validation, 3000-step LoRA; five draws, remaining 102 donors per draw."
+    },
+    {
+      "page_131_deck": 55,
+      "original_76_page": "",
+      "title": "Composition: error and donor-level wins",
+      "target": "Figure 3",
+      "use": "MAIN_OPTIONAL",
+      "role": "exact-five의 cell-type composition 오차·donor별 우위",
+      "data_regime": "ImMicro measured bulk–sc paired",
+      "required_action": "54쪽 보완 패널 또는 Supp S05로 배치. per-draw 절대오차 후 donor 평균 정의 유지",
+      "secondary_target": "Supp S05",
+      "figure_files": "figures/expanded_master_20261013/palette/review_style_20261013_healthy_composition.png",
+      "current_protocol_note": "Exactly five pairs per draw. Absolute errors averaged within each held-out donor; 107 donors, equal weight. Descriptive."
+    },
+    {
+      "page_131_deck": 56,
+      "original_76_page": 31,
+      "title": "Healthy paired-data learning curves",
+      "target": "Supp S06",
+      "use": "SUPPLEMENT",
+      "role": "rank·pair budget·K·시간 상세",
+      "data_regime": "healthy 또는 mixed paired cohorts",
+      "required_action": "Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig3_healthy_pairs.png",
+      "current_protocol_note": "LoRA: n training + 7 validation. scSemi: n + 7 representatives, with CV-HVG reuse. Not the strict total-n protocol."
+    },
+    {
+      "page_131_deck": 57,
+      "original_76_page": 32,
+      "title": "Representative-dependent composition",
+      "target": "Supp S05",
+      "use": "SUPPLEMENT",
+      "role": "healthy paired 재구성·FT mode·대표 샘플 대조",
+      "data_regime": "ImMicro measured bulk–sc paired",
+      "required_action": "48의 condition-encoder는 fold0 pilot. 57은 literal cell copying이 아닌 composition dependence",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig3_healthy_copy.png",
+      "current_protocol_note": "Existing healthy five-draw diagnostic. This evaluates composition dependence, not literal copying of every cell."
+    },
+    {
+      "page_131_deck": 58,
+      "original_76_page": 33,
+      "title": "Time: days for scSemiProfiler, minutes for LoRA",
+      "target": "Supp S06",
+      "use": "SUPPLEMENT",
+      "role": "rank·pair budget·K·시간 상세",
+      "data_regime": "healthy 또는 mixed paired cohorts",
+      "required_action": "Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig3_time_days.png",
+      "current_protocol_note": "One GB10 GPU, job alone on it. GPU time from receipts; waits for the GPU excluded."
+    },
+    {
+      "page_131_deck": 59,
+      "original_76_page": 34,
+      "title": "5 donors: time to apply",
+      "target": "Supp S06",
+      "use": "SUPPLEMENT",
+      "role": "rank·pair budget·K·시간 상세",
+      "data_regime": "healthy 또는 mixed paired cohorts",
+      "required_action": "Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_time_to_apply.png",
+      "current_protocol_note": "Existing measurements: adaptation + target generation; pretraining, GPU wait and scoring excluded."
+    },
+    {
+      "page_131_deck": 60,
+      "original_76_page": 35,
+      "title": "Figure 4. Disease signal on paired cohorts",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 61,
+      "original_76_page": 36,
+      "title": "Figure 4 plan: Disease signal on paired cohorts",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_fig4.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 62,
+      "original_76_page": 37,
+      "title": "Five paired cohorts",
+      "target": "Figure 4",
+      "use": "MAIN_EXTRACT",
+      "role": "질환 paired cohorts와 sample/donor 단위",
+      "data_regime": "ImMicro + 질환 paired cohort metadata",
+      "required_action": "Figure4에서는 질환 행만 사용; 전체 cohort 지도는 Figure1 선택 inset 가능",
+      "secondary_target": "Figure 1",
+      "figure_files": "figures/sec09_cohorts.png",
+      "current_protocol_note": "Existing paired-cohort map. Five donors means 5 pairs in Sepsis, 12-15 Imperial, 12-14 Giroux, 7-9 Ji."
+    },
+    {
+      "page_131_deck": 63,
+      "original_76_page": 38,
+      "title": "Disease signals with five training donors",
+      "target": "Figure 4",
+      "use": "MAIN_REANALYSIS",
+      "role": "five-donor LoRA와 기존 방법의 disease integrity",
+      "data_regime": "질환 measured bulk–sc paired; historical five-donor draws",
+      "required_action": "multi-visit의 five donors ≠ five samples. cohort별 예외와 동일한 donor-aware estimator 경계를 표시",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig4_disease_fewpair.png",
+      "current_protocol_note": "Existing five-donor protocol; sample counts differ by cohort. Original sample-level estimator; corrected update separate."
+    },
+    {
+      "page_131_deck": 64,
+      "original_76_page": "",
+      "title": "Sepsis: exact-five reconstruction across five draws",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/strict_sepsis_quality.png",
+      "current_protocol_note": "Completed exact-five Sepsis draws. Bars = means; dots = support draws. Five matched methods; no new generation."
+    },
+    {
+      "page_131_deck": 65,
+      "original_76_page": "",
+      "title": "Sepsis: realism beyond precision and coverage",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/strict_sepsis_realism.png",
+      "current_protocol_note": "Completed exact-five outputs; raw readout, fixed R1 reference. Real/real control shown. Dots are five support draws."
+    },
+    {
+      "page_131_deck": 66,
+      "original_76_page": "",
+      "title": "Imperial: first completed exact-five comparison",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/strict_imperial_draw0_quality.png",
+      "current_protocol_note": "One completed support draw (draw 0). Exact five paired samples; all five methods scored. Not the final five-draw comparison."
+    },
+    {
+      "page_131_deck": 67,
+      "original_76_page": 39,
+      "title": "5-fold CV: disease signal",
+      "target": "Figure 4",
+      "use": "MAIN_REANALYSIS",
+      "role": "five-fold held-out disease integrity",
+      "data_regime": "질환 measured bulk–sc paired; pooled five-fold",
+      "required_action": "63쪽과 design을 분리; 같은 평가축/estimators로 기존 방법을 비교",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig4_disease_fivefold.png",
+      "current_protocol_note": "Original donor-level split and original signal estimator. These results are distinct from the exact-five update."
+    },
+    {
+      "page_131_deck": 68,
+      "original_76_page": 40,
+      "title": "Disease genes at every list length (CAT)",
+      "target": "Figure 4",
+      "use": "MAIN_REPLOT",
+      "role": "gene-list 길이에 따른 CAT 비교",
+      "data_regime": "질환 measured bulk–sc paired; five-donor + five-fold",
+      "required_action": "이미 저장된 scSemi five-fold CAT를 추가. generated predictor 행만 사용하고 bulk-broadcast 행 제외",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_cat_curves.png",
+      "current_protocol_note": "Existing CAT curves; ranking agreement and chance line. Patient-aware reanalysis reuses the same predictions."
+    },
+    {
+      "page_131_deck": 69,
+      "original_76_page": "",
+      "title": "Sepsis: patient-aware disease-gene agreement",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/strict_ye2026_sepsis_deg.png",
+      "current_protocol_note": "Completed corrected count-based DEG fits; mean of five support draws."
+    },
+    {
+      "page_131_deck": 70,
+      "original_76_page": "",
+      "title": "Imperial: patient-aware disease-gene agreement",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/strict_imperial_eskd_deg.png",
+      "current_protocol_note": "Single completed exact-five draw 0; patient-aware DEG fits. Other draws are outside this frozen snapshot."
+    },
+    {
+      "page_131_deck": 71,
+      "original_76_page": 41,
+      "title": "Cell-type deconvolution, same pairs: disease signal",
+      "target": "Figure 4",
+      "use": "MAIN_QUALIFIED",
+      "role": "BayesPrism/CIBERSORTx 대비 cell-type disease expression",
+      "data_regime": "질환 measured bulk–sc paired; supported expression endpoints",
+      "required_action": "동일 support/공통 gene axis/가용 draw 수를 명시. 미계산 상관은0으로 처리하지 않고 DEG의 비우위도 유지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/secS_deconv_disease.png",
+      "current_protocol_note": "Completed BayesPrism/CIBERSORTx comparisons; same five-donor support, method-supported expression metrics."
+    },
+    {
+      "page_131_deck": 72,
+      "original_76_page": 42,
+      "title": "SLE benchmark: PBISC vs scSemiProfiler",
+      "target": "Supp S02",
+      "use": "SUPPLEMENT",
+      "role": "SLE pseudobulk 모델 검증과 대조군",
+      "data_regime": "pretraining validation 또는 SLE pseudobulk",
+      "required_action": "12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec04_sle_benchmark.png",
+      "current_protocol_note": "Completed pseudobulk benchmark; separate from external paired real-bulk validation. PBISC five seeds, scSemi one run."
+    },
+    {
+      "page_131_deck": 73,
+      "original_76_page": 43,
+      "title": "SLE benchmark: per cell type",
+      "target": "Supp S02",
+      "use": "SUPPLEMENT",
+      "role": "SLE pseudobulk 모델 검증과 대조군",
+      "data_regime": "pretraining validation 또는 SLE pseudobulk",
+      "required_action": "12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec04_sle_per_celltype.png",
+      "current_protocol_note": "Same 42 donors, 1,006 genes. PBISC: 5 seeds, mean ± SD; scSemiProfiler: 1 run. Other: top-50 of 12."
+    },
+    {
+      "page_131_deck": 74,
+      "original_76_page": "",
+      "title": "Disjoint-cell controls test the bulk-broadcast baseline",
+      "target": "Supp S02",
+      "use": "SUPPLEMENT",
+      "role": "SLE pseudobulk 모델 검증과 대조군",
+      "data_regime": "pretraining validation 또는 SLE pseudobulk",
+      "required_action": "12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/healthy/sle_disjoint_truth_control.png",
+      "current_protocol_note": "SLE pseudobulk benchmark; five existing half-cell splits. Error bars are saved split SDs. No new cell splitting or scoring."
+    },
+    {
+      "page_131_deck": 75,
+      "original_76_page": "",
+      "title": "SLE: what is added beyond the bulk broadcast?",
+      "target": "Supp S02",
+      "use": "SUPPLEMENT",
+      "role": "SLE pseudobulk 모델 검증과 대조군",
+      "data_regime": "pretraining validation 또는 SLE pseudobulk",
+      "required_action": "12쪽은 corpus validation loss, 나머지는 SLE development. paired disease Figure4에 넣지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec04_sle_broadcast.png",
+      "current_protocol_note": "Existing SLE pseudobulk development benchmark; bulk broadcast is an explicit comparator. Separate from paired real-bulk validation."
+    },
+    {
+      "page_131_deck": 76,
+      "original_76_page": "",
+      "title": "Influenza: response order versus response genes",
+      "target": "Supp S09",
+      "use": "SUPPLEMENT",
+      "role": "외부 질환 real-bulk diagnostic",
+      "data_regime": "influenza/COVID historical zero-shot/calibration",
+      "required_action": "LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig_randolph_celltype_response.png",
+      "current_protocol_note": "Historical zero-shot influenza test. Response-strength ordering is recovered; gene-wise agreement remains below the matched bulk broadcast."
+    },
+    {
+      "page_131_deck": 77,
+      "original_76_page": "",
+      "title": "COVID: calibration on a shared real-cell map",
+      "target": "Supp S09",
+      "use": "SUPPLEMENT",
+      "role": "외부 질환 real-bulk diagnostic",
+      "data_regime": "influenza/COVID historical zero-shot/calibration",
+      "required_action": "LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리",
+      "secondary_target": "",
+      "figure_files": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/refcal_scoring/arunachalam/placement/figures/fig1_umap_seven_panels_by_label_depth1000.png",
+      "current_protocol_note": "Historical 7 COVID / 5 HC samples, common depth. High-resolution reference overview; following detail panels provide readable quantitative results."
+    },
+    {
+      "page_131_deck": 78,
+      "original_76_page": "",
+      "title": "COVID monocytes: direction after depth matching",
+      "target": "Supp S09",
+      "use": "SUPPLEMENT",
+      "role": "외부 질환 real-bulk diagnostic",
+      "data_regime": "influenza/COVID historical zero-shot/calibration",
+      "required_action": "LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리",
+      "secondary_target": "",
+      "figure_files": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/refcal_scoring/arunachalam/placement/figures/fig2_covid_hc_monocyte_arrows_depth1000.png",
+      "current_protocol_note": "Historical 7 COVID / 5 HC samples, common depth. High-resolution reference overview; following panels separate direction, magnitude and relabeling null."
+    },
+    {
+      "page_131_deck": 79,
+      "original_76_page": "",
+      "title": "COVID monocytes: direction, magnitude and null",
+      "target": "Supp S09",
+      "use": "SUPPLEMENT",
+      "role": "외부 질환 real-bulk diagnostic",
+      "data_regime": "influenza/COVID historical zero-shot/calibration",
+      "required_action": "LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/covid_monocyte_direction_detail.png",
+      "current_protocol_note": "Readable detail of the retained COVID overview. Existing 7 COVID / 5 HC paired samples; native and depth-matched results stay separate."
+    },
+    {
+      "page_131_deck": 80,
+      "original_76_page": "",
+      "title": "COVID CD14 monocytes: direction, magnitude and null",
+      "target": "Supp S09",
+      "use": "SUPPLEMENT",
+      "role": "외부 질환 real-bulk diagnostic",
+      "data_regime": "influenza/COVID historical zero-shot/calibration",
+      "required_action": "LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/covid_cd14_direction_detail.png",
+      "current_protocol_note": "Readable detail of the retained COVID overview. Existing 7 COVID / 5 HC paired samples; native and depth-matched results stay separate."
+    },
+    {
+      "page_131_deck": 81,
+      "original_76_page": "",
+      "title": "COVID: biological endpoints give a mixed verdict",
+      "target": "Supp S09",
+      "use": "SUPPLEMENT",
+      "role": "외부 질환 real-bulk diagnostic",
+      "data_regime": "influenza/COVID historical zero-shot/calibration",
+      "required_action": "LoRA-five/scSemi paired rare-state 증거가 아님. Figure4 핵심 adaptation 비교와 분리",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/covid_endpoints.png",
+      "current_protocol_note": "Historical 7 COVID / 5 HC samples. Saved effect ratios and donor ranks; no new fits. Reference calibration is not uniformly best."
+    },
+    {
+      "page_131_deck": 82,
+      "original_76_page": 44,
+      "title": "Sepsis: fine-tuning variants",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig4_disease_variants.png",
+      "current_protocol_note": "Sepsis, 82 donors, pooled 5-fold. LoRA r = 16 is the model used throughout."
+    },
+    {
+      "page_131_deck": 83,
+      "original_76_page": 45,
+      "title": "Pairs needed: donor and disease signal",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec08_pairs_signal.png",
+      "current_protocol_note": "Sepsis: 61 case vs 21 control. logFC r = mean over T / B / Mono / NK. 0 = no pairs."
+    },
+    {
+      "page_131_deck": 84,
+      "original_76_page": 46,
+      "title": "Cell-type deconvolution, same pairs: composition",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/secS_deconv_composition.png",
+      "current_protocol_note": "Completed deconvolution outputs already exist. Verified identical Sepsis supports use adoption instead of method reruns."
+    },
+    {
+      "page_131_deck": 85,
+      "original_76_page": "",
+      "title": "Deconvolution: expression coverage limits the comparison",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/deconv_expression_coverage.png",
+      "current_protocol_note": "Completed historical CIBERSORTx five-donor runs. Missing disease correlations remain unavailable; deconvolution does not generate individual cells."
+    },
+    {
+      "page_131_deck": 86,
+      "original_76_page": 47,
+      "title": "Figure 5. Rare cell states in disease cohorts",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 87,
+      "original_76_page": 48,
+      "title": "Figure 5 plan: Rare cell states in disease cohorts",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않��. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_fig5.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 88,
+      "original_76_page": 49,
+      "title": "Rare-state fractions: original analysis",
+      "target": "Historical reference",
+      "use": "HOLD",
+      "role": "기존 rare-state scoring 기록",
+      "data_regime": "historical paired rare analysis",
+      "required_action": "zero-drop/sample-unit 문제가 있어 현재 manuscript 본문/정식 supplementary에서는 사용 보류; 수정된5donor결과는 S10",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig5_rare_fraction.png",
+      "current_protocol_note": "Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate."
+    },
+    {
+      "page_131_deck": 89,
+      "original_76_page": "",
+      "title": "Rare-state frequencies in observed cells",
+      "target": "Supp S10",
+      "use": "SUPPLEMENT",
+      "role": "paired rare-state 정량 검증과 case 선택 근거",
+      "data_regime": "paired disease rare-state quantitative",
+      "required_action": "Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/review_style_20261013_rare_observed.png",
+      "current_protocol_note": "Original five-donor outputs. Median across draws; state share within its parent population. Controls included."
+    },
+    {
+      "page_131_deck": 90,
+      "original_76_page": 50,
+      "title": "Rare-state ranking: original analysis",
+      "target": "Historical reference",
+      "use": "HOLD",
+      "role": "기존 rare-state scoring 기록",
+      "data_regime": "historical paired rare analysis",
+      "required_action": "zero-drop/sample-unit 문제가 있어 현재 manuscript 본문/정식 supplementary에서는 사용 보류; 수정된5donor결과는 S10",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/legacy_rare/fig5_rare_donors.png",
+      "current_protocol_note": "Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate."
+    },
+    {
+      "page_131_deck": 91,
+      "original_76_page": "",
+      "title": "Rare-state misses: rules 1–8",
+      "target": "Supp S10",
+      "use": "SUPPLEMENT",
+      "role": "paired rare-state 정량 검증과 case 선택 근거",
+      "data_regime": "paired disease rare-state quantitative",
+      "required_action": "Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/rare_misses_1.png",
+      "current_protocol_note": "Saved zero-aware diagnostic; all 15 predefined rules are shown across two slides. NA means not evaluable, not zero error."
+    },
+    {
+      "page_131_deck": 92,
+      "original_76_page": "",
+      "title": "Rare-state misses: rules 9–15",
+      "target": "Supp S10",
+      "use": "SUPPLEMENT",
+      "role": "paired rare-state 정량 검증과 case 선택 근거",
+      "data_regime": "paired disease rare-state quantitative",
+      "required_action": "Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/rare_misses_2.png",
+      "current_protocol_note": "Saved zero-aware diagnostic; all 15 predefined rules are shown across two slides. NA means not evaluable, not zero error."
+    },
+    {
+      "page_131_deck": 93,
+      "original_76_page": "",
+      "title": "Rare-state fidelity: zero-aware reanalysis",
+      "target": "Supp S10",
+      "use": "SUPPLEMENT",
+      "role": "paired rare-state 정량 검증과 case 선택 근거",
+      "data_regime": "paired disease rare-state quantitative",
+      "required_action": "Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/review_style_20261013_rare_corrected_comparison.png",
+      "current_protocol_note": "Original five-donor outputs; zero draws retained and actual donor IDs. 59 state/cohort combinations, including controls."
+    },
+    {
+      "page_131_deck": 94,
+      "original_76_page": 51,
+      "title": "Rare states: original five-fold analysis",
+      "target": "Historical reference",
+      "use": "HOLD",
+      "role": "기존 rare-state scoring 기록",
+      "data_regime": "historical paired rare analysis",
+      "required_action": "zero-drop/sample-unit 문제가 있어 현재 manuscript 본문/정식 supplementary에서는 사용 보류; 수정된5donor결과는 S10",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/fig5_rare_fivefold.png",
+      "current_protocol_note": "Historical scoring: zero-draw and sample-unit limitations apply. Corrected five-donor analyses follow; five-fold correction is separate."
+    },
+    {
+      "page_131_deck": 95,
+      "original_76_page": "",
+      "title": "Sepsis rare states across paired-data budgets",
+      "target": "Supp S10",
+      "use": "SUPPLEMENT",
+      "role": "paired rare-state 정량 검증과 case 선택 근거",
+      "data_regime": "paired disease rare-state quantitative",
+      "required_action": "Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/sepsis_rare_support.png",
+      "current_protocol_note": "Historical cross-fitted learning curve: n training pairs plus validation, not strict total n. All 15 rules; arm-specific depth matching."
+    },
+    {
+      "page_131_deck": 96,
+      "original_76_page": "",
+      "title": "Sepsis rare-state direction across support budgets",
+      "target": "Supp S10",
+      "use": "SUPPLEMENT",
+      "role": "paired rare-state 정량 검증과 case 선택 근거",
+      "data_regime": "paired disease rare-state quantitative",
+      "required_action": "Fig5 정성 사례를 뒷받침하되 빈도/상관 heatmap을 population-identity 그림으로 대체하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/sepsis_rare_direction.png",
+      "current_protocol_note": "Saved sign-agreement counts across all 15 rules, including controls. Historical n + validation; directions are not effect-size accuracy."
+    },
+    {
+      "page_131_deck": 97,
+      "original_76_page": "",
+      "title": "Rare-state screen in DRESS, Kawasaki and SLE",
+      "target": "Supp S11",
+      "use": "SUPPLEMENT",
+      "role": "pseudobulk rare screen·unmatched Behcet 한계",
+      "data_regime": "own-sc pseudobulk 또는 unmatched observed reference",
+      "required_action": "Fig5의 독립 paired qualitative validation으로 사용하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/external_rare_screen.png",
+      "current_protocol_note": "Historical one-seed zero-shot screen conditioned on each sample's own sc pseudobulk. DRESS was in training; real-bulk validation is separate."
+    },
+    {
+      "page_131_deck": 98,
+      "original_76_page": 52,
+      "title": "Consequence: disease direction flips",
+      "target": "Supp S11",
+      "use": "SUPPLEMENT",
+      "role": "pseudobulk rare screen·unmatched Behcet 한계",
+      "data_regime": "own-sc pseudobulk 또는 unmatched observed reference",
+      "required_action": "Fig5의 독립 paired qualitative validation으로 사용하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec05_behcet_direction.png",
+      "current_protocol_note": "Historical Behcet diagnostic; reference is unmatched and includes training donors. This is not independent patient-level subtype validation."
+    },
+    {
+      "page_131_deck": 99,
+      "original_76_page": "",
+      "title": "C1Q: single-gene detection versus co-detection",
+      "target": "Supp S11",
+      "use": "SUPPLEMENT",
+      "role": "pseudobulk rare screen·unmatched Behcet 한계",
+      "data_regime": "own-sc pseudobulk 또는 unmatched observed reference",
+      "required_action": "Fig5의 독립 paired qualitative validation으로 사용하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/c1q_cross_disease.png",
+      "current_protocol_note": "Existing pseudobulk-conditioned comparison at matched depth. Multi-gene C1Q rates remain compressed, including in-corpus DRESS."
+    },
+    {
+      "page_131_deck": 100,
+      "original_76_page": "",
+      "title": "Behçet: K=4,000 changes direction more than identity",
+      "target": "Supp S11",
+      "use": "SUPPLEMENT",
+      "role": "pseudobulk rare screen·unmatched Behcet 한계",
+      "data_regime": "own-sc pseudobulk 또는 unmatched observed reference",
+      "required_action": "Fig5의 독립 paired qualitative validation으로 사용하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/behcet_k4000.png",
+      "current_protocol_note": "Two existing sampling seeds; 6 BD / 10 HC held-out bulks. K is condition size; each bulk still generates 5,000 cells. Descriptive."
+    },
+    {
+      "page_131_deck": 101,
+      "original_76_page": "",
+      "title": "Behcet: one-gene and multi-gene C1Q rules",
+      "target": "Supp S11",
+      "use": "SUPPLEMENT",
+      "role": "pseudobulk rare screen·unmatched Behcet 한계",
+      "data_regime": "own-sc pseudobulk 또는 unmatched observed reference",
+      "required_action": "Fig5의 독립 paired qualitative validation으로 사용하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/review_style_20261013_behcet.png",
+      "current_protocol_note": "Descriptive: unmatched reference includes training donors. Same three pairs; two identity-pass and one uncertain."
+    },
+    {
+      "page_131_deck": 102,
+      "original_76_page": 53,
+      "title": "Figure 6. Discovery from public bulk-only data",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 103,
+      "original_76_page": 54,
+      "title": "Figure 6 plan: Discovery from public bulk-only data",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_fig6.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 104,
+      "original_76_page": 55,
+      "title": "Bulk-only cohort plan: original snapshot",
+      "target": "Figure 6",
+      "use": "TODO_SETUP",
+      "role": "adaptation / bulk-only cohort 설계 후보",
+      "data_regime": "과거 cohort inventory",
+      "required_action": "새 재현 대상 논문·cohort가 확정된 후 다시 작성; 현재 수치를 최종 재현 cohort로 간주하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/sec10_paired_subset_cohorts.png",
+      "current_protocol_note": "Historical cohort counts. Existing exploratory outputs are reviewed separately; reproduction with the user-prepared new dataset remains pending."
+    },
+    {
+      "page_131_deck": 105,
+      "original_76_page": "",
+      "title": "Historical bulk-only search: no confirmed new finding",
+      "target": "Figure 6 reference",
+      "use": "WORKFLOW_REFERENCE",
+      "role": "과거 bulk-only 탐색·검토 기록",
+      "data_regime": "historical exploratory candidate audit",
+      "required_action": "0 confirmed는 특정 기존 논문 재현의 성공/실패를 뜻하지 않음; main Figure6 결과로 사용 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/bulkonly_verification_history.png",
+      "current_protocol_note": "Historical exploratory analysis of existing bulk-only generations. Review dispositions are not experimental validation. Figure 6 reproduction still awaits new data."
+    },
+    {
+      "page_131_deck": 106,
+      "original_76_page": "",
+      "title": "Bulk-only exploration: calibration populations differ",
+      "target": "Supp S14",
+      "use": "SUPPLEMENT",
+      "role": "bulk-only target cohort 차이",
+      "data_regime": "historical paired vs bulk-only population metadata",
+      "required_action": "Fig6 설계용 shift 점검; published finding의 재현 결과가 아님",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/bulkonly_imperial_wave_shift.png",
+      "current_protocol_note": "Historical Imperial sample composition. Paired and bulk-only sets differ in wave; repeated samples are not independent donors."
+    },
+    {
+      "page_131_deck": 107,
+      "original_76_page": "",
+      "title": "What remains to establish Figure 6",
+      "target": "Figure 6",
+      "use": "TODO_PLAN",
+      "role": "기존 논문 결과 재현의 검증 요건",
+      "data_regime": "계획",
+      "required_action": "실제 재현 결과 패널은 현재 없음",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": "Historical candidate analyses do not substitute for independent reproduction in the new cohort."
+    },
+    {
+      "page_131_deck": 108,
+      "original_76_page": 56,
+      "title": "Supplementary",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 109,
+      "original_76_page": 57,
+      "title": "Supplementary plan",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "figures/figplan_supp.png",
+      "current_protocol_note": "Original 8 October planning snapshot, retained for context. Current evidence and incomplete work are distinguished in this expanded deck."
+    },
+    {
+      "page_131_deck": 110,
+      "original_76_page": 58,
+      "title": "5 donors: LoRA vs scSemiProfiler — cells",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_fewpair_cells.png",
+      "current_protocol_note": "Completed five-donor comparisons. Multi-visit cohorts used more than five paired samples."
+    },
+    {
+      "page_131_deck": 111,
+      "original_76_page": 59,
+      "title": "5 donors: realism beyond precision/coverage",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_extra_fewpair.png",
+      "current_protocol_note": "E_N: scSemiProfiler's own error metric. Misses only vs calibration, e.g. Imperial E_N 1.28 vs 1.22."
+    },
+    {
+      "page_131_deck": 112,
+      "original_76_page": 60,
+      "title": "5 donors: precision/coverage at k = 3, 5, 10",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_extra_ksens.png",
+      "current_protocol_note": "Completed sensitivity at k=3,5,10; no new sensitivity campaign required."
+    },
+    {
+      "page_131_deck": 113,
+      "original_76_page": 61,
+      "title": "5-fold: realism beyond precision/coverage",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_extra_fivefold.png",
+      "current_protocol_note": "Donor-level 5-fold CV, one pooled call per cohort. E_N: scSemiProfiler's own error metric."
+    },
+    {
+      "page_131_deck": 114,
+      "original_76_page": 62,
+      "title": "Same training pairs: LoRA vs scSemiProfiler",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_same_pairs.png",
+      "current_protocol_note": "Completed original five-fold LoRA/scSemi comparison; not the new total-n curve."
+    },
+    {
+      "page_131_deck": 115,
+      "original_76_page": 63,
+      "title": "scSemiProfiler and the representative-copy baseline",
+      "target": "Supp S07",
+      "use": "SUPPLEMENT",
+      "role": "질환/다중 cohort 생성 품질과 비교 안정성",
+      "data_regime": "paired cell-level/realism/composition",
+      "required_action": "질환 신호 자체의 증거와 구분. strict6-unit snapshot과 historical designs를 혼합평균하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec09_scsemi_representative.png",
+      "current_protocol_note": "Copy = composition of the one training sample scSemiProfiler assigns to each test sample."
+    },
+    {
+      "page_131_deck": 116,
+      "original_76_page": 64,
+      "title": "Full vs LoRA vs cond.-encoder-only",
+      "target": "Figure 3",
+      "use": "MAIN_EXTRACT",
+      "role": "raw/calibrated baseline 대비 full FT 개선",
+      "data_regime": "ImMicro + Sepsis paired; pooled와 fold0 pilot 혼재",
+      "required_action": "dataset=immicro, call=pooled만 추출. Sepsis와 condition-encoder fold0 pilot을 본문 healthy 비교에 혼합하지 않음",
+      "secondary_target": "Supp S05",
+      "figure_files": "figures/expanded_master_20261013/palette/sec07_ft_modes.png",
+      "current_protocol_note": "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)."
+    },
+    {
+      "page_131_deck": 117,
+      "original_76_page": 65,
+      "title": "Pairs needed: composition, coverage, training steps",
+      "target": "Supp S06",
+      "use": "SUPPLEMENT",
+      "role": "rank·pair budget·K·시간 상세",
+      "data_regime": "healthy 또는 mixed paired cohorts",
+      "required_action": "Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec08_pairs_realism.png",
+      "current_protocol_note": "ImMicro / sepsis, 5-fold CV (107 / 82 donors). 0 = no pairs; squares = calibration fitted on the same n pairs."
+    },
+    {
+      "page_131_deck": 118,
+      "original_76_page": 66,
+      "title": "LoRA rank: r = 4, 16, 64",
+      "target": "Supp S06",
+      "use": "SUPPLEMENT",
+      "role": "rank·pair budget·K·시간 상세",
+      "data_regime": "healthy 또는 mixed paired cohorts",
+      "required_action": "Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec07_rank_sweep.png",
+      "current_protocol_note": "Sepsis, 58–59 pairs/fold: r64 logFC r 0.53 → 0.55, DEG 23 → 26; r4 0.48, DEG 18. Pooled 5-fold scores."
+    },
+    {
+      "page_131_deck": 119,
+      "original_76_page": 67,
+      "title": "K diversification: no gain",
+      "target": "Supp S06",
+      "use": "SUPPLEMENT",
+      "role": "rank·pair budget·K·시간 상세",
+      "data_regime": "healthy 또는 mixed paired cohorts",
+      "required_action": "Fig3 재사용 시 healthy만 추출. historical n+validation/CV-HVG와 exact-five를 구분; r16 universal-best 금지",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec07_k_aug.png",
+      "current_protocol_note": "LoRA r16, 5-fold pooled. p = paired Wilcoxon on donor L1, uncorrected."
+    },
+    {
+      "page_131_deck": 120,
+      "original_76_page": "",
+      "title": "K=4,000 recipe: paired reconstruction quality",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/k4000_quality.png",
+      "current_protocol_note": "Existing paired benchmark, five support draws per cohort and 1,500 cells/test sample. Bars = means; dots = draws. No new runs."
+    },
+    {
+      "page_131_deck": 121,
+      "original_76_page": "",
+      "title": "K=4,000 recipe: disease signal does not improve uniformly",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/k4000_disease.png",
+      "current_protocol_note": "Existing paired benchmark, five support draws per cohort and 1,500 cells/test sample. Bars = means; dots = draws. No new runs."
+    },
+    {
+      "page_131_deck": 122,
+      "original_76_page": "",
+      "title": "K=4,000 recipe: calibration gains are selective",
+      "target": "Supp S08",
+      "use": "SUPPLEMENT",
+      "role": "환자 단위 DEG·질환 sensitivity·deconvolution/K4000",
+      "data_regime": "disease paired; 일부 mixed/partial",
+      "required_action": "69는 Sepsis5draw,70은 Imperial1draw의 고정 snapshot. 전체 cohort 완성 결과처럼 제시하지 않음",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/disease/k4000_calibration.png",
+      "current_protocol_note": "Saved paired-test gates: 50 features × 5 draws per cohort. Descriptive counts, not 250 independent biological replicates."
+    },
+    {
+      "page_131_deck": 123,
+      "original_76_page": 68,
+      "title": "Pairs from other cohorts do not transfer",
+      "target": "Supp S12",
+      "use": "SUPPLEMENT",
+      "role": "pretraining·cross-cohort·scratch 대조",
+      "data_regime": "paired controls; 일부 fold0/one-seed pilot",
+      "required_action": "131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec08_cross_cohort.png",
+      "current_protocol_note": "Completed leave-one-cohort-out test at K=1000; target-cohort sc was excluded from training."
+    },
+    {
+      "page_131_deck": 124,
+      "original_76_page": 69,
+      "title": "Longer training on other cohorts: gap stays",
+      "target": "Supp S12",
+      "use": "SUPPLEMENT",
+      "role": "pretraining·cross-cohort·scratch 대조",
+      "data_regime": "paired controls; 일부 fold0/one-seed pilot",
+      "required_action": "131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec08_cross_cohort_steps.png",
+      "current_protocol_note": "Existing saturation experiment; no additional saturation run introduced here."
+    },
+    {
+      "page_131_deck": 125,
+      "original_76_page": 70,
+      "title": "No pretraining: same cohort, from scratch",
+      "target": "Supp S12",
+      "use": "SUPPLEMENT",
+      "role": "pretraining·cross-cohort·scratch 대조",
+      "data_regime": "paired controls; 일부 fold0/one-seed pilot",
+      "required_action": "131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec07_no_pretraining.png",
+      "current_protocol_note": "Completed Sepsis fold0 pilot, not all folds. Pretraining supports speed/few-pair adaptation, not universal superiority."
+    },
+    {
+      "page_131_deck": 126,
+      "original_76_page": 71,
+      "title": "No pretraining: other cohorts, from scratch",
+      "target": "Supp S12",
+      "use": "SUPPLEMENT",
+      "role": "pretraining·cross-cohort·scratch 대조",
+      "data_regime": "paired controls; 일부 fold0/one-seed pilot",
+      "required_action": "131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec08_cross_scratch.png",
+      "current_protocol_note": "Pilot: held-out sepsis, 82 samples. Genes no training cohort measured are zeroed in every arm."
+    },
+    {
+      "page_131_deck": 127,
+      "original_76_page": 72,
+      "title": "Rare states: from scratch vs pretrained",
+      "target": "Supp S12",
+      "use": "SUPPLEMENT",
+      "role": "pretraining·cross-cohort·scratch 대조",
+      "data_regime": "paired controls; 일부 fold0/one-seed pilot",
+      "required_action": "131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/secS_scratch_rare_states.png",
+      "current_protocol_note": "Pilot, one seed: sepsis fold 0 (17 donors), held-out sepsis (82, masked). 1 = observed rate; grey = within 2-fold."
+    },
+    {
+      "page_131_deck": 128,
+      "original_76_page": 73,
+      "title": "Appendix",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 129,
+      "original_76_page": 74,
+      "title": "Appendix: scSemiProfiler without output threshold",
+      "target": "Supp S13",
+      "use": "SUPPLEMENT",
+      "role": "scSemi output-threshold sensitivity",
+      "data_regime": "paired comparative readout sensitivity",
+      "required_action": "본문은 official output. 수정 readout은 별도 sensitivity로 표기",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/secA_scsemi_readout.png",
+      "current_protocol_note": "Completed modified-readout check; main results use official output. Models were not retrained for the sensitivity."
+    },
+    {
+      "page_131_deck": 130,
+      "original_76_page": 75,
+      "title": "Discussion",
+      "target": "Deck / planning",
+      "use": "ADMIN",
+      "role": "표지·목차·구분·이전 계획·읽기 안내",
+      "data_regime": "non-result material",
+      "required_action": "원고 Figure 증거에는 배정하지 않음. 덱의 기존 Figure 번호는 새 사용자 정의로 대체",
+      "secondary_target": "",
+      "figure_files": "",
+      "current_protocol_note": ""
+    },
+    {
+      "page_131_deck": 131,
+      "original_76_page": 76,
+      "title": "Next: other tissues from a few dozen pairs",
+      "target": "Supp S12",
+      "use": "SUPPLEMENT",
+      "role": "pretraining·cross-cohort·scratch 대조",
+      "data_regime": "paired controls; 일부 fold0/one-seed pilot",
+      "required_action": "131의 다른 tissue 전이는 전망이며 실제 검증 결과가 아님;127은 rare 정량 대조",
+      "secondary_target": "",
+      "figure_files": "figures/expanded_master_20261013/palette/sec10_other_tissues.png",
+      "current_protocol_note": "Pilots: sepsis fold 0 (17 donors); last row held-out sepsis (82). Folds 1–4 and a second cohort pending."
+    }
+  ]
+}