Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261012T130738Z-strictpairs-implementation/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/archive/20261012T130738Z-strictpairs-implementation/snapshot/manuscript/sections/fig2.tex

@@ -0,0 +1,3 @@
+\mainfigure{fig:fig2}{Pretraining ablations and paired-support design}
+\figpanelh{}{figures/strict_20261012/fig2_ablation_and_support.pdf}{0.70}
+\figlegend{\textbf{Design and available results.} Held-out native-pseudobulk precision and coverage are measured on a common 1,006-gene axis with 42 donors; architecture and panel training floors differ and are plotted separately. Saved epoch-16 outputs contribute one generation seed per arm. Historical rank and full-parameter five-fold comparisons are labelled separately. The nested exact-support $n=5,10,20,40$ curve and external AIFI reference-calibration comparison are pending and must not be read as results.}

paper-releases/pbisc-diffusion/updates/20261012T130738Z-strictpairs-implementation/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/archive/20261012T130738Z-strictpairs-implementation/snapshot/manuscript/sections/fig3.tex

@@ -0,0 +1,3 @@
+\mainfigure{fig:fig3}{Healthy-donor strict-support evaluation pending}
+\tbcpanel{Five complete ImMicro held-out folds, real-reference displays, six cell-quality measures and operational time are required.}
+\figlegend{\textbf{Pending panel.} The planned comparison fixes five independent training donors and one paired sample per donor in each held-out fold. No strict-support quantitative inference is displayed in this draft.}

paper-releases/pbisc-diffusion/updates/20261012T130738Z-strictpairs-implementation/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/archive/20261012T130738Z-strictpairs-implementation/snapshot/manuscript/sections/fig4.tex

@@ -0,0 +1,4 @@
+\mainfigure{fig:fig4}{Disease strict-support draws pending; historical five-fold analysis separate}
+\tbcpanel{Five exact-support draws in each of four disease cohorts, matched deconvolution and donor-aware disease contrasts are required.}
+\figpanelh{}{figures/strict_20261012/fig4_historical_5fold_separate.pdf}{0.42}
+\figlegend{\textbf{Historical panel only.} The lower panel reports earlier five-fold results. Its pair and validation budgets are different from the pending exact-five-pair design; the designs are not pooled. The upper strict-support slot is a placeholder and has no result.}

paper-releases/pbisc-diffusion/updates/20261012T130738Z-strictpairs-implementation/diffusion/dataset/server_1/model8/plans/paper_figures_strictpairs_20261012/archive/20261012T130738Z-strictpairs-implementation/snapshot/manuscript/sections/fig5.tex

@@ -0,0 +1,3 @@
+\mainfigure{fig:fig5}{Rare-state strict-support analysis pending}
+\tbcpanel{Exact-five-pair rare-state caches for all 20 disease draws and complete source checks are required.}
+\figlegend{\textbf{Pending panel.} The existing rare-state plot uses earlier five-donor caches and is diagnostic only. It is deliberately absent from this manuscript draft. Beh\c{c}et bulk and single-cell donors are mostly unmatched and will be described separately.}