Archive 2 checkpoints (1-2)

paper-releases/pbisc-diffusion/updates/20261002T022002Z-deck-time-to-apply-scsemi-reps/diffusion/metadata/NOTE.md

@@ -0,0 +1,5 @@
+# PBISC-diffusion update 20261002T022002Z-deck-time-to-apply-scsemi-reps
+
+Paper summary deck update 2026-10-02 (presentations/2026-09-30--paper-summary--r00, now 34 slides): new slide 24 '5 donors: time to apply' (GPU time per draw, LoRA-5 3-11 min vs scSemiProfiler-5 1.0-4.1 h on the same 25 draws, 11-63x less, from receipts and queue logs on both hosts; tools/sec09_time_to_apply.py) and backup slide 34 'scSemiProfiler copies its representative' (tools/sec09_scsemi_representative.py); deck, PDF, spec, README, scripts, figures and number tables. CPU only; no new checkpoint or inference. || scSemiProfiler representative-dependence analysis (plans/paired_finetune_intradataset_20260927/scsemi/analysis/representative_dependence_20261002; CPU only): per test sample, composition L1 of copying its assigned representative vs scSemiProfiler-5 and LoRA-5 over the 25 five-donor draws, plus a KMeans proxy for upstream's own representative choice (not a rerun); its inputs (assignment.json of every draw, draw-0 bulk PCA per cohort; Imperial copied from server_2 with sha256). Corrected as-run host record of the scSemiProfiler draws (sepsis 4: server_2 guard stop at 36/77 targets, finished on server_1) and the archive request builder (server_2 scSemiProfiler locations, as-run queue copies).
+
+Payload states are recorded as supplied. Inference locations are pointers, not archived generated outputs.

paper-releases/pbisc-diffusion/updates/20261002T022002Z-deck-time-to-apply-scsemi-reps/diffusion/metadata/inference_locations.json

@@ -0,0 +1,5 @@
+{
+  "locations": [],
+  "project_id": "pbisc-diffusion",
+  "update_id": "20261002T022002Z-deck-time-to-apply-scsemi-reps"
+}