Archive 2 checkpoints (1-2)

paper-releases/pbisc-diffusion/updates/20261003T174003Z-scsemi-readout-sensitivity/diffusion/metadata/NOTE.md

@@ -0,0 +1,5 @@
+# PBISC-diffusion update 20261003T174003Z-scsemi-readout-sensitivity
+
+scSemiProfiler read-out sensitivity on the 25 five-donor draws (2026-10-04): arms scsemi5nt (same trained per-target models, upstream's hard-coded output threshold of 10 normalized units removed) and scsemi5rt (control, same re-read with the threshold; reproduces the official output), CPU only, no retraining. New tools (scsemi/tools/resample_output_variants.py, export_variant_npz.py, validate_variant_export.py, run_output_variants.sh, compare_output_variants.py, index_variant_receipts.py), scsemi/README.md section, queues, score tables in scores/ and scores_nosex/ (Giroux), comparison and receipt index (scsemi/analysis/output_variants_20261003/), queue logs; generated NPZs registered as locations (nt on server_2, rt on server_1 or server_2). Result: threshold removal helps scSemiProfiler on every cell metric but no LoRA-5 comparison flips. Listed with the whole paired fine-tuning benchmark request (unchanged files are skipped).
+
+Payload states are recorded as supplied. Inference locations are pointers, not archived generated outputs.