Conditioning-noise regularization for diffusion-based artifact detection in histopathology — model outputs and evaluation data

Data deposit for the study 'Conditioning noise is a free regularizer for LoRA fine-tuning: no pathology encoder required for diffusion-based artifact detection in histopathology' (Moutselos & Maglogiannis, submitted). Contents: UNI2-h embedding caches (real and synthetic-Gaussian variance ladder); LoRA adapter checkpoints (baseline and the two final-configuration seeds); per-run evaluation outputs including the two frozen pre-registered external looks on the GrandQC MPP10 cohort with their canvases; and slide-level heatmap arrays for the baseline and endpoint models. See DEPOSIT_MANIFEST.md for per-archive contents, md5 checksums, and the mapping to the paper's sections. Code and the curated laboratory record: https://github.com/kmouts/condnoise-histoqc (doi:10.5281/zenodo.22702198). Companion study: arXiv:2608.30835.