Numeric artifacts for "Auditing Metric Dilution and Optimization Coupling in Mixed-Label YOLO Pose Fine-Tuning"

Numeric artifacts underlying every table and figure of the accompanying article. The deposit contains the frozen evaluation matrix (15 fine-tuned runs, each evaluated on a target split and a fine-tuning-disjoint reference pool, at both the framework-default best checkpoint and a uniform epoch-60 checkpoint under one recorded validation protocol), per-run training logs and recorded arguments, the invariant-checkpoint metric audit, layer-wise centered kernel alignment values, the reference-pool construction audit and image-ID manifest, and desktop-CPU deployment measurements. Model checkpoints are not included because of their size, and COCO photographs are not redistributed. Local filesystem paths inside the JSON records were replaced with placeholders before deposit; no other content was altered.