Paired one-percent-total-budget results for frozen official-DINOv2 probes. Each line connects held-out macro-F1 for linear and MLP heads using the same task, seed, budget, and test partition; colors identify seeds 42–44, dashed lines show majority baselines, and each panel reports the mean paired difference. Although every paired run favors the MLP head, the spiral values remain near trivial baselines. The comparison does not establish unconditional MLP superiority for other tasks, label regimes, or backbone states.