Figure 1: Model-ranking consistency across metrics. Each point shows Spearman ρ between rankings induced by two official metrics, all direction-aligned so higher is better. Colors show correlation ranges, the dashed line marks ρ ≥0.85, and right-side annotations give metric counts and high-correlation pairs. Weak and negative correlations are common.
Overview
Studies multi-metric benchmark compression. HoloCosm jointly minimizes score-reconstruction error while constraining ranking distortion across 15 benchmark configurations and 85 component metrics.
