Explanation Correctness Thresholds

Auto-generated stub. Edit this file to add more details.

Empirically determined thresholds in explanation correctness (e.g., 70%) below which further degradation does not significantly impact human performance or learning of the model’s decision patterns.

Why It Matters

The finding that functional correctness degradation beyond a certain (70%) threshold does not cause further human understanding loss is a critical insight for metric design.

Evidence

performance dropped at 70% and 55% correctness relative to fully correct explanations, while further degradation below 70% produced no additional loss

Related Papers

• openalex-2603.25251-does-explanation-correctness-matter-linking-computational-xa