Tag
This paper presents a decision-theoretic framework for detecting data leakage in predictive models using only model outputs and outcomes, proving that certain leakage types can be identified without external benchmarks or training code.