Announcement_7

Our paper “Beyond Site Decodability: When Does a Scanner Shortcut Actually Inflate Diagnostic Performance?” (solo author) concluded review for the ML4H 2026 Findings track. While the paper was not accepted in the final decision, it received encouraging reviews—scores of 4 (Marginal Accept), 3 (Marginal Reject), 4 (Marginal Accept), 4 (Marginal Accept), and 3 (Marginal Reject)—with reviewers highlighting the importance of the Shortcut Gap metric. Our extensive rebuttal and new analyses across corrected IXI ($N=563$) and ABIDE I cohorts successfully addressed key methodological feedback, demonstrating that the gap serves as a sensitive screening diagnostic that scales from 0.074 to 0.240 under induced site-label correlation. We thank the reviewers and program chairs on OpenReview as we prepare the extended manuscript for future submission!