Anti-p-hack Protocol Retains Detection Power but Requires ~10-Day Wait

krisgligoric · x · 2026-07-06

Research indicates that this preregistration protocol still retains the ability to detect true effects: when a real effect is injected, the detection results are comparable to repeated measurements on the same model. The main tradeoff is the need to wait for the next model release, with a median wait time of about 10 days in the study.

Related event: Study Proposes Preregistration Protocol to Curb LLM Evaluation P-Hacking(7 posts)→

Original post →

More from Research

Research channel →