{"id":"arxiv/2610.16789","url":"https://arxiv.org/abs/2610.16789","tags":["stat.ME","math.ST"],"score":0,"title":"Adaptive Cauchy inference from invariant information","domain":"stat.ME","source":"arxiv","authors":["Yiming Liu","Shaochen Wang"],"reviews":0,"subject":"stat.ME","summary":"Selecting weights and projection directions from the observations used for testing can invalidate a reference distribution. We develop a framework for exact conditional inference under rotation symmetry. A maximal invariant permits adaptation while leaving a uniform common phase for calibration. This yields explicit Cauchy references for invariantly selected weights and projections. Under separate block rotations, the same principle gives exact normal-score combinations with coefficients selected from all blockwise invariants. The calibration results require neither densities nor moments. To…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-10-12","reactions":{}}