{"id":"arxiv/2609.05286","url":"https://arxiv.org/abs/2609.05286","tags":["math.ST"],"score":0,"title":"Statistical reduction before the target is known: two boundary results","domain":"math.ST","source":"arxiv","authors":["Rianne de Heide"],"reviews":0,"subject":"math.ST","summary":"Suppose that the eventual use of data is not known when the data are reduced or collected. This note considers two simple boundary cases. In a finite statistical experiment, a statistic preserves the Bayes risk for every finite later decision problem if and only if it is sufficient. Hence, when the minimal sufficient statistic is one-to-one, exact preservation of all later decision problems permits no nontrivial reduction. We then consider adaptive sampling from $m$ independent Gaussian streams when an external query specifies the coordinate to be classified only after sampling stops. Under…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-04","reactions":{}}