{"id":"arxiv/2607.00224","url":"https://arxiv.org/abs/2607.00224","tags":["math.ST","cs.IT","cs.LG","stat.ML"],"score":0,"title":"Sample Complexities of Estimating Gumbel--Max Watermark Proportions with and without Reduction to Pivotal Statistics","domain":"math.ST","source":"arxiv","authors":["Shuwen Chai","Qiaosen Wang"],"reviews":0,"subject":"math.ST","summary":"Watermarking promises statistical traceability of large language model (LLM) uses, but real documents rarely arrive as purely human-written or purely LLM-generated. This motivates a quantitative question beyond detection: what proportion of a document is generated from a pre-specified watermarked LLM? We study this watermark proportion estimation problem under the Gumbel--max watermarking mechanism, treating the next-token prediction distributions as unknown and arbitrary nuisance parameters subject to a non-degeneracy condition. We compare two observation regimes: in the full observation…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-30","reactions":{}}