arxiv6.0 / 10

Variance-adaptive confidence regions for fixed-benchmark binomial mixtures, with application to LLM evaluation

Kuancheng Ye

Abstract

We construct confidence intervals for linear functionals $θ(F_m)=\int h\,dF_m$ of the empirical law $F_m$ of success probabilities $p_1,\dots,p_m$, from independent counts $c_i\sim\mathrm{Bin}(n,p_i)$, as for a language model run $n$ times per problem. Coverage is at least $1-α$ for every $n$, $m$ and fixed $(p_i)$, simultaneously for all bounded $h$. The variance-adaptive (VA) region tests each candidate mixing law with the conditional variance of the tail counts it implies. This variance is linear in the law, so the region is a linear program, and cell envelopes keep it valid off-grid. Regions calibrated for iid sampling target a super-population law and are conservative for $θ(F_m)$. With all $n$ tail counts, fixed $n$, continuous $h$ and a grid refined as $m\to\infty$, the identification part of VA's width is first-order worst-case optimal, as for any localization with the same index set. Under a rule fixed before the main comparator runs, VA was narrower than a two-stage plug-in band at every $n$ (median width ratio 0.60 at $n=128$), but not than the asymptotic, pointwise AMARI interval at $n=128$; on real rollout pools it was wider than both pointwise and Bonferroni-adjusted AMARI at $n\le128$. For $P(p>0)$ and pass@$k$ with $k\gg n$ the data support only lower bounds.

Research area

evaluationsLLM evaluationuncertainty quantification
Published
16 Nov 2026
Source
arxiv
Org
University of California, Irvine
View paper
Sign in to read and join the discussion.