cs.AIOct 31, 2026

Few Trials, Many Metrics: Beta Mixtures for Low-Budget AI Evaluation

Authors: Gustavo Tironi, Felipe Maia Polo, Diego Mesquita

Organizations: Getulio Vargas Foundation · Institute of Foundation Models, MBZUAI

Abstract

Evaluating a stochastic AI system often requires many repeated samples per instance, especially for metrics defined over long horizons. Recent work has shown that modeling instance-level success probabilities with a Beta distribution recovers Pass@kk from far fewer samples than direct estimation. We show that this framework supports a broader range of evaluation targets. We fit a mixture of Beta distributions by marginal maximum likelihood, with the number of components chosen by BIC. From that one fit, we compute Passk^k and AUC@kk at any horizon in closed form. We also derive delta-method confidence intervals for these metrics and for Pass@kk, which require no additional model calls. Treating configurations rather than tasks as instances extends the same model to sensitivity analysis. For repeated success metrics, we find that mixture priors yield more accurate estimators than naive estimation across competition mathematics and graduate-level science benchmarks when extrapolating beyond the observed number of trials, and the resulting intervals cover the full-data reference. When prompts are treated as instances, it estimates the distribution of prompt sensitivity more accurately than a comparable item-response baseline without external covariates. Direct estimators remain preferable once trials reach the target horizon.

Figures & tables

Appendix figures & tables15 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Nov 18, 2026

Fast Bayesian Methods for Uncertainty Quantification in Sparse AI Evaluation

Estimating how an AI system will perform on a new benchmark need not require another large model. We use public evaluation counts to choose a weak Beta prior, then update it exactly from at most 31 revealed outcomes. The predictor uses no item semantics, pretrained weights, or inference API. Its single online evaluation in the AIMS competition returned a team-reported Brier ALC of 0.188477, a 24.6% reduction from the constant-0.5 reference. The complete submission is 7,754 bytes; a server microbenchmark measures 12.71 microseconds per prediction with 31 matching labels. We describe how source evaluations are compressed into 301 prior means and show why a concentration of two lets new evidence rapidly determine the prediction. The resulting posterior gives exact uncertainty about the modelled success rate without numerical integration or neural inference. A public evaluation covers 79 system-benchmark pairs and eight seeds, reaching Brier ALC 0.200250 for the submitted method. Accompanying code and sufficient statistics reconstruct the fitted prior byte for byte and replay all 632 evaluation trajectories. The result demonstrates how much predictive value a compact Bayesian model can extract from sparse target feedback at very low online cost.
Jul 16, 2026

Can We Trust Item Response Theory for AI Evaluation?

AI benchmarks increasingly leverage item-level statistical models, particularly item response theory (IRT), to estimate capabilities, rank systems, select informative examples, and diagnose benchmark quality. However, AI benchmark data often departs from the data regimes encountered in human testing, for which standard IRT estimation tools were originally developed: these benchmarks typically involve fewer evaluated models, far more items, and capability distributions that may be skewed, clustered, or multimodal. We examine how these regime mismatches challenge the reliability of IRT modeling for AI evaluation. Using item parameters and capability distributions derived from six widely used LLM benchmarks, we simulate response matrices under three common IRT models and compare four estimators used in recent benchmark studies: marginal maximum likelihood, Markov chain Monte Carlo, variational inference, and a neural pseudo-Siamese estimator. Across 18,000 simulation runs, we systematically evaluate computational feasibility, scalability, and the reliability of IRT inferences about model rankings, predicted performance, and item characteristics. Results show that classical estimators can become computationally infeasible for large benchmarks, whereas scalable estimators can produce unreliable item-level and ranking inferences with small or non-normally distributed model sets. This study identifies when latent trait models reliably support or risk distorting AI benchmarking claims, and what sample sizes and diagnostics are needed for trustworthy use.
Sep 20, 2026

How to Run Statistics over LLM Judges and Trust the Results: Calibrated Inference for Small-Sample AI Evaluation with evalstats

Researchers across academia increasingly base significance claims on LLM judge scores and small-sample AI evaluations. Yet without well-calibrated confidence intervals (CIs), hypothesis tests, and judge-bias corrections, such claims are unreliable. We address these issues in several contributions. First, we find that running statistics over raw LLM judge scores leads to inflated false positives: counterintuitively, for many inter-rater agreement metrics, false positive risk peaks at "almost perfect" human-LLM agreement. To help researchers understand how to run statistics over LLM judges responsibly, we present guidance and tooling for the statistical analysis of mixed human-AI judge designs, and implement nine hypothesis tests via prediction-powered inference (PPI), including the first known PPI corrections for four rank-based tests (Wilcoxon signed-rank, Mann-Whitney U, and omnibus variants). To keep PPI++ stable with small human-labeled calibration sets, we introduce bootstrap-adaptive power tuning, which shrinks the estimated weight toward a target estimated from the labeled data, and accounts for that weight's own sampling variance. Second, through Monte Carlo simulations, we derive recommendations for what CI, p-value, and FWER correction methods to use for small-sample AI evaluations (N<100), and warn researchers against bootstrap CIs. We package these recommendations into evalstats, an open-source Python package that selects calibrated methods automatically, and demonstrate it in three scenarios, including one where a real LLM judge validated at "substantial agreement" would have led a researcher to publish a spurious finding. evalstats is publicly available at https://github.com/ianarawjo/evalstats.