On Randomness in Agentic Evals

TL;DR

This paper reveals significant variability in agentic system evaluations due to randomness, emphasizing the need for multiple runs and statistical methods to reliably measure true performance improvements.

Contribution

It provides empirical evidence of high variance in agentic evaluation scores and offers concrete recommendations for more reliable assessment practices.

Findings

Single-run pass@1 scores vary by 2.2 to 6.0 percentage points.

Trajectory divergence occurs early, within the first few tokens.

Small score differences often reflect noise, not true progress.

Abstract

Agentic systems are evaluated on benchmarks where agents interact with environments to solve tasks. Most papers report a pass@1 score computed from a single run per task, assuming this gives a reliable performance estimate. We test this assumption by collecting 60,000 agentic trajectories on SWE-Bench-Verified, spanning three models and two scaffolds. We find substantial variance: single-run pass@1 estimates vary by 2.2 to 6.0 percentage points depending on which run is selected, with standard deviations exceeding 1.5 percentage points even at temperature 0. This variance has critical implications: reported improvements of 2--3 percentage points may reflect evaluation noise rather than genuine algorithmic progress. Through token-level analysis, we show that trajectories diverge early, often within the first few percent of tokens, and that these small differences cascade into different solution strategies. To enable reliable evaluation of agentic systems, we recommend three concrete practices: (1) estimate pass@1 from multiple independent runs per task, especially when measuring small improvements, (2) use statistical power analysis to determine the number of runs needed to detect expected effect sizes, and (3) consider metrics like pass@k (optimistic bound) and pass^k (pessimistic bound) with k>1 to better characterize the full performance envelope. While these practices increase evaluation cost, they are essential for distinguishing genuine scientific progress from statistical noise.