Agent psychometrics: Task-level performance prediction in agentic coding benchmarks

TL;DR

This paper introduces a framework that predicts individual task success in agentic coding benchmarks by augmenting Item Response Theory with rich task features and decomposing agent ability.

Contribution

It presents a novel method combining IRT with detailed task features and ability decomposition to predict task-level performance across diverse benchmarks.

Findings

Accurately predicts success on unseen tasks and benchmarks.

Enables better calibration of task difficulty for benchmark design.

Decomposes agent ability into LLM and scaffold components.

Abstract

As the focus in LLM-based coding shifts from static single-step code generation to multi-step agentic interaction with tools and environments, understanding which tasks will challenge agents and why becomes increasingly difficult. This is compounded by current practice: agent performance is typically measured by aggregate pass rates on benchmarks, but single-number metrics obscure the diversity of tasks within a benchmark. We present a framework for predicting success or failure on individual tasks tailored to the agentic coding regime. Our approach augments Item Response Theory (IRT) with rich features extracted from tasks, including issue statements, repository contexts, solutions, and test cases, and introduces a novel decomposition of agent ability into LLM and scaffold ability components. This parameterization enables us to aggregate evaluation data across heterogeneous leaderboards and accurately predict task-level performance for unseen benchmarks, as well as unseen LLM-scaffold combinations. Our methods have practical utility for benchmark designers, who can better calibrate the difficulty of their new tasks without running computationally expensive agent evaluations.