Benchmarking the Benchmarks: A Validity Audit of Tool-Calling Evaluation

2026-06-30

AI Summary

This study addresses a critical gap in existing tool-calling evaluation benchmarks: the lack of validation of the evaluators themselves, which risks conflating assessment artifacts with agents’ true capabilities. Through a systematic audit of four prominent benchmarks—BFCL v4, τ2-Bench, LiveMCPBench, and MCP-Atlas—the authors conduct expert review of 496 tasks, replicate experiments, and perform trajectory-level analysis, revealing an 18.5% disagreement rate between automated evaluators and human judgment. Notably, LiveMCPBench exhibits a score variance of up to 18.9 percentage points upon re-evaluation, sufficient to overturn leaderboard rankings. To address these issues, the work introduces the first unified taxonomy of tool-calling evaluation failures, advocates for distinct measurement of tool invocation, task completion, and result verification, and releases Tool-Veritas—a configurable benchmark—and Harness Lab, an open-source evaluation platform.

Abstract

Tool-calling benchmarks are increasingly used to rank language-model agents, yet their scores are often treated as ground truth without validating the evaluators themselves. We present a systematic validity and reproducibility audit of four major tool-calling benchmark families: BFCL v4, τ2-Bench, LiveMCPBench, and MCP-Atlas. Across 496 expert-reviewed benchmark tasks, we find 92 evaluator-human disagreements, corresponding to an 18.5% misalignment rate. The failures are not isolated annotation mistakes: deterministic benchmarks exhibit brittle state matching, trajectory lock-in, incorrect ground truths, substring-based communication failures, and reward-basis misalignment, while LLM-judge benchmarks exhibit rubric drift, hallucinated completion, answer-only scoring, and substantial run-to-run variance. In LiveMCPBench, 23 repeated evaluations of the same setup produce scores ranging from 57.9% to 76.8%, a spread of 18.9 percentage points, large enough to change leaderboard conclusions. These results show that current tool-calling scores can reflect evaluator artifacts rather than agent capability. We introduce a unified taxonomy of tool-calling evaluation failures, release trace-level audit artifacts and corrected evaluation components, and argue for decomposed metrics that separately measure tool invocation, task completion, and outcome verification. Our findings suggest that progress in tool-using agents requires benchmarks whose evaluators are themselves reproducible, auditable, and aligned with human judgments of task success. We further introduce Tool-Veritas, a configurable benchmark that combines deterministic state verification with optional qualitative judging, and Harness Lab, an open-source system for benchmark execution, trace inspection, repeated-run comparison, and evaluator debugging.

Problem

• tool-calling evaluation
• benchmark validity
• evaluator alignment
• reproducibility
• LLM benchmarks

Innovation

• tool-calling evaluation
• benchmark validity
• evaluator alignment
• reproducibility audit
• decomposed metrics