Toward Scalable Verifiable Reward: Proxy State-Based Evaluation for Multi-turn Tool-Calling LLM Agents

TL;DR

This paper introduces Proxy State-Based Evaluation, a scalable LLM-driven simulation framework for benchmarking multi-turn tool-using LLM agents, providing reliable, on-policy evaluation without costly deterministic backends.

Contribution

It proposes a novel proxy state-based framework that enables scalable, reliable evaluation of LLM agents in multi-turn interactions without deterministic databases.

Findings

Produces stable, model-differentiating rankings.

Transfers supervision to unseen scenarios.

Achieves over 90% human-LLM judge agreement.

Abstract

Interactive large language model (LLM) agents operating via multi-turn dialogue and multi-step tool calling are increasingly used in production. Benchmarks for these agents must both reliably compare models and yield on-policy training data. Prior agentic benchmarks, such as tau-bench, tau^2-bench, and AppWorld, rely on fully deterministic backends, which are costly to build and iterate. We propose Proxy State-Based Evaluation, an LLM-driven simulation framework that preserves final state-based evaluation without a deterministic database. Specifically, a scenario specifies the user goal, user/system facts, expected final state, and expected agent behavior, and an LLM state tracker infers a structured proxy state from the full interaction trace. LLM judges then verify goal completion and detect tool/user hallucinations against scenario constraints. Empirically, our benchmark produces stable, model-differentiating rankings across model families and inference-time reasoning efforts, and its on-/off-policy rollouts provide supervision that transfers to unseen scenarios. Careful scenario specification yields near-zero simulator hallucination rates, as supported by ablation studies. The framework also supports sensitivity analyses over user personas. Human-LLM judge agreement exceeds 90%, indicating reliable automated evaluation. Overall, proxy state-based evaluation offers a practical, scalable alternative to deterministic agentic benchmarks for industrial LLM agents.