BenchTrace: A Benchmark for Testing Reflection Ability and Controlled Evolution in LLM Agents

2026-05-27

AI Summary

This work addresses the limitations of current evaluation methods for large language model (LLM) agents, which rely primarily on task scores and fail to adequately assess the quality of agent reflection or provide controlled testing mechanisms for specific failure modes. To bridge this gap, we propose BenchTrace—the first model-agnostic benchmark framework for evaluating LLM agents’ reflective and evolutionary capabilities—featuring a snapshot-reflection dataset with 1,821 annotated trajectories and introducing a novel metric, Failure Avoidance Rate (FAR). Through targeted questioning, controlled evolution simulations, and cross-task analysis, BenchTrace systematically measures an agent’s ability to learn from failures. Experiments reveal that state-of-the-art models such as Qwen3-32B and GPT-4.1 achieve end-to-end reflection success rates below 30%, with diagnostic reasoning identified as the primary bottleneck; FAR improves significantly only when reflections are fully correct, while self-evolution remains vulnerable to catastrophic forgetting and negative transfer.

Abstract

Self-evolving agents improve over time by reflecting on past failures, but existing evaluation is limited in two ways: it measures only task scores, leaving reflection quality unknown, and it relies on agents' own episode runs, offering no mechanism to target specific failure patterns. We present \textbf{BenchTrace}, a benchmark for evaluating self-evolution ability in LLM agents. BenchTrace is built on a snapshot-reflection dataset of 1,821 annotated episodes spanning six diverse tasks, and comprises a \textbf{Reflection Evaluation} that probes failure identification through targeted QA tasks, and an \textbf{Evolution Evaluation} that tests whether past failure experience translates into avoidance behavior in a controlled self-evolution simulation. Building on BenchTrace, we propose \textbf{failure avoidance rate (FAR)}, a new evaluation metric measuring the fraction of test cases in which the agent successfully avoids the target failure instance. Experiments with Qwen3-32B and GPT-4.1 reveal that both models fall below a 30\% end-to-end pass rate on reflection evaluation, with diagnosis as the primary bottleneck. Evolution evaluation shows that self-evolution methods generally improve FAR over the non-evolving baseline, but agents forget early lessons as noise episodes accumulate, and agents fail to generalize their reflections beyond the specific context, causing negative transfer across task contexts. Our correlation analysis further reveals that only a fully correct reflection is strongly associated with higher FAR. BenchTrace exposes concrete limits of current self-evolution approaches and provides a controlled, model-agnostic framework for targeted evaluation.

Problem

• self-evolving agents
• reflection ability
• evaluation benchmark
• failure patterns
• controlled evolution

Innovation

• BenchTrace
• reflection evaluation
• controlled evolution
• failure avoidance rate
• self-evolving agents