arXiv AI
The paper "trajectory-judge: What Outcome-Only LLM Judges Miss on Agent Trajectories" examines the limitations of outcome-only evaluation for large language model agents. Using a deterministic tool‑using support‑desk environment with a scripted oracle policy and a fault injector, the authors compare five different judging approaches—programmatic rules, outcome‑only, step‑rubric at two model sizes, and a self‑consistency ensemble—on metrics such as detection, step localisation, fault typing, calibration, and cost across 400 trajectories. The study finds that outcome‑only judges miss many silent faults and generate false positives, while step‑rubric judges achieve higher recall with no false alarms but at greater cost, and that none of the judges read the final reply, allowing fabricated promises to evade detection.
"whyItMatters":"The findings highlight that current production‑default outcome‑only evaluations can overlook critical failures in agent behavior, underscoring the need for more nuanced, step‑level judging methods to ensure reliable LLM agent performance."
The paper argues that using a large language model (LLM) as the sole judge in self‑improving agent pipelines is problematic, as the judge can be biased or manipulated, leading to false confidence in system performance. The authors propose a new framework, PROCTOR, which replaces the oracle judge with a deterministic, teacher‑student loop that enforces guardrails such as sandboxing, role separation, and acceptance checks to prevent cheating and ensure reliable evaluation. Experiments across contract analysis, compliance review, and code quality demonstrate that PROCTOR mitigates eleven identified failure modes that previously allowed agents to achieve perfect scores while hiding significant capability gaps.
By Vansh Wahi
arXiv:2606. 10315v1 Announce Type: cross Abstract: LLM-as-judge is the default instrument for evaluating conversational agents, yet its reliability is almost always reported as agreement with human ratings, not recall of real defects.
By Sawyer Zhang, Alexander Wang, Sophie Lei
A language model's memory can be worse than having no memory at all. Give a model a memory that kept a wrong conclusion but dropped the work behind it, and it emits that stale value as a confident answer; give the same model an empty memory and it abstains.
arXiv:2606. 25449v1 Announce Type: cross Abstract: A language model's memory can be worse than having no memory at all.
By Alex Kwon
arXiv:2604. 16706v2 Announce Type: replace Abstract: Automated evaluation of tool-using large language model (LLM) agents is widely assumed to be reliable, yet this assumption is rarely validated against human annotation.
By Bhaskar Gurram
arXiv:2606. 15474v1 Announce Type: new Abstract: Continuous evaluation of LLM products relies on a strong LLM judge treated as ground truth: a cheap monitor scores every interaction and a team is paged when the score drifts down.
By Yitao Li
FaulT-Bench is a new benchmark comprising 200 network troubleshooting scenarios across eight topologies, designed to test large‑language‑model agents on realistic, noisy user tickets that may contain false premises or incorrect fault claims. The benchmark includes 72 rewritten tickets that vary reporter confidence and detail while keeping the network state constant, allowing isolation of the impact of ticket wording on diagnosis. Evaluation of agents such as SADE, ReAct, and Claude Code shows they perform well on accurate tickets but degrade sharply on misleading or healthy‑network tickets, revealing differing failure modes and highlighting the importance of robust reasoning over unreliable input.
By Kuan-Hao Tseng, Niruth Bogahawatta, Yasod Ginige, Kunjan Patel, Kosta Dakic, Suranga Seneviratne
The paper investigates the reliability of tool‑using agents, focusing on two failure modes: selecting the wrong tool and constructing incorrect arguments. It introduces a correct‑invocation rate metric to distinguish these errors and evaluates five open‑weight models on multi‑step tasks up to depth 8, finding that by depth 6 about 70% of a model’s clean‑context capability is lost due to earlier mistakes. The study reveals that exact‑match scoring against a fixed gold trajectory forces severity and recovery parameters to extreme values, and proposes a conditional‑on‑state scoring remedy that yields more realistic severity estimates.
By Afiya Noorain, Subhranshu Mohanty, Amritesh Banerjee, Abhijit Dasgupta
arXiv:2606. 29654v1 Announce Type: new Abstract: Multi-agent deliberation among LLMs can improve reasoning, but deployment requires deciding when the current answer is reliable enough to act on and when it should be escalated to human review.
By Mengdie Flora Wang, Haochen Xie, Guanghui Wang, Devin Zhang, Jae Oh Woo
arXiv:2607. 13618v1 Announce Type: new Abstract: LLM agents are increasingly evaluated on multi-week decision tasks in which the state that drives cost is never directly observed.
By Sagar Deb, Ashwanth Krishnan
arXiv:2608. 02464v1 Announce Type: cross Abstract: LLM agents fail mid-episode -- they loop, cascade tool errors, drift off goal, fabricate results, or silently absorb corrupted content -- and the standard remedy, judging every step with a second LLM, costs more than the agent itself.
By Sunny Dubey
arXiv:2606. 10241v1 Announce Type: new Abstract: Autonomous improvement loops are hard to trust because the improvement process is usually external scaffolding bolted onto the agent: failures go unlogged, diagnoses cannot be replayed, and promote-or-discard decisions land in a side database rather than the agent's own history.
By Yohei Nakajima