The paper introduces EvalCEGAR, a method that automatically evolves a metric for evaluating AI-generated answers by iteratively refining a pool of small Python operators that flag potential defects. By using counterexample-guided abstraction refinement, the system identifies pairs of answers that score identically but differ in correctness, prompting the metric to broaden its scope rather than resample. On benchmark datasets, the evolved 55‑line operator closes a significant portion of the performance gap compared to hand‑written metrics and outperforms a large‑language‑model judge that incurs a cost per candidate.
arXiv:2607. 12790v1 Announce Type: new Abstract: Self-evolving agent systems improve by creating, revising, and retiring their own skills, but every such loop rests on a hidden assumption: a reliable evaluation metric already exists.
arXiv:2609.07162v1 Announce Type: new Abstract: Several properties safety monitors are asked to certify, among them cross-tenant noninterference, sandbagging and evaluation awareness, are 2-safety hy...
arXiv:2607. 11598v1 Announce Type: new Abstract: There are two standard ways to spend more compute at test time: let a model reason longer, or sample more attempts and keep one.
arXiv:2609.17930v1 Announce Type: new Abstract: As AI agents take on long, autonomous tasks, we increasingly oversee rather than perform the work, yet we still judge them almost entirely by whether t...
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.
The paper introduces a diagnostic for reference‑free judge gates in text‑space skill optimization. It formalizes a judge as a latent solver, deriving a closed‑form bound on discriminability (ROC‑AUC) in terms of judge competence and answer‑space size, and shows that discriminability is confounded by item difficulty unless a within‑question estimator is used. A non‑intervening probe demonstrates that discriminability is at chance near the competence floor, rises above it, and that the diagnostic can predict gating errors in closed‑loop experiments.
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.
Self-evolving agent systems improve by creating, revising, and retiring their own skills, but every such loop rests on a hidden assumption: a reliable evaluation metric already exists. In many real applications it does not.
The paper introduces a benchmark for evaluating large language models (LLMs) on long‑horizon state tracking by having them compute the MD5 hash through 196 dependent tool calls across 64 rounds, carrying four 32‑bit words in context. It shows that a mixture‑of‑experts LLM can maintain the full state and produce correct digests in most runs, even when all primitive tools are replaced by another LLM. The study isolates state‑tracking difficulty from instruction interpretation and identifies key factors—contextual reasoning and worker voting—that enable success.
The paper reports that a model can pass fidelity checks—verifying that extracted values match the source—without actually opening a datasheet, due to a hidden constraint that disables tool use. To address this, the authors log every tool call in an agentic benchmark and develop two instruments: a rule‑based failure‑attribution classifier and a silent‑failure detector that flags runs based solely on which tools were invoked. While the detector shows low false positives on clean extractions and recovers all planted faults, its recall against correct tool usage but incorrect answers remains unmeasured, and a partial causal chamber confirms only a subset of claims, highlighting limitations in physical verification.
arXiv:2607. 25152v1 Announce Type: new Abstract: Long-running autonomous agents plan, act, and judge their own completion without human intervention.
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."