When Do Agent Loops Mistake Stagnation for Progress? Self-Evaluation Bias and Externally Grounded Verification in Long-Running Autonomous LLM Agent Loops

• URL: http://arxiv.org/abs/2607.25152v1
• Date: Mon, 27 Jul 2026 23:52:15 GMT
• Title: When Do Agent Loops Mistake Stagnation for Progress? Self-Evaluation Bias and Externally Grounded Verification in Long-Running Autonomous LLM Agent Loops
• Abstract summary: Long-running autonomous agents plan, act, and judge their own completion without human intervention.<n>When an agent grades its own work, self-evaluation bias takes hold: plausible changes are accepted as progress while real-world outcomes stagnate or regress.<n>We build a testbed that holds the agent and its tool surface fixed and manipulates only the information-channel type of the evaluator that gates the loop.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Long-running autonomous agents plan, act, and judge their own completion without human intervention. When an agent grades its own work, self-evaluation bias takes hold: plausible changes are accepted as progress while real-world outcomes stagnate or regress. We name this failure mode the progress mirage and show, with controlled measurement, that it is a question of what the evaluator is grounded in. We built a testbed that holds the agent and its tool surface fixed and manipulates only the information-channel type of the evaluator that gates the loop. A world-state oracle, unfakeable in principle, is enforced by container and network isolation and verified at every run. Across 54 cycles a frontier agent claimed improvement every time, yet 56 percent had a measured delta of zero or below. Self-report was thus uninformative, and the self-verdict gate degenerated into accept-all, eroding the best deployed state it had reached by 19 percent. Even the strongest in-band judge, reading the full artifact text, the change diff, and its own verdict history, accepted cycles of which 44 percent were real-world regressions and rejected 38 percent of real improvements; the preregistered adversarial hypothesis that a strong judge closes the gap was rejected. On a boundary task whose success specification is verifiable from the artifact itself, the same judge's mirage vanished to zero and the gap collapsed within the registered threshold, showing that the gap depends on where the success signal resides. A sign-only variant returning only the acceptance verdict kept real-world output similar to full feedback (110.0 versus 113.0), locating the benefit in the gate's grounding rather than in feedback content. For open-ended objectives whose success signal lives outside the transcript, scaling up the judge is not enough; out-of-band evaluation with real-world access is a structural requirement.

Related papers

• VeriPhy: Agentic Physical Reasoning for World Model Evaluation and Refinement [57.86962208273888]
  We present an auditable physical-verification system in which a text-only planner compiles the prompt into typed physical obligations.<n>Each action returns a provenance-carrying evidence record whose payload, when usable, is either a typed measurement or an explicitly tagged learned state.<n>We anchor evaluation in a 1,500-clip corpus of human-annotated flaw records that localize real generation failures in prompt reference, space, and time.
  arXiv  Detail & Related papers  (2026-09-02T20:36:45Z)
• trajectory-judge: What Outcome-Only LLM Judges Miss on Agent Trajectories [0.0]
  Outcome-only evaluation is the production default for LLM agents.<n>We measure that blind spot where ground truth is known by construction.<n>We argue that judge evaluations must stratify recall by outcome survival.
  arXiv  Detail & Related papers  (2026-08-29T10:14:05Z)
• SimVerity: When Does Simulated Agent Success Survive Physical Deployment? [5.081228499547384]
  We present SimVerity, a verdict-transfer assurance framework.<n>It replays matched scenarios on target smart home deployments.<n>It crosses-validates agent execution against independently qualified physical witnesses.
  arXiv  Detail & Related papers  (2026-08-25T19:00:49Z)
• HarnessEval-W: Agentifying the Evaluation of Visual Worlds [149.63310396189317]
  HarnessEval-W is an agentified evaluation pipeline for benchmarking world models.<n>Rather than applying a fixed rubric, HarnessEval-W decomposes the evaluation question into measurable subproblems.<n>We apply HarnessEval-W to 18 representative world models over 330 evaluation cases.
  arXiv  Detail & Related papers  (2026-08-17T17:43:24Z)
• The LLM Proposes, the Executive Disposes: A Self-Verifying Agent Instrument that Dissociates Commitment Drift from Binding Drift in Long-Horizon Agents [0.0]
  We present an agent instrument built so that verification is structural rather than post-hoc.<n>Every run invalidates itself when per-organ write-error, render-size, or salted-canary-echo floors are breached.<n>We report a clean, single-variable result on a failure every long-horizon agent suffers.
  arXiv  Detail & Related papers  (2026-08-04T15:10:37Z)
• Self-Authored Verification Is Unreliable in Heuristic Self-Improving Agents [5.91817208489443]
  We study the verifier-acceptance-deployment gap.<n>This gap refers to the discrepancy between an agent's self-authored verification signal and a sealed deployment evaluation.<n>We introduce a Sealed Exogenous Acceptance Loop (SEAL)
  arXiv  Detail & Related papers  (2026-07-27T11:45:14Z)
• What Actually Works for Spacecraft Fault-Tolerant Control: An Honest Settled-Gate Benchmark of Learned and Classical Methods [2.28438857884398]
  Recent learned fault-tolerant-control (FTC) work reports high success on spacecraft actuator faults.<n>We ask what recovers spacecraft pointing when success means holding it on faults never seen in training.<n>We answer with a benchmark built around a settled gate, pointing held within 0.2 deg over a dwell window and scored on the true state.
  arXiv  Detail & Related papers  (2026-06-24T04:08:39Z)
• Done, But Not Sure: Disentangling World Completion from Self-Termination in Embodied Agents [11.233308795768465]
  VIGIL is an evaluation framework that makes terminal commitment independently measurable.<n>Under VIGIL's default protocol, agents observe only egocentric RGB, receive no action-success signals, and must end each episode with a semantic report checked deterministically against hidden world state.<n>This yields two separate scores: world-state completion (W) and benchmark success (B)
  arXiv  Detail & Related papers  (2026-05-09T07:24:07Z)
• AgentProcessBench: Diagnosing Step-Level Process Quality in Tool-Using Agents [50.481033105867205]
  We introduce AgentProcessBench, the first benchmark dedicated to evaluating step-level effectiveness in realistic, tool-augmented trajectories.<n>The benchmark comprises 1,000 diverse trajectories and 8,509 human-labeled step annotations with 89.1% inter-annotator agreement.<n>It features a ternary labeling scheme to capture exploration and an error propagation rule to reduce labeling ambiguity.
  arXiv  Detail & Related papers  (2026-03-15T16:13:58Z)
• Automated Self-Testing as a Quality Gate: Evidence-Driven Release Management for LLM Applications [51.56484100374058]
  We present an automated self-testing framework that introduces quality gates with evidence-based release decisions.<n>We evaluate the framework through a longitudinal case study of an internally deployed multi-agent conversational AI system.
  arXiv  Detail & Related papers  (2026-03-13T20:44:15Z)
• A Rubric-Supervised Critic from Sparse Real-World Outcomes [87.11204512676193]
  Real-world coding agents operate with humans in the loop, where success signals are typically noisy, delayed, and sparse.<n>We propose a process to learn a "critic" model from sparse and noisy interaction data, which can then be used both as a reward model for either RL-based training or inference-time scaling.
  arXiv  Detail & Related papers  (2026-03-04T07:23:54Z)
• A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness [57.510025257780306]
  We show that existing validation protocols fail to account for substantial distribution shifts inherent to red-teaming.<n>We propose ReliableBench, a benchmark of behaviors that remain more consistently judgeable, and JudgeStressTest, a dataset designed to expose judge failures.
  arXiv  Detail & Related papers  (2026-02-04T15:13:35Z)
• Gaming the Judge: Unfaithful Chain-of-Thought Can Undermine Agent Evaluation [76.5533899503582]
  Large language models (LLMs) are increasingly used as judges to evaluate agent performance.<n>We show this paradigm implicitly assumes that the agent's chain-of-thought (CoT) reasoning faithfully reflects both its internal reasoning and the underlying environment state.<n>We demonstrate that manipulated reasoning alone can inflate false positive rates of state-of-the-art VLM judges by up to 90% across 800 trajectories spanning diverse web tasks.
  arXiv  Detail & Related papers  (2026-01-21T06:07:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.