StrictPRMBench is a replayable evaluation benchmark designed to audit fixed‑score PRM aggregation rules. It standardizes mathematical reasoning pools, scorer evidence, and deletion interventions so that different reducers confront identical selection problems. Across seven checkpoints and two benchmarks, the benchmark reveals that product aggregation yields higher deletion‑induced selection instability than geometric mean, with statistically significant differences in most cells.
arXiv:2608. 08008v1 Announce Type: new Abstract: Process reward models (PRMs) score intermediate reasoning steps and are widely used for search, ranking, and training, but optimization can exploit these learned proxies by increasing reward while turning correct reasoning into incorrect reasoning.
The paper investigates how language‑model judges can make version‑dependent errors when evaluating upgraded agents. Using 35 public coding‑agent submissions, two customer‑service agents, and over a thousand expert‑labeled trajectories, the authors show that fixed judges often reject task‑conditioned error invariance and can incorrectly approve failed patches, especially as agent capability increases. Paired audits of current outputs reduce interval width only marginally, and the study concludes that independent human patch review is still necessary.
The paper investigates whether providing candidate solutions during test‑time aggregation improves or harms accuracy compared to a fresh solve that does not use any candidates. Using Qwen3‑4B on AIME‑2025 and HMMT‑2025, the authors find that conditioning on multiple correct candidates boosts accuracy (+0.290), while conditioning on an all‑wrong candidate pool reduces accuracy (−0.123); the effect for a single correct candidate remains unclear. The study also explores structured interventions and placebo controls, but the underlying mechanisms of these effects are not resolved.
arXiv:2607. 17136v1 Announce Type: cross Abstract: Agentic computer-use RL is reported in single runs, and those numbers mislead.
arXiv:2609.24194v1 Announce Type: new Abstract: Evaluation scores used around LLM systems -- including reward models, rerankers, and LLM judges -- can track surface form instead of the quality they c...
arXiv:2608. 08944v1 Announce Type: cross Abstract: A failed retrieval-augmented generation (RAG) answer can be consistent with several unseen responses to evidence repair.
arXiv:2607. 03386v1 Announce Type: new Abstract: Agentic AI systems are increasingly used to edit, refine, and repair decision policies, but evaluating these edits is difficult when per-state expert action labels are unavailable.
arXiv:2607. 28871v1 Announce Type: cross Abstract: When a repair agent runs a test and sees it pass, the result is treated as evidence about the reported defect.
arXiv:2606. 31511v1 Announce Type: cross Abstract: In deployment settings where retraining is infeasible, small frozen code models are routinely asked to repair a failed program after seeing their own failing output, usually treated as a retry mechanism.
The paper introduces TRACE, a protocol designed to diagnose whether changes in verifier scores for large language model agents reflect actual changes in agent behavior or merely alterations in the evaluation process. TRACE works by applying targeted changes to evaluation components, running paired experiments, and re‑scoring unchanged trajectories to isolate the source of score variation. Experiments on synthetic tasks and public benchmark tasks demonstrate that seemingly significant score shifts can often be attributed to evaluation artifacts, while TRACE can also detect genuine behavioral changes.
arXiv:2609.06107v1 Announce Type: new Abstract: Data policies for reinforcement learning with verifiable rewards (RLVR) determine which rollouts are used, how strongly they are weighted, and which do...
arXiv:2608. 07813v1 Announce Type: new Abstract: An LLM judge deployed inside a reasoning pipeline does not merely measure quality, it decides which answer ships.