Hacker News Front Page · [Submitted on 28 Sep 2026] ·
Targeting Pivotal Decisions for Credit Assignment in Agentic Reinforcement Learning
ProVer es un marco que mejora el entrenamiento de agentes de lenguaje grande mediante asignación fina de crédito en aprendizaje por refuerzo. A diferencia de GRPO, que asigna uniformemente ventajas a todos los tokens, ProVer identifica decisiones potencialmente pivotales usando un juez basado en modelos que contrasta trayectorias exitosas y fallidas, luego verifica esas decisiones midiendo su ventaja real a partir de las tasas de éxito en continuaciones muestreadas. Los resultados en ALFWorld, WebShop y SearchQA muestran mejoras relativas del 9.91% y 7.12% sobre GRPO en modelos de 2B y 4B, demostrando que dirigirse selectivamente a decisiones clave mejora el entrenamiento de la política con sobrecarga de generación modesta.
Group Relative Policy Optimization (GRPO) has become a promising approach for training large language model agents. However, its uniform assignment of trajectory-level advantages to all policy tokens fails to distinguish consequential decisions from less relevant ones, obscuring which intermediate decisions contributed to success. We introduce ProVer, a framework that targets potentially pivotal decisions for fine-grained credit assignment in agentic reinforcement learning. Given a rollout group, an agentic judge contrasts successful and failed trajectories to propose a segment potentially responsible for their divergent outcomes. Rather than directly trusting the judge's assessment, ProVer verifies the proposed segment by estimating its advantage from the difference in terminal success rates between current-policy continuations sampled before and after the segment. Positive estimates are then incorporated into the GRPO advantages of policy tokens within the proposed segment. By using model judgment only to select where to verify, ProVer grounds local credit in observed outcomes without exhaustively evaluating every intermediate state. Across ALFWorld, WebShop, and SearchQA, ProVer achieves the strongest average performance at both model scales, with relative improvements over GRPO of 9.91% and 7.12% for Qwen3.5-2B and Qwen3.5-4B, respectively. Further analyses demonstrate that informed segment selection improves policy training with modest additional generation overhead, even without a frontier-scale judge model, highlighting the effectiveness and efficiency of selectively targeting pivotal decisions for fine-grained credit assignment in agentic reinforcement learning.