SLCA-GRPO: Resolving Cross-Segment Credit Misattribution in Tool-Calling RL

A new method called SLCA-GRPO addresses cross-segment credit misattribution in tool-calling reinforcement learning, targeting a structural failure mode where standard on-policy RL algorithms such as GRPO indiscriminately broadcast a single reward signal across heterogeneous outputs that interleave structured tool invocations with user-facing natural language summaries.

Tool-calling agents produce heterogeneous outputs, interleaving structured tool invocations with user-facing natural language summaries. This output heterogeneity presents a structural failure mode in standard on-policy Reinforcement Learning RL : algorithms like GRPO indiscriminately broadcast a h