# SLCA-GRPO: Resolving Cross-Segment Credit Misattribution in Tool-Calling RL > Source: > Published: 2026-09-28 02:30:20+00:00 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