Teach the Magnitude, Not the Direction: Verifier-Bounded Credit Assignment for Multi-Turn Multi-step LLM Agents

CrEST implements a hierarchical credit assignment framework to resolve the dilution of trajectory-level rewards in multi-turn tool-use agents. By combining turn-segmented verified advantages with entropy-gated self-teacher modulation, the method outperforms standard RL and distillation baselines on BFCL V3 and WildToolBench benchmarks across multiple model scales.