Training large language models to reason with search engines via reinforcement learning is hindered by a fundamental credit assignment problem: existing methods such as Search-R1 provide only a sparse outcome reward after an entire multi-step trajectory, making it infeasible to attribute success or failure to individual reasoning and retrieval decisions. Process-reward methods like StepSearch alleviate this by introducing step-level supervision, but rely on heuristic rewards such as TF-IDF overlap with gold documents, and still sample $k$ complete trajectories per example, retaining high gradient variance. We propose SLATE, a framework built on two complementary ideas: (1) truncated step-level sampling, which generates $k$ trajectories that share a common prefix and differ only at the next step, isolating variation to a single decision point; and (2) dense, decomposed LLM-as-judge rewards, which score each reasoning step, search query, and answer on a ternary scale with separate quality dimensions, providing richer supervision than binary outcome signals or undifferentiated step-level judgments. We theoretically prove that under the same dense reward structure, truncated sampling reduces the variance of advantage estimates by up to a factor of $T$ compared to full-trajectory sampling for $T$-step trajectories, yielding lower-variance and better-targeted policy gradients. Experiments on seven QA benchmarks confirm that SLATE consistently outperforms both sparse-reward and process-reward baselines, with the largest gains on harder multi-hop tasks and smaller models.

翻译：暂无翻译