How reproducible are report results across multiple runs?

The structure and tool execution remain consistent, but natural-language generation may exhibit small differences due to inherent stochasticity in LLM outputs. For fully reproducible results, minimize generative components and use fixed prompts, detailed substeps, documents, and datasets.