Presented at R for HTA 2026. A published health economic model description is incomplete by design: it assumes the reader brings the tacit modeling knowledge that was never written down. Hand that same description to an LLM and it fills those gaps silently, producing code that runs whether or not the interpretation is right. This talk lays out a decomposition framework that breaks the translation into discrete, inspectable stages and surfaces the points where a model description is ambiguous. Used this way, the LLM becomes a research companion that keeps the modeler in control of every judgment call and enables full transparency, traceability, and reproducibility, so the resulting R code is something you can actually trust.