What is an LLM input package?

An LLM input package is the controlled interface between governed report data and a language model.

Its purpose is to allow language generation without transferring analytical authority to the model. The package contains only bound content: approved statements, structured values, limitations, citation references, section constraints, and forbidden outputs. It defines what the model may verbalize and what it must not create.

In SDX, a language model may help turn machine-readable report data into readable institutional language. It may not decide, aggregate evidence, invent claims, reinterpret uncertainty, or create new analytical meaning.

The LLM input package therefore acts as a boundary object. It allows language to be produced while keeping governance, evidence, and decision authority outside the model. This protects the report from becoming an AI-generated interpretation rather than a rendering of controlled evidence.