Brief: Code complements weights and prompts

Type: types/write-brief.md

Intent

Develop the general distinction between model-mediated and runtime-executed operations, so readers can reason about their composition without confusing an artifact's form or origin with how it operates. Complement the bookkeeping note's concrete argument with conceptual precision, not a second sales pitch.

Must keep

• Make the distinction usable for model-generated code and code supplied as prompt context, including the same artifact serving both roles.
• Preserve the role of weights alongside prompts and the limited sense of execution independent of model interpretation. Do not imply independence from requirements, inputs, or the runtime.
• Remain independently citable. Point to the bookkeeping note for the concrete reliability argument; allow enough overlap that neither requires reading the other first.

Must exclude

• A natural-language-versus-code dichotomy or a global ranking of the two operation classes: either would defeat the intended distinction.
• A parallel treatment of bookkeeping's leverage, scheduling architectures, or error-correction evidence. Those belong to the companion's argument.
• Defensive disclaimers against hostile readings. Keep conditions that explain the distinction; do not repeatedly deny claims the note never makes.