The compile endpoint needs to accept a specification describing the function to compile. Users may want to include input/output examples. Should the API have separate fields for spec and examples?

The API takes a single spec string field. No separate examples field. Users include examples naturally in the spec text if they want. The website frontend provides structured UI (spec box + examples editor) for UX convenience, but concatenates them client-side before calling the API.

• The compiler was trained on specs that include examples as text
• The compile function internally takes a single spec string
• Separate fields create ambiguity (override? append? format?)
• Simpler API, simpler logging (one field = one column)
• Training data pipeline is cleaner (each row = one spec string)

• API is simpler and matches the underlying model interface
• Website does the structured -> flat conversion in frontend JS
• Logging stores the merged spec string
• Power users can format examples however they want in the spec text