Compare before you commit. Treating a first draft as final skips a step — comparing candidate responses, or reviewing a draft against the original criteria, catches issues before they ship (this is the same self-correction pattern from Module 2's task decomposition).
Editing is a distinct skill from generating. Tightening prose, correcting tone drift, and fixing structural issues are their own pass — not something that should happen automatically just because the first draft exists.
Choosing the right container for the output matters as much as the content itself:
| Output need | Container |
|---|---|
| Quick, disposable answer | Inline chat text |
| Data meant for further processing | Structured data (table, JSON) |
| Something the user will reuse, edit, or hand off | Artifact |
This ties directly back to Module 1's feature-selection judgment — an Artifact earns its place when the output has a life beyond this one conversation, not by default.
Organizing multi-part output means sequencing and formatting it for the reader's benefit, not just dumping content in the order it was generated.