Key answer
Use a preliminary artifact when a wrong direction would be replicated widely or you must see something to judge preferences. Match it to uncertainty: outlines test coverage/structure, samples tone/presentation, and trials whether materials, tools and processes actually work.
Full explanation
An outline reveals missing chapters in a long tutorial, not whether explanations suit beginners; a representative section tests difficulty. A polished sample table does not validate batch extraction: use a real file through reading, extraction and checking. Showing a suitable direction and demonstrating a feasible process have different acceptance targets.
AI Chains participants used intermediate results to locate problems and adjust local steps, supporting opportunities to inspect key artifacts. Choose samples exposing the actual difficulty. Study
Do not choose only easy content. A knowledge-card series could sample a concept, a numerical/cited question and a confusing boundary. Check accuracy, adequate explanation and fit within the layout before applying the approach elsewhere.
Complete three representative cards covering concepts, data and confusing cases; explain the selection. Check them against the supplied criteria yourself. Present concrete alternatives only for style decisions I must make. Once rules are clear, finish the rest without approval per card.
With a mature template, complete sources and little content, a sample round can cost more than direct completion. Check consistency and omissions across the final batch: approved samples do not verify everything else.