One prompt is not a strategy.

It is one roll of the dice.

LLMs make candidates. Your system decides what survives.

The real work is the loop around the model: criteria, examples, checks, memory, taste, and receipts.

Use the model for range. Use the room around it for judgment.