Evaluation

Evaluation is saying how to tell the answer worked: the test it should pass, or what would make it ready.

What it looks like

• Stating what “done” means: “ready to send to the board.”
• Asking the AI to check its work against something specific.
• Defining what “good,” “better” or “awesome” means for this task.

When it doesn't count

When success is obvious from the task, Evaluation counts only when the prompt needs it, or not at all. It always counts in tasks like tests, code review and decisions, and on a Correction it counts double: say how to tell the fix is right.