Prompt Evaluation Loop

Core idea

Prompt quality improves fastest when you evaluate outputs against explicit criteria instead of relying on gut feeling.

Revision points

• Start with a baseline prompt.
• Define success conditions.
• Compare outputs using the same test set.

{
  "task": "summarize a note",
  "criteria": ["accuracy", "brevity", "coverage"]
}

What to remember

Iteration without evaluation usually becomes guesswork.