Original prompt Engineering
Evaluate prompt variants
The prompt
Create an evaluation plan comparing [variants] on [criteria], with normal, ambiguous and insufficient-evidence cases. Define what counts as failure and how disagreements will be reviewed.
Copyable text only. No model runs here, and no source material is uploaded.
When to use it
Design a prompt evaluation set before changing the instruction again.
Prepare the source material
- Prompt variants
- Quality criteria
- Representative tasks
Review before using the result
Use the same test cases and conditions for each variant, and report uncertainty rather than selecting a winner from one output.