Start with a handful of hand-graded cases and a model-graded rubric. It is rough, but it turns vibes into a number you can move.
Evaluating prompts without a labeled dataset
Bootstrapping an eval when you have no ground truth.
cover image
Related articles
16:9 - cover
Guides
A field guide to prompt design
What actually moves output quality - and the small set of habits that beat clever one-liners every time.
16:9 - diagram
Frameworks
The PTCF framework: Persona, Task, Context, Format
A repeatable four-part structure for prompts that stay reliable as your product and models change.
16:9 - cover
Prompt Examples
12 prompt patterns that ship better outputs
Copy-ready building blocks - from role priming to output contracts - with annotated before/after examples.