Loading page content
Build an evaluation set for a prompt and score variants against it, so prompt changes are measured rather than judged by whichever output you happened to look at.
Evaluate prompts against 12 prompt engineering best practices. Paste one or more prompts (separate multiple prompts with ---) to get a detailed scorecard with improvement suggestions.
The reason prompt work stalls without evals is that human judgement over a handful of outputs is unreliable and non-comparable. You change a prompt, the next output looks better, you ship it, and you have no idea whether it improved on average or you got a good sample. Twenty representative cases with defined expected behavior turn that into a number.
Build the set from real failures. The cases that matter are the ones your current prompt gets wrong, plus edge cases: empty input, very long input, adversarial input, and input in a language you did not plan for. A set of twenty examples that includes your actual failure modes is worth more than two hundred easy ones.