Compare prompt variants with real outcomes
Run prompt experiments, review results, and use evidence-backed suggestions to guide the next version.
Prompts now have a way to be measured as well as edited. Teams can create variant arms for an instruction, run an experiment, and compare outcomes before deciding which version should become the next release.
A structured experiment
Set up a control and one or more challengers, choose how traffic is assigned, and review the results in the prompt workspace. The experiment view keeps versions and their performance together, so a decision has context beyond a single example conversation.
Improve with evidence
Rylvo can propose prompt improvements informed by observed runs. Teams can inspect a suggestion, test it, and decide when to promote a version. Version history and release controls make the change traceable after it goes live.
This gives prompt owners a repeatable loop for iteration: formulate a change, compare it against the current behavior, and publish with a clear record of why the new version was chosen.