onrup

Evaluation

Evaluation gate

What is evaluation gate?

An evaluation gate is a mandatory check between training a model and deploying it: the candidate is scored against the model it would replace, and only a passing result permits the deploy. It converts "is this better?" from an argument into a measurement.

The value is in it being blocking rather than advisory. A dashboard that shows a regression is a dashboard somebody can decide to overlook at the end of a long week; a gate that returns fail is not.

A well-designed gate returns three outcomes rather than two. Pass and fail are obvious; inconclusive means the difference was inside the noise, which is a genuinely different situation and should not be silently rounded to either.

It should measure against the incumbent, not against an absolute bar. The decision is always comparative — this model or the one already running — and an absolute score cannot answer that.

Related terms

All terms in the glossary →

Start with the free tier

A magic link creates your account, your tenant and your first API key. No card until you ask for compute.