Baselines before fancy models

A baseline is the simplest honest comparison. For a classifier, it might be "always choose the most common label" or "use the old rules the team already trusts." If a trained model cannot beat that, the model is not earning its complexity.

Baselines are beginner-safe because they turn vague model excitement into a visible bar. The question becomes: does the new workflow improve a real decision, on held-back examples, by enough to matter?

In workplace terms, a baseline stops a lead scorer, ticket router, or research-note tagger from shipping just because it has a model inside.