How do we know it's any good?

You trained a model — but is it actually good? Generalisation, the train/test idea, and the one-liner that does what you just built by hand.