A model is only worth keeping if it can do one thing: make a prediction you can test. You feed in the situation, the model tells you what should happen, and then — this is the important bit — you go and measure what actually happens, and compare.
Calling the Result Before It Lands
Think about a weather model. Nobody cares that it can describe yesterday's rain — yesterday already happened, you were there. What makes it useful is that it says “78% chance of rain on Saturday” before Saturday, and then Saturday rolls around and you find out whether it was right. A prediction is a model sticking its neck out. The test is reality calling its bluff.
Predict, Then Check
Here's the move in the toy. You've got a bean plant. You decide how much water it gets each day, and a little growth model predicts how tall it'll be after a fortnight — more water, taller plant, up to a point. The model draws its predicted plant. Then you reveal the actual measured plant that got that much water. Sometimes the model nails it. Sometimes the real plant is a bit shorter, because real beans have their own ideas. Either way, you can now see the gap between prediction and reality — and that gap is the most useful number in the whole topic.
Notice you didn't just trust the model because it looked clever. You made it commit to a number, then held it up against the real measurement. That's the difference between a model that's decoration and a model that earns its keep: a real one risks being wrong, out loud, every time you use it.