Scientists almost never get a model right on the first go. What they do instead is run a cycle: build a rough model, use it to predict, test the prediction against data, refine the model so it fits better, then predict again. Round and round, getting closer each lap.
The Modelling Cycle, in Five Moves
One — build a model. Start with a simple guess at the pattern. A straight line through the data is the classic one.
Two — predict. Use your model to call a result: for this input, the line says that output.
Three — test against data. Compare the prediction to the real measurements. How far off is it?
Four — refine. Adjust the model to shrink the gap — shift the line, change its slope — until it hugs the data better.
Five — predict again. Now use the improved model on a fresh case, and check it once more.
The Key Idea: Fit Means Trust
Here's the rule that makes this worth doing. A model that matches the data you already have is trustworthy for situations like it. If your line passes neatly through ten measured points, you can reasonably believe it for an eleventh point sitting right among them. The better the fit, the more confident your prediction — as long as you stay in the territory you measured. Stray too far outside it and things get sneaky — that's the whole of the next rung.
A Worked One, Slowly
Say you've measured how far a toy car rolls for five different ramp heights, and you want to predict the roll for a ramp you haven't tried. You build a straight line through your five points — that's your model. You predict: read off where the line sits at the new ramp height. You test by checking how snugly the line already sits among the five real points; if one point is miles off the line, your fit is poor and the prediction is shaky. So you refine — tilt and shift the line until the total miss is as small as you can make it. Now you predict again for the new ramp, and this time you can trust it, because the model has earned that trust on the data you do have. Build, predict, test, refine, predict — same loop every time.