Opening the book…
A model keeps the features that matter for a chosen question and discards the rest; that deliberate omission is the source of both its power and its limits. Frictionless planes and point masses are not lies but scoped approximations, accurate wherever the neglected effects stay small. Trouble comes from forgetting the scope: the map is not the territory, and a model applied outside the regime it was built for returns confident, wrong answers.
For every model, write down what it assumes and where those assumptions break: the ideal gas law fails at high density, Hooke's law beyond the elastic limit, projectile motion once drag matters. Check that your problem sits inside that domain before trusting the result. When a prediction diverges from experiment, suspect a violated assumption before you suspect the data, and reach for a richer model only when the discrepancy demands it.
This rule is itself unconditional: no model is the world, including our deepest theories, which remain effective descriptions valid within a tested range. Even so, a coarse model can outperform a detailed one when its assumptions fit the problem better.