The one line to carry out of this rung is older than you'd think: the map is not the territory. A model is a stand-in, not the real thing — and because it dropped stuff back in rung 1, it is guaranteed to be wrong somewhere. That's fine. A wrong-somewhere model is still brilliant. The danger is forgetting it's a model and over-trusting it past the point where it breaks.
Trap One: Mistaking the Model for Reality
This is the big one. The globe is not the Earth; the particle diagram is not matter; the equation is not the ball. People slide into talking as if the model is the thing, and then they trust details the model never actually got right. When you catch yourself defending a feature of the model, stop and ask: is that true of the real thing, or just an artefact of how we drew it?
Trap Two: the Map That Lies About Size
A flat world map has to peel a round Earth onto a rectangle, and something has to give. On the common one, the bits near the poles get stretched horribly — Greenland looks about the size of Africa, when really Africa is fourteen times bigger. The map is genuinely useful for "which way is north" and useless for "compare these countries' sizes". Same model, brilliant for one job, a liar for another. That's the lesson: a model's wrongness depends on what you ask of it.
Trap Three: the Atom That Isn't a Tiny Solar System
The shell model of the atom — a nucleus with electrons whizzing round it in neat rings, like planets round the sun — is a fantastic teaching picture. It is also not literally true. Electrons don't sit in tidy circular orbits; they live in fuzzy clouds of "probably-here", and the real maths is far stranger. The shell model keeps "how many electrons and roughly how grouped" and quietly drops the truth about where they are. Lean on it to count electrons: great. Lean on it to picture an electron as a little planet on a track: you've walked off the edge of what it can do.
And the Quiet One: a Model Is Only as Good as What Went In
A computer model that predicts the weather, or a population, runs on the numbers and rules we feed it. Feed it dodgy data or leave out something that matters, and it will hand back a confident, tidy, wrong answer. The model isn't lying — it's faithfully working out the consequences of what we told it. "Garbage in, garbage out" is the whole warning: a beautiful simulation built on bad assumptions is still bad.