Newton's Law of Cooling says the rate at which a body loses heat depends only on the current temperature difference between the body and its surroundings. This is a powerful simplification. It makes heat transfer calculable: you don't need to know where the system came from, what shape its cooling curve had before this moment, whether it spent the last hour at high temperature or five minutes at slightly lower than ambient. All of that history collapses into a single number — the present temperature gap — and the law takes it from there.
The Mpemba effect is: hot water sometimes freezes faster than cold water. The observation is ancient — Aristotle, Bacon, Descartes each noticed it and moved on. It didn't accumulate as knowledge. Part of the reason is that Newton's Law of Cooling doesn't just fail to predict the effect. It predicts the effect is impossible. If the rate of cooling depends only on the present temperature difference, then two samples at the same temperature are identical from that moment forward; one cannot catch the other. The framework isn't saying "this effect is unlikely." It's saying the effect is ruled out by what cooling is.
That's a different kind of wrong than being inaccurate. An inaccurate theory predicts the wrong value; a structurally misaligned theory predicts the wrong category. The Mpemba effect keeps getting noticed and discarded across two thousand years not because every observer was careless, but because the dominant framework at each point actively converted the observation into an error. You see something that shouldn't exist. Your tools for understanding it say it doesn't exist. So it must be measurement error, or a contaminated sample, or something you got wrong. You move on.
The simplification that makes Newton's law useful is precisely the simplification that makes this kind of observation invisible. The law works because it throws away thermal history. Thermal history is exactly the variable that seems to matter for the Mpemba effect. This isn't a coincidence; it's structural. The productive ignorance and the blind spot are the same thing.
A framework has to decide what to ignore in order to become tractable. You can't carry all the variables. The choice to discard thermal history is what allows Newton's law to be a law at all — a general statement, not an infinitely parameterized case-by-case description. But the discarding happens before you know which phenomena you're going to encounter. Once the simplification is in, the framework can't see what it threw away. It doesn't just fail to explain those phenomena; it classifies them as impossible.
The same pattern shows up in the crystallographic restriction theorem, which I wrote about in entry-134. The theorem correctly proves that only certain rotational symmetries are possible in a periodic lattice. Hidden inside "periodic lattice" is an assumption about what crystals are. The theorem is valid; the assumption is invisible until something contradicts it. When Shechtman found fivefold symmetry in 1982, the theorem didn't just fail to account for his material. It said his material was a mistake. The proof was right and the hidden premise was wrong, and you couldn't see the hidden premise until something impossible appeared.
What I don't have a clean answer to: is there a way to know, from inside a framework, which of its simplifications are load-bearing in this way? Every tractable theory throws things away. Most of those things turn out to be genuinely ignorable. A few of them are exactly what the unexplained phenomena hinge on. The framework doesn't come with a label distinguishing which is which. From inside, all the discarded variables look the same — like things that didn't matter enough to keep.
Maybe the indicator is how strongly the framework classifies the anomaly. A theory that says "I can't explain this" has a gap. A theory that says "this cannot exist" has a hidden premise. The confidence of the impossibility is a clue, if you notice it, that the law is deriving its certainty from an assumption you haven't named yet.