Entropy is not disorder, and never was

Ask what entropy measures and you will be told: disorder. The metaphor has survived a century of counterexamples because it is vivid, not because it is right. Entropy counts something precise — the number of microscopic arrangements compatible with what you know about a system. It is a statement about information, and it always was.

The formula @Boltzmann is buried under says it plainly: S = k log W. Here W is the number of #microstates — exact molecular configurations — consistent with the macrostate, meaning the handful of numbers you actually track: pressure, volume, temperature. Nothing in the formula mentions messiness, and no procedure for computing W ever asks how a state looks. W counts the arrangements your coarse description fails to distinguish, and entropy is the logarithm of your ignorance.

Where the disorder picture meets real systems, it loses. Compress hard spheres and at high enough density they spontaneously crystallize — the neat lattice has more entropy than the jammed disordered fluid, because spheres arranged on a lattice each gain local room to rattle, and the rattling multiplies microstates faster than the ordering spends them.¹ Entropy built the crystal. The same accounting drives oil out of water, stretches a rubber band back when you let go, and assembles colloidal particles into ordered phases no energy term asked for. A concept that cannot survive "order can be the high-entropy state" was never the concept.

The paradox that gives the game away

@Gibbs saw the deepest crack in the 1870s. Remove a partition between two different gases and the entropy of mixing appears, a definite computable amount. Remove the same partition between two samples of the same gas and nothing happens — though the molecules wander across just as freely. The molecules have not consulted your chemistry. What changed is whether your description can tell the two sides apart; the #gibbs-paradox dissolves the moment you admit that W is counted relative to a set of distinctions.²

@Jaynes made the lesson systematic in 1957. Take Shannon's measure of missing information, demand the probability distribution that assumes nothing beyond what is actually measured, and the entire apparatus of equilibrium statistical mechanics — canonical ensembles, partition functions, the lot — falls out as inference rather than as extra physics. Nothing needs to be postulated about ergodicity or molecular chaos; you are simply reasoning as honestly as possible from partial data. On the #maxent reading, thermodynamic entropy is the amount of microscopic detail your macroscopic variables leave unspecified, expressed in Boltzmann's units.³

Does this make entropy subjective? No — and the distinction is worth being careful about, because it is where the argument usually goes wrong. Fix the set of variables you track, and W is as objective as arithmetic; two observers holding the same description compute the same number, and a steam table does not care who reads it. Observers with different descriptions compute different entropies, and each makes correct predictions for the measurements their own variables support. The second law holds for all of them, because macrostates with larger W are overwhelmingly larger, and dynamics wanders into them and stays.

So retire the metaphor. A teenager's bedroom is not high-entropy; a shuffled deck is not disordered to the cards, which sit in exactly one arrangement either way. Entropy is the size of the crowd of microstates hiding behind your summary of a system — a count of what the summary leaves out. The strangest fact in the whole story is chronological: Boltzmann carved an information formula onto physics in 1877, and it took seventy years, and the arrival of information theory, for anyone to be able to read what it said.