The Free Energy Principle is the most ambitious unifying proposal in the contemporary brain and behavioural sciences. From a single premise — that any system persisting as a distinguishable thing can be described as minimising variational free energy — its proponents recover perception, action, attention, learning, and the boundary of the organism itself.
This audit does not dispute the framework's mathematical seriousness. It examines one structural commitment: the framework declares a division between an unfalsifiable principle and a falsifiable periphery of process theories. The argument is that the division is not kept under load — that when a peripheral prediction fails, the immunity granted to the principle is extended to cover it.
The audit is directed at the structure of the claim, not at intent. No motive is imputed to any researcher at any point. The argument stands or falls on whether the declared division survives its own test cases.
An entire section is given to the framework's case at the strength its own authors currently give it: the inversion, what follows from persistence, the unification, what the framework says about its own status, what it has produced, and the revisions it has undergone — presented as counting in its favour rather than against it.
This is not a formality and the paper says so. An audit that begins by weakening its target has proved nothing when it wins.
The first fault is a category error: the collapse of what a system wants into what it expects. Under the framework, a goal is represented as a prediction the system holds about its own future states — and the machinery offered to distinguish the two is precision weighting.
The paper's argument is that precision weighting cannot separate a goal being set from a goal merely being predicted, except after the fact. Post-hoc separability is not a distinction the framework can use predictively, which is what it needs the distinction for.
The base imperative implies that an environment without variation is optimal — the dark room. The framework has an answer, and the paper grants that the naive form of the objection fails.
The audit then tests the second answer against a kind of evidence the thought experiment usually avoids: the clinical record of real environments in which homeostatic needs are met while variation is withdrawn. The record demonstrates the reverse of what the imperative implies. The paper also asks whether the exploratory drive invoked to solve the problem is genuinely derived from the framework or added to it, and traces the accommodation as it appears in print — including what that accommodation costs.
The third fault is the one the title names. The framework claims immunity for its principle at the point of attack, while advertising the empirical reach of its models elsewhere — and runs both readings within a single publication.
The audit's finding is that the advertised reach has been left untouched across two decades in which the formal machinery supporting it was rebuilt three times. A claim that survives the replacement of its own mathematics three times over is not being carried by that mathematics.
The paper argues that the phenomena the framework reaches for first — stabilisation against a varying environment, the persistence of goals under disconfirmation, the rejection of disturbances that were never modelled — are accounted for more economically by a closed negative-feedback architecture specified before the framework existed and never compared against.
It comes with something the framework has no equivalent of: an operational procedure for identifying what a system is actually controlling, rather than inferring it from the model afterwards.
This is not offered as a victory. Section 6.6 states what that architecture does not yet have, in its own section, before anyone else has to point it out.
Two features of the method are flagged in the abstract rather than buried. Where the framework has an answer, the answer is conceded before the objection is made — and in two cases the concession is total. Each of the three faults is tested against a different kind of evidence, so that no single class of objection carries the argument.
Section 8 states the conditions under which each argument here would be shown wrong. Two of those conditions have already been met. They are reported as met rather than quietly dropped.
The framework's mathematical seriousness is granted throughout. The target is the structural relationship between an immunised principle and an advertised empirical reach, not the formalism itself.
The deposited text carries an explicit list of references awaiting completion, and marks which claims rest on verified sources and which remain provisional. Nothing is presented as settled that the author cannot currently source.
Section 6.6 sets out what the proposed architecture does not yet have. The paper does not claim the alternative is finished.
Current version. Diener, Ł. (2026). The Description Dressed as a Law: A Structural Audit of the Free Energy Principle and the Case for Perceptual Control (version 2.0). Zenodo. https://doi.org/10.5281/zenodo.21761071
Previous version. Version 1.0, 11 July 2026. https://doi.org/10.5281/zenodo.21309527
Preprint, not peer reviewed by a journal. Licensed CC BY 4.0. A companion analysis in plain language is published on this site.