N1 - The Free Energy Principle as a public lens
How to read this page
Three ways to read this page. Precise is the document itself, exactly as it is written in the repository. Plain and Clear were written for this website to help you meet that document — they are about it. They are not it, and they are not evidence.
The Encyclopedia is the UNI method written out as a reference work: 39 pages, arranged in wings, setting out what the programme is attempting and why it is built the way it is. This is where the ideas are explained in order and in prose, rather than as code, as runbooks, or as dated receipts.
Every chapter is authored against two ledgers and never ahead of them. One records what UNI has built, and the evidence class of each claim. The other records nature's own regularities, kept separate on purpose. That way a fact about biology is never quietly reused as a fact about the software. Where a chapter and a ledger disagree, the chapter is the thing that is wrong. Every chapter closes with an invitation to falsify it, and a recorded negative is published beside the result it qualifies rather than after it.
Read "How to read this work" first. It is the evidence constitution: the classes, the four ledger states, and the rule that a finished chapter is not the same as a working system. Then the calibration ledger, which carries the figures every other chapter is required to use.
What it is not: a description of a person or of a mind. The programme calls itself a developmental active-inference simulation, a bounded peek into a toy world, and its own index prints how much of the developmental ladder has actually been earned — roughly two rungs out of eleven or more. It is also not a report of what is running today. For what ran, and when, go to the evidence record.
Your browser cannot switch reading levels, so the document itself is shown.
Precise — the source document
This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.
The Free Energy Principle (FEP) is the textbook-level lens this encyclopedia uses to organize the UNI program: a single, citable way of talking about agents that predict their world and act to keep prediction and observation in line. It is a lens, not a verdict. Nothing in this chapter claims that active inference is the correct theory of mind, that the FEP has been demonstrated inside UNI, or that the lens itself proves anything. UNI remains a developmental active-inference SIMULATION, a bounded peek into a toy world, never a person and never a mind. The honest program position is printed here and never softened: about 2 of 11+ developmental rungs earned. This is a meaning chapter. It asserts a framing, carries its corrections as first-class content, and cites exactly one reference for the math: Parr, Pezzulo, and Friston, Active Inference (MIT Press, 2022), plus the fenced, unrefereed Zenodo preprint as a mathematical foundation only.
The lens, stated plainly (M16)
The plain-language spine is taught laypeople-first, as a ladder of four examples that climb the abstraction: a thermostat, a bird catching a moth, a person crossing a street, and your own body. The unifying idea is that an organism keeps its predictions and its observations in line by changing the prediction or by changing the world. Inferring the hidden causes is the perception half; acting to make predictions come true is the active half; together they are "active inference." Extended outward, the lens offers a moral grammar the program uses publicly: distorted maps lead to extraction and harm, while accurate maps with calibrated signals and cooperation tend toward regeneration.
This is recorded as method claim M16 (class: method): "Free Energy Principle as a textbook-level public lens." A method claim is a definitional, governance, or framing pattern. It is proven and reusable as a discipline, but it is not an empirical capability claim and cannot be raised into one. The lens does not measure anything; it tells the reader how to read what was measured.
The Popper note travels with M16 and must never be stripped. "No one can falsify it" is NOT evidence FOR a claim. An unfalsifiable statement sits outside science by definition, so the very breadth that makes the FEP feel all-explaining is a liability, not a credential. UNI's claimed strength is the opposite of unfalsifiability: a verified core plus a fenced frontier. Every load-bearing capability row in the ledger ships with a stated condition that would prove it false, and the open rungs are parked as questions, not paraded as answers. The lens earns its place only because the work behind it is falsifiable; the lens itself proves nothing.
The three-layer framing the lens licenses (M12)
The lens is given structure by method claim M12 (class: method): the WORLD perpendicular BODY perpendicular MIND three-layer framing, built from two typed Markov blankets. Interoception enters as hardware signals; the mind runs a discrete POMDP loop of perceive, plan by expected free energy, act, and learn, with exact conjugate-Dirichlet learning and no backpropagation. Surprisal and variational free energy are carried at textbook level only, anchored by the inequality F[q] >= -ln p(o|m), with precision acting as the dial that bifurcates behavior into distinct regimes.
M12 carries its own load-bearing fences inline, and they are part of the claim, not footnotes to it. The composed appliance is variationally-controlled (an audited description), and it is module-exact only at the single-step categorical body-to-mind interface, NOT globally exact. Carding "variationally-controlled" as "globally exact" is the recorded violation. A second violation the framing guards against is an agent reading raw hidden host state, which is caught by an isolation assertion and an abstract-syntax-tree whitelist. The three layers are a way to keep the bookkeeping honest, not a proof that the layers correspond to anything real in a brain.
The single most-misread point: minimizing VFE does NOT erase surprise (M14)
If a reader takes one correction from this chapter, it is this one, recorded as method claim M14 (class: method, with an E-tier correction held for human review). Minimizing variational free energy does not reduce already-observed surprise. For a fixed observed outcome the surprisal -ln p(o|m) is what it is; minimizing free energy tightens an upper bound on it by improving the posterior approximation q(s) toward the true posterior p(s|o,m). The anchor inequality is F[q] >= -ln p(o|m), with equality exactly when q(s) = p(s|o,m). Perception tightens the bound. Action lives one layer up, as expected free energy, and is deferred to a later layer of the framing; it is not what is happening when the bound is tightened.
The paired negative that travels with M14, and must be cited alongside it, is its provenance: this correction came from a human science reviewer who would not sign off on the first draft. The first draft got it wrong, in the ordinary and tempting way, by implying that doing inference makes the surprise itself go down. That is a measurement that the original framing was incomplete, and under this program a recorded negative of that kind is content, not an embarrassment to hide. The correction is held at the E tier (test-covered, negative-corrected) and would only be reopened if the inequality F[q] >= -ln p(o|m) were itself revised. Stating "minimizing VFE reduces already-observed surprise" anywhere in the encyclopedia is a falsification of this chapter's conceptual integrity, by definition.
Why hierarchy needs the cavity principle (M22)
The last piece the lens contributes is M22 (class: method), the cavity principle: a level in a hierarchy must never treat an upstream prior as if it were fresh evidence. The prior is divided out (the program records this as WS-B operating on the upward pass and WS-C on the downward pass) so that the same information is not counted twice. M22 also records that the factored mean-field approximation was rejected as lossy, so the exact joint posterior is used everywhere instead. The recorded violations that would falsify M22 are double-counting a prior as evidence, which inflates belief, or showing the mean-field variant to be non-lossy after all. The cavity principle is why the framing can stack layers at all without quietly manufacturing confidence out of nothing; it is a bookkeeping discipline, not a capability.
What is NOT claimed in N1 - The Free Energy Principle as a public lens
- Ceiling. That UNI demonstrates active inference, or that the FEP is proven to be the correct theory of mind (or that the breadth of the FEP is itself evidence), is NOT shown. The most we claim is that the Free Energy Principle is a textbook-level public lens for organizing the program (M16, method), that the WORLD/BODY/MIND three-layer framing (M12, method) is a coherent and audited way to describe the appliance at the single-step categorical interface only, and that two conceptual corrections are held: minimizing VFE tightens an upper bound rather than reducing observed surprise (M14, method/E), and a hierarchy must divide out upstream priors via the cavity principle (M22, method). All four are method claims and assert no capability.
- Fences engaged. Red line 3 (never "active inference demonstrated": the lens is the framing only, no AIF loop is claimed here). Red line 7 (never raise a claim above its source class: M16, M12, M14, M22 are all
method, with M14's correction held at E). Red line 10 (no patent-level UNI math: textbook-level framing only; no PII). Red line 11 (the Zenodo preprint, DOI 10.5281/zenodo.19785799, MIT, is the mathematical foundation only, always fenced unrefereed: Layer-1 AI-executable audit complete, Layer-2 human expert review PENDING; never proof the theory is correct). Red line 12 / vocabulary guard (this is a MEANING chapter, so "active inference" and "free energy principle" are permitted at textbook level only). - Negatives that travel with this claim (cite alongside, never strip). The Popper note on M16 ("no one can falsify it" is NOT evidence FOR a claim; unfalsifiable means outside science; UNI's strength is verified-core-plus-fenced-frontier, not unfalsifiability). The M14 provenance negative (the VFE-bound wording was wrong in the first draft and was corrected only because a human reviewer refused to sign). The M12 inline bound ("variationally-controlled" and "module-exact at the single-step interface only, NOT globally exact"). The M22 lossy-mean-field rejection (the factored approximation was discarded; double-counting a prior is the standing violation).
- Parked / owed. No empirical PASS is owed here, because this chapter makes no empirical claim. What is owed sits one level out, in the science it frames: the preprint's Layer-2 human expert review remains PENDING (the top falsifiable edge of the whole science posture), and M14's correction is held for human review at the E tier. The lens does not discharge either; it points at them.
- One-line honest summary a skeptic could not dispute. This chapter offers a citable, falsifiability-aware vocabulary for reading the program, plus four recorded conceptual corrections, and it claims no capability, no demonstration, and no proof of any theory.
Falsify this
The lead, operable falsifier for this chapter is conceptual integrity. N1 is falsified the moment any encyclopedia copy presents "minimizing variational free energy reduces already-observed surprise" as true: the recorded fact is that it tightens an upper bound (F[q] >= -ln p(o|m), equality iff q(s) = p(s|o,m)), while action lives one layer up as expected free energy (M14). It is equally falsified by externalizing patent-level UNI math under cover of the lens, by citing "no one can falsify it" as evidence FOR any claim (M16), by carding the variationally-controlled appliance as "globally exact" (M12), or by double-counting an upstream prior as fresh evidence in a hierarchy (M22). Any of these overturns the chapter as written.
Sources
- Ledger rows: M16 (FEP as a textbook-level public lens; the Popper note), M12 (WORLD perpendicular BODY perpendicular MIND framing; variationally-controlled; module-exact at the single-step interface only), M14 (the VFE-bound correction, held for human review), M22 (the cavity principle; mean-field rejected as lossy) -
encyclopedia/CLAIM-LEDGER.md, Section 3 (Method claims). - Authoring spec and constitution:
encyclopedia/MASTER-PLAN.md, Section N1 and Part I front matter (FM-1 Evidence Constitution, FM-2 A-U rubric, FM-3 red lines, FM-4 not-claimed template). - Narrative grounding (PII-redacted digests):
curated/worldmodels-digest.md(the thermostat to body teaching ladder; the VFE-bound correction and its human-reviewer provenance; the verified discrete-time engine and three precision knobs),curated/website-digest.md(the public-facing textbook-level UNI framing; the unrefereed-preprint gating). - Reference (the single citable foundation): Parr, Pezzulo, Friston, Active Inference (MIT Press, 2022); plus the fenced, unrefereed preprint - Polzin et al. (2026), Zenodo DOI 10.5281/zenodo.19785799, MIT (Layer-1 audit complete; Layer-2 human expert review PENDING).
- Archive pointers (no PII):
...-WorldModels,...-SolutionWrightUniversal-Website.
sha256 28cb3fb03f484424 — of the original file, so what was ingested stays checkable.
Plain — written for this website, not the source document
The free energy principle, taught at textbook level, is the lens the encyclopedia uses to organise everything else. A lens is not a verdict. Nothing here claims that active inference is the correct theory of mind, or that it has been shown inside the program. The program itself is a simulation, a toy world rather than a person. The plainest idea is that an organism keeps its predictions and its observations in line either by changing the prediction or by changing the world, and the chapter teaches this with a ladder of everyday examples. It also carries a correction it is careful not to lose: minimising variational free energy does not reduce surprise that has already happened. It tightens an upper bound on it by improving an approximate posterior. That correction came from a human reviewer who would not sign off on the first draft, and the chapter treats where it came from as content.
Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 28cb3fb03f484424
Clear — written for this website, not the source document
This is a meaning chapter. It asserts a framing and no capability. The program remains a developmental simulation, a bounded peek into a toy world.
The lens is taught from four everyday examples that climb the abstraction, and the unifying idea is that inferring the hidden causes is the perception half while acting to make predictions come true is the active half. Recorded as a governance pattern, the lens can be reused but cannot be raised into an empirical claim.
A note travels with it and must never be stripped. That nobody can falsify a statement is not evidence for it. An unfalsifiable statement sits outside science by definition, so the very breadth that makes this principle feel all-explaining is a liability rather than a credential. The claimed strength is the opposite: a checked core with a frontier marked off rather than claimed.
The lens is given structure by a three-layer framing built from two typed boundaries, in which hardware signals enter as interoception and the mind runs a discrete loop of perceiving, planning, acting and learning, with no gradient step. The framing carries its own limits inline. The composed appliance is described as variationally controlled and exact only at a single-step interface, not globally exact, and recording it as globally exact is the violation on file.
The most-misread point gets its own section. For a fixed observed outcome the surprisal is what it is; minimising free energy tightens an upper bound on it by moving the approximate posterior toward the true one, with equality exactly when the two coincide. Perception tightens the bound; action lives one layer up. Stating that minimisation reduces already-observed surprise falsifies the chapter by definition.
A final piece covers a bookkeeping rule for hierarchies: a level must never treat an upstream prior as if it were fresh evidence, so the prior is divided out. A cheaper factored approximation was implemented, measured as lossy, and rejected, and that rejection is recorded rather than glossed.
The chapter closes with what it does not claim, and with the observation that what is owed sits one level out, in the science it frames.
Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 28cb3fb03f484424