S-L4 - Affect-as-precision and the Z modulator
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.
Wing S - Science (UNI), the L0-L12 developmental ladder. This is rung four of the HUMAN-HGM-001 developmental design: a no-backprop, nested-Markov-blanket developmental SIMULATION, never a person, never a mind. One engine, shown across scales. Honest program position, printed and never softened: ~2 of 11+ developmental rungs earned.
The honest position, stated first
This chapter describes how the UNI simulation models affect: not as a feeling, but as a setting on a dial. In the active-inference framing used throughout this work (a textbook-level lens, after Parr, Pezzulo and Friston, Active Inference, MIT Press 2022, and never a demonstrated loop in itself), "affect-as-precision" means that an emotional state changes how confidently the agent weights its sensory evidence against its predictions. Turn precision up and perception sharpens; the agent leans on what it is seeing. Turn it down and the agent leans on its priors and on exploration. The balance between pragmatic value (act to reach a preferred state) and epistemic value (act to reduce uncertainty) flips as that dial moves.
The load-bearing fence sits at the top, not the bottom: affect here is modeled, never felt. Nothing in this chapter, and nothing in the ledger row behind it, claims that the simulation experiences energy, arousal, pain, or threat. Phenomenal feeling and sentience are explicitly disclaimed. No falsifier is offered for sentience because it is disclaimed, not tested. What is claimed is narrow, functional, and held to a single registered class.
What the ledger records (L4.1)
The single carded claim is L4.1, recorded at Class C (a dev-gate / held-out evaluation), status proven (functional):
Affect-as-precision: emotion modulates precision so perception sharpens and the pragmatic-to-epistemic balance flips. The global Z modulator
[energy, arousal, valence, fatigue, pain, threat, safety, inflammation]sets precision, preferences, habits, learning-rate, and horizon.
The mechanism is a single eight-dimensional vector, Z, that the developmental design carries globally and that feeds the same discrete categorical engine used at every other rung. Z does not act as a separate sub-network; it modulates the existing precision parameters, the preference (goal) terms, the habit priors, the learning-rate on the conjugate-Dirichlet count update, and the planning horizon. This is why it earns a place on the ladder rather than being a one-off lab: it is the lever by which a developmental "interoceptive / autonomic" state reaches into perception and policy on one shared loop.
Class C is the empirical-PASS tier in this work's rubric, but Class C is a functional verdict, not a phenomenal one. It says the modeled dial does what the model says it does under the registered evaluation; it says nothing about inner experience. The verdict is read as the calibrated, in-class statement and never raised above it.
The falsifier (stated operably)
L4.1 carries an explicit falsifier, and it is the right kind: an ablation that removes the dial and looks for the predicted effect to vanish.
An ablation of the Z modulator shows no change in precision-weighting and no pragmatic-to-epistemic flip under the grounded reader, OR the effect collapses under control.
In plain terms: cut Z out (or shuffle it into noise) and re-run the grounded reader. If precision-weighting does not change, and if the pragmatic-to-epistemic balance does not flip, the claim is false. Equally, if the apparent effect collapses the moment a matched control is introduced (a load-bearing discriminator, the program's standing discipline), the claim is false. The claim earns its Class-C standing precisely because it is pinned to a discriminator that can collapse it.
The negative that travels with this claim (M10 sub-bound)
Citing the L4.1 PASS without its paired negative would be an overclaim, and review treats it as one. The negative that must always travel alongside is the M10 honest boundary, recorded as a sub-bound of this rung:
Neuroticism does NOT change behavior under the grounded reader's bimodal surprise without a graded task. A graded task would be needed to expose any such effect.
This is exactly where the fence between modeled and felt becomes empirical rather than rhetorical. A personality-style affective parameter (neuroticism, the disposition to weight threat and negative valence more heavily) was probed against bimodal surprise in the grounded reader, and it did not move behavior on its own. The honest reading is bounded and specific: the Z-style modulation is demonstrably functional for precision-weighting and the pragmatic-to-epistemic flip (L4.1), but it is not a general behavioral driver across every affective axis. Some affective dimensions, under some tasks, produce no measured behavioral change. M10 is a recorded measurement that the design is incomplete on that axis, not a failure to hide. The negatives are the credibility, and this one delimits L4.1 rather than decorating it.
This pairing also guards the larger temptation. Affect language is the single most inviting place to slide from a dial into a feeling. The M10 sub-bound holds the line by showing, with a measurement, that the modeled affective parameters are levers on inference, not evidence of an inner life: a parameter that fails to move behavior under bimodal surprise is plainly a parameter, not a felt state.
What this rung is, and what it is not
L4 sits between L3 (organ-level physiological control, the toy cardio-renal Heart Lab) and L5 (the sensorimotor / motor hierarchy). It is the "interoceptive / autonomic plus affect" layer: the place where a developmental simulation's internal state (energy, fatigue, pain, threat, and so on) is allowed to shape how it perceives and what it pursues. That is a genuine, demonstrable encyclopedia entry on how a modeled feeling shapes modeled attention. It is not, and is never presented as, a claim that the simulation feels anything.
What is NOT claimed in S-L4 (Affect-as-precision and the Z modulator)
- Ceiling. That the UNI simulation feels emotion, or has phenomenal affect, sentience, or any inner experience, is NOT shown and is explicitly disclaimed. That affect is a general behavioral driver across all axes is NOT shown either. The most we claim is the exact, in-class statement: at Class C (proven, functional), an ablatable global Z modulator
[energy, arousal, valence, fatigue, pain, threat, safety, inflammation]modulates precision so that perception sharpens and the pragmatic-to-epistemic balance flips under the grounded reader, with the effect required to collapse under control. - Fences engaged. Red line 2 (never consciousness / sentience / aware; phenomenal sentience explicitly DISCLAIMED, no falsifier offered because it is disclaimed not tested). Red line 3 ("active inference" is the framing lens only, never a demonstrated loop; no AIF loop in the Rust crate, the live UNI.OS loop is a separate, not-gate-matched reimplementation). Red line 7 (never raise a claim above its source class; L4.1 stays Class C). Red line 10 (textbook-level framing only, no patent-level Z math, no PII). The standing program framing: developmental SIMULATION, toy world, ~2 of 11+ rungs earned.
- Negatives that travel with this claim (cite alongside, never strip). The M10 neuroticism sub-bound: neuroticism does NOT change behavior under bimodal surprise without a graded task. This negative must appear next to the L4.1 PASS in every citation; the PASS alone is an overclaim.
- Parked / owed. No sign-to-park is owed for L4 (it is a proven-functional rung, not a parked frontier). What is owed is permanent co-citation: L4.1 may never be quoted without M10 traveling with it, and the phenomenal-sentience disclaimer is a standing fence on this rung, not a discharge.
- One-line honest summary a skeptic could not dispute. A modeled, ablatable affect dial changes how the simulation weights evidence and what it pursues (Class C), one affective parameter provably did not move behavior without a graded task (M10), and nothing here is felt.
Falsify this
Lead falsifier, stated operably: ablate the Z modulator (remove it, or shuffle its eight components into uninformative noise) and re-run the grounded reader under the registered protocol. If precision-weighting does not change and the pragmatic-to-epistemic balance does not flip, L4.1 is false. If the effect survives ablation but collapses under a matched control, L4.1 is false. Independently, the paired M10 sub-bound is overturned only by a graded task in which a neuroticism-style affective parameter is shown to move behavior under bimodal surprise with a CI excluding zero; absent that, neuroticism remains a non-driver and the modeled-not-felt fence stands.
Sources
- Curated digest (PII-redacted):
curated/uni-gpt-digest.md(the HUMAN-HGM-001 ladder, the global Z modulator[energy, arousal, valence, fatigue, pain, threat, safety, inflammation], affect-as-precision as an encyclopedia entry, and the recorded M10 honest boundary: neuroticism does not change behavior under bimodal surprise without a graded task). - Curated digest (PII-redacted):
curated/uni-mind-digest.md(the one no-backprop discrete-categorical engine reused verbatim, the model/process Markov-blanket split, the textbook-level epistemic-vs-pragmatic decomposition, and the developmental-ladder fencing). - Ledger of record:
encyclopedia/CLAIM-LEDGER.md, row L4.1 (Class C, proven-functional) and its "what is NOT claimed at L4" line carrying the M10 sub-bound. - Authoring spec and constitution:
encyclopedia/MASTER-PLAN.md, section S-L4 plus PART I front matter (FM-1 Evidence Constitution, FM-2 A-U rubric, FM-3 red lines, FM-4 not-claimed template). - Source archives (local-only, PII, never read for this chapter):
...-UNI-GPT,...-uni-mind.
sha256 6094a5b604673391 — of the original file, so what was ingested stays checkable.
Plain — written for this website, not the source document
Affect appears in this simulation as a dial rather than a feeling. Where the dial sits changes how heavily the agent trusts what it is currently sensing against what it already expects. Push it one way and sensing dominates. Push it the other and priors and exploring take over, and the trade between chasing a preferred outcome and chasing information tips with it. Affect here is modelled and never felt. Nothing claims the simulation experiences energy, arousal, pain or threat, and phenomenal sentience is disclaimed outright rather than tested, which is why nothing is offered that could overturn it. What remains is a narrow functional claim held to one registered class, and it travels with a measured boundary marking where the dial does not reach.
Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 6094a5b604673391
Clear — written for this website, not the source document
All of this happens inside a simulation — a toy world, and not a person. The recorded claim is that emotion modulates precision so that perception sharpens and the balance between goal-seeking and uncertainty-reduction flips, driven by a single global modulator carrying a short list of interoceptive dimensions.
The mechanism is one vector that the developmental design carries globally and that feeds the same discrete engine used at every other rung of the ladder. It does not act as a separate sub-network; it modulates the existing precision parameters, the preference terms, the habit priors, the learning rate on the count update, and the planning horizon. That is why it earns a place on the ladder rather than being a one-off lab: it is the lever by which an internal bodily state reaches into perception and policy on one shared loop.
The class it is recorded at is the empirical pass tier in this corpus, but the verdict is functional rather than phenomenal. It says the modelled dial does what the model says it does under the registered evaluation, and says nothing about inner experience.
What would show it wrong is the right kind of test, and it is an ablation. Cut the modulator out, or shuffle it into noise, and re-run the reader. If precision-weighting does not change and the balance does not flip, the claim is false. Equally, if the apparent effect collapses the moment a matched control is introduced, the claim is false.
The negative that must travel with the pass is a measured boundary on a different axis. A personality-style affective parameter was probed against a particular kind of surprise and did not move behaviour on its own, and a graded task would be needed to expose any such effect. So the honest reading is bounded: the modulation is functional for precision-weighting and for the flip, and it is not a general behavioural driver across every affective axis.
The chapter treats that pairing as the guard on the largest temptation here. Affect language is the easiest place to slide from a dial into a feeling, and a parameter that fails to move behaviour under a registered probe is plainly a parameter rather than a felt state.
Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 6094a5b604673391