This instrument maps a dynamic state frame (or frame of intention) — the purposive sequence of reaching for, grasping, raising, and ingesting a cup of coffee — together with its pre-state frame: the prior state in which deep expectations are already active before intention forms. In computing, a 'cold start' describes a system booting with no prior state loaded; biological systems never cold-start — there is always prior activation, always a history of predictions already running. The instrument models the dynamics of an organic neural system using the framework of active inference, which applies concepts from mathematical physics to biological prediction and action. Three layers are distinguished throughout: the mathematical model — the repertoire space and upper/lower bounds within which the system can vary; the neuronal dynamics — the actual trajectory of precision weighting across the 11 systems; and the lived behaviour — the intersection, where the model meets the dynamics and produces minded behaviour. Each system's precision weighting at each moment represents its coupling coefficient with the intentional act — the strength and direction of its functional relationship to the task. A coupling coefficient, in the statistical sense, measures how much a change in one variable produces a change in another. A high coupling coefficient means the system is tightly bound to the act at that moment; a low one means it is running in the background. The dual-register ovals show gross anatomical wiring (grey inner core) and learned adaptation (coloured outer oval); the gap between them is the space within which conscious access operates. Panel 4 (moving thresholds) makes that access threshold explicit: it is not fixed but is set by the frame of intention through precision weighting, and is therefore multivariate — a bundle of per-channel thresholds — and continuously in motion. Panels 4 and 5 are explicitly theoretical.
This panel maps the dynamic state frame (or frame of intention) across seven body systems, showing how their precision weighting — their coupling coefficient with the intentional act — varies across the four phases of the coffee sequence. The instrument approaches the sequence from the perspective of 11 integrated bodily systems in total (the remaining four — persistent and canalised — appear in Panel 2). Each curve represents how tightly a given system is coupled to the task at each moment: a high value means the system is strongly bound to the act; a low value means it is running in the background, present but not leading.
The underlying substrate is a biochemical neural network — billions of neurones connected by synapses, transmitting electrochemical signals, continuously active. This instrument models the dynamics of that organic system using the framework of active inference — a theory of brain function developed by Karl Friston and colleagues, which applies concepts from mathematical physics to biological prediction and action. In this framework, the brain is understood as a prediction machine: rather than passively receiving sensory information, it continuously generates predictions about the world and updates them when prediction errors arise. The mathematical model describes what the neural system is doing in principle; the neural dynamics are what is happening in tissue; the lived behaviour is the intersection where the two produce minded action.
The mathematics of active inference is organised around a quantity called variational free energy. This is a mathematical construct — not electrochemical or metabolic energy, which cannot be directly measured in this context. The term is borrowed from statistical physics by analogy, but it lives entirely within the mathematical model. It measures how well the brain's current generative model accounts for incoming sensory data: minimising it means making the model as accurate as possible while penalising excessive revision of stored expectations. The weighting of each system — its coupling coefficient — reflects how much variational free energy is being allocated to it at any moment.
Central to active inference is the concept of priors — the brain's stored expectations about how something will unfold, built from past experience. When you reach for a coffee cup, you are not computing the reach from scratch; you are running a prediction shaped by thousands of prior reaches, correcting it as sensation arrives. A deep prior is a highly consolidated expectation, resistant to revision — the system runs largely on stored structure and its coupling coefficient changes slowly. A shallow prior is less consolidated: the system depends more heavily on current sensory data to guide behaviour, and its coupling coefficient can shift rapidly. This distinction drives the width of the mathematical envelope and the size of the dual-register ovals.
There is supporting evidence from magnetoencephalography research — particularly emerging OPM-MEG work consistent with hierarchical prediction error signalling in sensorimotor tasks — but no study has simultaneously mapped precision weighting across all 11 systems during a naturalistic action sequence. This instrument is therefore explicitly illustrative.
The faint coloured band around each curve is the mathematical envelope — the repertoire space within which the coupling coefficient can vary. Deep-prior systems (Visual, Musculoskeletal) have narrow envelopes; shallow-prior systems (Thermal — Lips, Thermal — Cheeks) have wider ones. The travelling ovals are notional and illustrative. The outer coloured oval represents the anticipation window — how far ahead the system is already predicting. This is simplified here as hippocampal anticipation, following recent work on hippocampal pre-excitation (Christoff et al., 2009). In practice, anticipation is distributed across multiple neural structures: the cerebellum runs forward models for motor prediction; the basal ganglia anticipate phase transitions; the anterior insula anticipates interoceptive and thermal state changes; the orbitofrontal cortex anticipates reward and outcome. The grey inner core of each oval represents the gross anatomical wiring — the neural connectivity underlying learned behaviour. The gap between core and outer oval is the space within which conscious access operates: the zone where learned anticipation exceeds anatomical structure. The coupling coefficient of each system is readable in that gap — a wide gap means high learned coupling, a narrow gap means the system is running mostly on structure. The sequence begins in the pre-state frame (shaded left region): no system starts from zero.
Four systems run continuously throughout and beyond the sequence — cardiovascular, respiratory, endocrine, and nervous/cerebellar. They do not take turns leading; their coupling coefficients with the intentional act remain narrow-band throughout. They are the biological substrate of the dynamic state frame: the always-present neural and physiological foundation that makes the sequence possible. These systems are not fixed — the cardiovascular system responds continuously to demand, respiratory rhythm shifts with exertion and arousal, the cerebellar system adapts its forward models — but they are dynamic within tight anatomical bounds. Their gross wiring constrains the range of variation, and within that range they are fully active. The distinction from the seven dynamic systems above is not that these four are static, but that their operating range is narrow and deep-prior: they run on consolidated structure rather than on moment-to-moment sensory updating.
The most visible event in this panel is the respiratory dip in Phase IV (Ingest). When liquid crosses the threshold of the lips and the swallowing reflex is triggered, breathing pauses automatically — the airway closes to prevent aspiration of fluid into the lungs. This is swallow apnea: a precisely coordinated deep prior executed entirely below the threshold of conscious access. The respiratory system does not stop; it executes a stored programme, then resumes. Its coupling coefficient with the intentional act briefly inverts — from passive substrate to active suppression — before returning to baseline. This is the clearest demonstration in the entire instrument of a canalised system managing a critical bodily event without any conscious direction.
This panel shows where conscious attention is directed throughout the sequence, and how wide or narrow the attentional beam is at each moment. The vertical axis runs from somatic (bottom) to cognitive (top). The named system reference lines on the right show where each body system sits in the cognitive-somatic register.
The amber centre line is the primary focus of attention at each moment. The filled amber band is the full field of simultaneous conscious access — what is in the foreground and middle ground at once. The faint blue region beyond the band shows what is available but not currently in focus — systems that could be attended to if directed, but are not. The outermost faint region is the neural bandwidth — how wide the attention band could span given the full repertoire of the neural architecture. The gap between this bandwidth and the actual band is where the organism lives: choosing, moment by moment, within the space that its neuronal connectivity makes possible. The mathematical model describes that space; the neural system enacts it; the lived behaviour is the selection between them.
The band is widest during Phase II (Grasp), when multiple systems are simultaneously available to conscious attention. It narrows sharply at the Limit Condition (Phase IV, Ingest), converging almost entirely on the somatic floor — Thermal-Lips dominates, and the rest of the ensemble recedes from conscious access while continuing to run.
The three registers in Panel 5 — fugitive, working, consolidation — are not three kinds of event. They are one kind of event measured against a threshold. This panel makes the threshold explicit, and shows that it is not fixed. In active inference the threshold is precision: the gain the system places on a channel's prediction errors, optimised by attention (Friston, 2010; Feldman and Friston, 2010). Precision and the coupling coefficient of Panel 1 are the same quantity seen from two sides — a high coupling coefficient lowers the threshold for material arriving on that channel — so the access threshold is drawn here as the inverse of the coupling envelope.
Because precision is allocated per channel, the threshold is multivariate: a bundle of per-channel thresholds rather than a single line. The band shows their spread — it narrows when the frame concentrates precision on one or two systems (the Grasp; the Limit Condition) and widens when precision is diffuse (the Phase III scatter). The plural is exact. There is no single criterion, and the bundle moves continuously as the frame re-weights. These are moving thresholds.
The consequence is that the same decay envelope clears the gate at one moment and falls short at another. A working-register activation at the opening of the frame crosses the low Phase I threshold; an activation of equal amplitude during the Grasp — when the frame has raised the gate against off-task material — does not. What reaches conscious access is decided not by amplitude alone but by amplitude measured against a threshold the frame is actively setting. The crossings are marked; sub-threshold activations are left as hollow ticks.
The threshold is also anticipatory. It dips ahead of the expected event — at contact in the Grasp, and at the Limit Condition in Ingest — because the frame pre-allocates precision before the signal arrives. This is the computational content of the pre-state frame: the pre-ignition activity of Panel 5 is, in part, the gate being lowered in advance of anything to gate.
Finally, the largest crossings feed back. A consolidation event does not only update the coupling priors; it updates the precision priors — the policy by which the frame will set its thresholds on the next reach. The frame of intention is therefore a recursive, multiple-weighting regime: it sets the thresholds, the crossings it admits revise the generative model, and the revised model sets the thresholds next time. The same gating governs the felt rate and grain of time, developed at length in the companion essay The Moving Threshold: Time as an Inference of the Coupled Organism (Feber, 2026).
As elsewhere in the instrument, the precision values are illustrative: no study has measured per-channel precision across eleven systems during a naturalistic action sequence. The construct is the claim; the curve is a sketch of it.
I first read Freud's Project for a Scientific Psychology in 1977. Written in 1895 and never published in his lifetime, it is his attempt to describe the workings of the mind in neurological terms — quantities of excitation passing between neurones, pathways worn by repeated use, contact-barriers that decide what is let through. It draws on the physicalist physiology of the Helmholtz school, in which Freud had trained, and on Fechner's psychophysics, and it marks the transition between Freud the scientist and Freud the psychoanalyst. It is intriguing that he was thinking in this way in the late nineteenth century, reaching for a physiology of mind long before there was any means to test one. What he wanted, and could not have, was the experimental apparatus — the technical ability — to carry his enquiries in neurology through to an answer. Some of that apparatus exists now, in the imaging and connectome work this panel draws on. What follows is a thought experiment undertaken in that spirit, and a frankly speculative one.
Panel 1 treats each system's coupling as a value that rises and falls. This panel looks at the imagined wiring underneath that value. The grey baseline is gross-wired anatomical adjacency — the structural connectivity between brain regions that is present regardless of learning, always active and low in amplitude. It is the biological floor on which the frame of intention runs, and none of it is available to consciousness.
Above that floor, the spikes mark topological activation. When the active inferential loop fires hard, networks lying close to it in the wiring tend to fire too. The spikes are not random: they cluster where activation is most intense — the Grasp, the Limit Condition — because strong firing in one region raises the probability of firing in the regions structurally adjacent to it. This follows the pattern established by connectome research, in which physical proximity in the brain's white-matter tracts predicts functional co-activation during cognitive and motor tasks (Human Connectome Project, from 2009).
When the frame fires, it does not respect the task's 'boundaries'. The wiring has adjacencies of its own, and activation serving the reach spills into regions that have nothing to do with coffee. A landscape. A face. A fragment of music. What surfaces is governed by the physical and learned topology of the brain rather than by the intention. The behavioural evidence is good. Dorthe Berntsen (Aarhus, from 1996) has shown that involuntary autobiographical memories — recollections that arrive without being sought — are not rare intrusions but a basic and frequent mode of remembering, observed in adults, young children and great apes, and biased toward material that overlaps the occurring situation. Functional imaging is consistent with this: the default mode network activates partway through well-learned motor tasks (Mason et al., 2007), the signature of spontaneous thought during habitual action of exactly the kind modelled here.
The mechanism suggested here is the how — the route by which such material reaches awareness during purposive action. It does not select by content. It excites whatever is structurally near: benign memories, neutral associations, and on occasion material that carries an affective charge. In the ordinary case these arisings sit at the edge of conscious access without disturbing the frame. The coffee gets picked up. The frame completes. On Berntsen's evidence this is the rule rather than the exception.
Freud returns here in a later guise. In the structural model (The Ego and the Id, 1923, and the papers on defence) the unconscious holds material kept from awareness by an active force — repression, Verdrängung — which exerts a continuous pressure to return. Berntsen treats the intrusive memories seen after trauma as a subclass of a functional process. The adjacency mechanism runs continuously. In Freudian dynamics, the adjacent arising carries enough affective charge to breach continuous suppression; in current-day neurophysiology, drawing on the active inference model, testing, modelling and selection constitute a continuous process in which 'errors' are generated constantly. Indeed, without them prediction and cognition are impossible. A symptom then becomes an inference error which breaches the bounds of the state frame.
Mark Solms has argued, on independent grounds, for grounding Freudian dynamics in the same free-energy framework used throughout this instrument. In The Hidden Spring (2021) he proposes that affect is not incidental but the primary form of consciousness — the felt registration of prediction error — and in The Only Cure (2026) he extends the case that neuroscience now bears out much of what Freud inferred. Read through the adjacency model, the account is illuminating: affectively charged material carries prediction error that has not been resolved, free energy not yet minimised, and that is what breaches the frame. The disturbance is not a by-product; it is the primary datum. The instrument stays agnostic, as the parent paper does, on Solms's further claim about the brainstem origin of consciousness (Solms and Friston, 2018).
The three registers shown here — fugitive, working and consolidation — are defined in Panel 4. Of them, the fugitive register, whose activations fade before they reach conscious access, is the most speculative element of the instrument, and is offered as such.
The five panels before this one display activity — coupling weighted and re-weighted, attention narrowed and widened, thresholds set and reset. What we have been calling the act, the reach for coffee, is not a steady state the system is in but a stabilisation it achieves: inference settling, briefly and contingently, into a form steady enough to act through. This coupling between organism and local ecology — the temporary boundaries we called the state frame — is intelligence.
One familiar move is to say that intelligence is not a noun but a verb. The phrasing is Buckminster Fuller's — "I am not a thing — a noun. I seem to be a verb" — from his 1970 book of that title, and it is a useful corrective to the habit of treating the mind/brain as an object. But it is not a particularly useful way to describe what is going on. A verb implies a steady, continuous action, and that is not what these panels show. They show a process that settles into an actionable form and then gives way — a contingent stabilising and releasing, not a doing that simply runs on. Trading the noun for a verb swaps one label for another. In the terms of Panel 1, each frame of intention is a transient low point in the free-energy landscape — a metastable state the system holds and then leaves, steady enough to organise the reach, unsteady enough to release into the next (Kelso, 1995; Tognoli and Kelso, 2014). This is not the claim that the appearance is an illusion. Nothing here is pretending to be something it is not. There is only process, held 'still' momentarily.
As elsewhere in the instrument, the construct is the claim and the dynamical language is its sketch. Metastability is an established idea in coordination dynamics, increasingly applied to large-scale brain activity; identifying it with the frame of intention, as this panel does, is a theoretical proposal and is offered as one.
The meta in meta-stabilisation marks a second key property: the system stabilises its capacity to keep producing frames; it is stably unstable. That is the recursive regime set out in Panel 4. What persists is not any one frame but the standing ability to keep forming them — a continuous temporal stabilisation loop, a short working space in which a settled form can be held just long enough to be acted from before it is let go.
This loop is not one property of the system among others; it is what lets the system act at all. A process that could not hold itself still, even briefly, could form no intention and complete no reach. It is the deeper reason there is no cold start: the pre-state frame is simply the prior stabilisation, not yet released, already serving as the ground of the next. Multitasking is being alive: an organism persists by holding itself, far from equilibrium, within a narrow band of viable states, a self re-stabilised from moment to moment against its own dissipation (Friston, 2013; Maturana and Varela, 1980; Thompson, 2007). Meta-stabilisation is not something intelligence does, it is what living is.
This is the plainest claim the instrument can make. Intelligence is not a faculty the organism consults, not a quantity it stores, not a separate realm laid over the body's systems. It is the name we give to this continuous, contingent stabilising of inference across coupled systems, and there is no separate thing to find because there is nothing for it to be made of but the loop, held. This is what is meant by the being of being human: not a property we have, but a process we are, and have to go on being in order to be at all.