Chapter 9 — The Embodied Present
Stratum 2: allocentric spatial governance as the second cognitive closure; the embodied stream of Stratum 3 (valence as real-time pricing of trajectories) fused with it; the reafference debt received from Chapter 8 and discharged here, with the hard problem of felt experience handed forward to Chapter 10.
***
§9.0 — The Floor Is Not Where the Floor Is
She has done this ten thousand times. Sit at the edge of the bed, plant both feet, shift weight forward, rise. The procedure is older than memory. It is so deeply below intention that procedure is the wrong word — there is no step one and step two, there is only the unanalysed motion of getting up in the morning, a motion so trusted that its machinery has never once required her attention.
This morning the machinery is not there.
She plants her feet and the feet report back correctly — pressure on the heels, the grain of the rug, the small coolness of the floorboards where the rug ends. She shifts her weight forward and the weight shifts. The room is exactly where it was last night. The window is in the window's place; the doorframe is in the doorframe's place; the light through the blinds falls in the stripes it has always fallen in. Nothing in the visual field has moved. Nothing in what her feet are telling her has moved. And yet as she comes up off the mattress the room begins, impossibly, to tilt — not to rotate around her, which she would recognise as dizziness and would know what to do with, but to detach, as though the floor she can see and the floor she can feel were two different floors reporting two different stories about where the centre of the earth is, and neither one of them is willing to defer to the other.
She puts a hand out for the nightstand. The nightstand is where it has always been. She can see it. Her hand arrives at it. Her fingers close on the wood. The wood is solid. None of this helps. The information is correct and the information is useless, because the problem is not that any single signal has lied to her; the problem is that the signals no longer compose. Vision says the room is level. The soles of her feet say the room is level. Some third thing, something she has never had to name because she has never had to notice it, is no longer arriving at all — and in its absence the level room will not stay level, because the levelness of the room was never a fact about vision and never a fact about the soles of the feet. It was a fact about the agreement among them, and the agreement has lost its convenor.
This is the eighth morning. The first morning, three weeks ago, she thought she had an ear infection. The second morning she thought it was vertigo and waited for it to pass. By the fifth morning the word gentamicin had been introduced into her vocabulary, and with it the information that the drug that had saved her life in the intensive care unit had also, in a small percentage of patients, destroyed the hair cells of both vestibular apparatuses — bilaterally, symmetrically, permanently — and that she was one of them. She has read the pamphlet. She knows, in the propositional sense, what has been lost. What the pamphlet did not tell her, because the pamphlet could not tell her, is that knowing what has been lost does nothing to restore the composition. The composition is not a belief. It is prior to belief. It is the silent referee without which the scene cannot hold together as a scene, and no quantity of propositional knowledge about its absence can stand in for its work.
She closes her eyes. Immediately it is worse. With vision gone, even the illusion of a stable reference collapses, and she is not standing in a dark room but floating in a direction she cannot name, falling in a direction she cannot name, being pulled toward a down that is not underneath her and may not exist. She opens her eyes. The room returns, stable-looking, a picture of a room. It is a picture she can no longer inhabit. It arrives as information rather than as place. She is looking at the bedroom the way a person looks at a photograph of a bedroom — correctly, accurately, and from outside it.
What has failed is not a sense. She has all her senses. What has failed is the operation that made the senses into a world — that stitched the seen floor to the felt floor to the remembered floor and delivered, beneath every morning of her life, the single unexamined fact that the floor was where the floor was. The stitch is gone. The threads are intact. She can see each thread. She cannot, by any effort of attention, re-tie them. The attempt to tie them — the deliberate, conscious effort to reconcile the visual horizon with the bodily axis — is itself exhausting in a way that no physical exertion has ever been exhausting, because she is doing in the bright daylight of conscious effort what a deeper system used to do silently, continuously, and for free.
She takes a step. The step lands. The next step lands. She crosses the room by treating each contact with the floor as a separate event requiring separate verification, and by the time she reaches the door she is drenched in sweat. It has taken her forty seconds to walk four metres. The caloric cost of this crossing is many times what it was three weeks ago, because every correction that her vestibular system used to perform beneath notice now has to be performed through vision and proprioception under conscious control, and the control loop runs at a fraction of the speed and at many times the metabolic price. She leans against the doorframe. The doorframe is where the doorframe is. That, at least, has not changed. What has changed is that this no longer feels like an observation. It feels like an achievement.
The neurologist, two weeks earlier, had asked her to describe the experience, and she had said — groping for the words, which did not exist in ordinary language because ordinary language presupposed the operation that had failed — that it was not that the room moved and not that her body moved but that the relationship between them had come apart. He had nodded carefully. He had not told her that what she was describing was the live signature of a coordination architecture whose operation is ordinarily invisible because continuous and costless, and whose failure therefore presents not as the loss of a signal but as the loss of a ground — the silent condition for there being a scene at all.
She will learn, over the coming year, to walk again. She will never again walk the way she walked three weeks ago. She will walk the way a skilled operator walks, consciously deploying vision and proprioception to reconstruct, at metabolic expense and cognitive cost, something that previously arrived for free. Under good lighting and on flat ground she will pass for unimpaired. In the dark, on uneven terrain, with her eyes closed in the shower, she will be reminded — every day, for the rest of her life — that the world she inhabits as a world is the continuous output of a machine she cannot feel running, and that when the machine stops the world does not become a different world. It becomes a picture of a world, correct in every particular and impossible to stand up in.
This is the disparation the present chapter will name, mechanise, and price. For now it is enough that the reader has stood, for a moment, on a floor that was not where the floor was.
§9.1 — The Debt Is Specific
Chapter 8 closed with an IOU whose terms are now callable. The bioelectric closure solved the morphogenetic coordination problem at the tissue scale — convergence on a stable anatomical target under a fixed genome, with cell-level noise compressed into tissue-level invariance through a voltage pattern distributed across a gap-junction network. That solution was purchased at the ordinary currency of ion-pump ATP, and its success installed the platform on which multicellular motility became possible. Precisely that success is what generated the reafference problem. Once the organism moves — once the eyes saccade, once the head turns, once the body translates through a structured environment — the sensory surface is continuously rewritten by the organism's own actions, and the morphogenetic architecture that stabilised the tissue has no operators for distinguishing a change in the world from a change produced by the organism's movement against a stationary world. The patient of §9.0 lost a component of this distinction late in life and paid the metabolic price in sweat and sustained attention; the evolutionary problem is that no organism capable of locomotion can afford to pay that price continuously. The reafference problem is not a perceptual puzzle; it is a thermodynamic one. The organism that cannot subtract its own contribution from the sensory stream must either move slowly enough that the contribution is negligible, or move quickly and pay, at every step, the full metabolic cost of reconstructing the scene from below. Neither option scales to the coordination demands of a vertebrate ecology.1
The specific form of the debt is sharper than this. The reafference problem does not merely require that self-generated sensory change be subtracted. It requires that the residue — what remains after subtraction — be interpretable as distal structure: a layout of surfaces, obstacles, resources, and conspecifics whose positions are stable across the organism's movement through them. The residue must have a coordinate frame, and the coordinate frame must not be tied to the moving body, because a body-centred frame offers no invariance across locomotion. A different frame is required — one whose axes are referenced to the environment rather than to the mover — and the operators that read this frame must compose consistently with the body-centred frame in which limbs are actually commanded. Two frames, then, not one, and the machinery to translate between them in real time at a speed the locomoting organism can afford. This is the structural shape of the debt: a second coordinate system, maintained in parallel, coupled through a translation whose failure is diagnostic of the whole apparatus. §9.0's patient is the live demonstration of that coupling's failure — the two frames remain, each internally coherent, but the translator has been denied its input.
The chapter discharges the debt through the installation of a distributed governance architecture for allocentric space, running on a cortical substrate whose metabolic cost — measured as the cerebral metabolic rate of oxygen, CMRO₂, in the tissue volumes that implement it — is of a different order of magnitude from the bioelectric layer beneath. What the prior closure spent on maintaining a morphological target across tissue, the present closure spends on maintaining a coordinate frame across motion, and the signature of cessation is not morphological dissolution but the precise disorganisation §9.0 described: a scene that arrives as information rather than as place. Success at this stratum generates its own downstream crises, two of which this chapter must also begin to discharge. The first is that a coordinate frame, once installed, does not by itself tell the organism which trajectories through it are worth taking. The second is that trajectories must be evaluated in time — before consequences are realised — and the evaluation must price lethal costs high enough to prevent the organism from traversing them even once. These two pressures fuse the embodied-present component of Stratum 3 onto the allocentric closure of Stratum 2, and the fusion is what §9.3 will name. The full hard-problem work on valence as phenomenal character is Chapter 10's business and is not done here.
The specific form of the debt admits one further precision, which the chapter will rely on but not re-argue. What the organism needs is not merely a coordinate frame but a frame whose axes are allocentric — anchored to the environment — while the motor system that acts within it remains necessarily egocentric — anchored to the moving body. The two frames are not redundant. An allocentric frame without an egocentric translator produces a patient who can describe the layout of a room without being able to walk through it; an egocentric frame without allocentric anchoring produces an organism locked in stimulus-response reactivity, capable of chasing the signal in front of its sensors but incapable of returning to a goal it has lost sight of. The spatial closure this chapter installs is not a single representation but a coupling architecture that maintains both frames in parallel and performs the translation between them in real time. Neither frame alone is the closure; the closure is the coupling, and the coupling's metabolic cost is what §9.2 will price.
One further inheritance from the prior stratum. The bioelectric Witness distributed its morphological target across gap-junction-coupled tissue; the allocentric closure distributes its coordinate frame across neural populations coupled through fast, directional synaptic signalling. The substrate has changed from electrochemical coupling at millisecond latencies to synaptic coupling at sub-millisecond latencies, and the topology has tightened from diffuse tissue-scale propagation to specified axonal projection. The functional logic — distributed publication, attractor compression, constitutive coupling — is invariant across the substrate shift. What the Stabilisation Engine of Chapter 6 theorised at substrate-indifferent generality, and what Chapter 8 installed at the bioelectric substrate, re-instantiates here at the neural one. The re-instantiation is noted and not re-installed; the apparatus travels.
***
§9.2 — The Floor Where the Floor Is
The canonical mammalian solution to the allocentric-frame problem is a distributed circuit whose main components have been individually characterised and jointly modelled to a degree that no other cognitive closure in this book will match. Head-direction cells in the presubiculum and anterior thalamic nuclei fire as a function of the animal's facing direction in the environment, each cell tuned to a preferred compass bearing and the population tiling the full 360° as a ring attractor whose bump tracks head rotation at the speed and precision of the movement itself. Place cells in the hippocampus fire when the animal occupies specific locations in a familiar environment, each cell coding a roughly Gaussian firing field and the ensemble jointly tiling the space. Grid cells in the medial entorhinal cortex fire at the vertices of a hexagonal lattice that tessellates the environment at a scale specific to the cell's dorsal–ventral position, the lattice maintaining its geometry across environments and rescaling systematically when the enclosure is deformed. Boundary-vector cells, object-vector cells, and speed-tuned populations in the entorhinal–hippocampal network supplement this apparatus with signals coding distance and direction to environmental features and the animal's own translational velocity. The empirical lineage — O'Keefe's discovery of place cells in 1971, the Mosers' identification of grid cells in 2005, and the subsequent mapping of the full ensemble across the entorhinal–hippocampal circuit — is the laboratory anchor on which this section rests.
The architecture's significance for the present argument is not that it represents the environment. It is that it stabilises a coordinate frame whose cells fire as the organism moves through a world that is not moving, and whose firing is sustained across the sensory transformations movement generates. Head-direction cells maintain their tuning when the animal walks in a straight line through a visual field that translates past the eyes at the locomotor rate. Place cells continue to fire at their preferred locations when the visual scene changes radically between entries — when lights go off, when a landmark is removed, when the animal enters a familiar environment from a novel direction. Grid cells maintain their hexagonal lattice during locomotion in the dark, driven by self-motion signals alone, and resynchronise to visual landmarks when the landmarks return. The pattern is not a picture of the environment stored in a neural screen. It is a set of firing regularities that preserve their structural relations to the environment across the transformations that moving through the environment imposes on the sensory surface. That is not a representation. That is an attractor — a stable configuration of a coupled dynamical system — whose governance extends across the tissue volumes it recruits, and whose invariants are the invariants of place, not of image.
The distributed publication of this attractor — the pattern's realisation across independent cellular supports, none of which individually carries the coordinate frame and all of which jointly sustain it — is the architecture's inheritance of the Stabilisation Engine's publication criterion from Chapter 6. No single head-direction cell encodes north; the population does. No single place cell encodes the animal's current location; the ensemble does, through a pattern of coincident firings whose decoding recovers position to centimetre precision in laboratory enclosures. Ablate ten percent of the hippocampus and the animal's spatial behaviour is mildly impaired; ablate most of it and the behaviour collapses, but the collapse is graceful in a diagnostically specific way — the animal does not become confused about where things are in a body-centred sense, it becomes confused about where things are relative to each other and to the environment as a whole, which is the signature of an allocentric rather than egocentric failure. The distributed nature of the publication is the architecture's robustness.1
The compression of this publication into a re-usable invariant is the architecture's inheritance of the Canon criterion. Hippocampal replay — the high-frequency, compressed re-activation of place-cell sequences during sharp-wave ripples, occurring in rest and in sleep — performs the compression in vivo and on the record. The same sequence of place cells that fired during the animal's traversal of a corridor fires again, in the same order at a twentyfold temporal compression, when the animal pauses at the corridor's end or sleeps after the day. These replays are not passive echoes. Their disruption — by targeted stimulation during ripple events, in the now-standard experimental paradigm — impairs subsequent performance on the spatial task whose trajectories were replayed, while leaving locomotor and sensory capacities intact. The replay is the operation by which a traversed trajectory becomes a re-usable compression of the space — an invariant that can be retrieved, re-activated, and combined with other such invariants without the animal having to re-traverse the terrain. Publication across the ensemble; compression through replay; the two constitutively coupled, because replay is only coherent if the publication was coherent, and the publication is only useful if the replay can compress it. This is the Gluing threshold met at the cognitive stratum.
The publication side of the architecture extends further than the hippocampal–entorhinal circuit's own tissue. Spatial structure is inscribed redundantly across topographic maps — retinotopic in primary visual cortex, somatotopic in primary somatosensory and motor cortices, tonotopic in primary auditory cortex — each of which preserves neighbourhood relations of its sensory surface and each of which contributes an independent support for the distal regularities the hippocampal–entorhinal ensemble compresses. Blindness does not abolish navigation; the congenitally blind subject navigates through somatic, auditory, and motor-efference sources whose spatial structure the intact topographic maps continue to publish. Deafness does not abolish navigation. The loss of any single modality degrades the publication without abolishing it, and the graceful degradation is the diagnostic signature of a distributed Witness whose redundancy was built for precisely this kind of insult. Motor repertoire itself functions as spatial inscription: the geometry of reachable positions, the kinematic envelope of the body's own traversable trajectories, encodes the layout in the grammar of action as well as in the grammar of sensation. Publication, at this stratum, is not a single neural population's business. It is the distributed labour of every cortical map whose neighbourhood structure preserves a spatial regularity, and the compression that reads them all is what the hippocampal–entorhinal architecture performs.
The production of spatial invariants through action rather than through passive registration deserves its own sentence, because the distinction is the structural signature of this stratum's governance. The organism does not receive a spatial world and then compute trajectories through it. It generates the spatial world by the systematic perturbation of its own sensory surfaces through motion and the continuous correction of motor prediction against sensory return. Buzsáki's inside-out framing of this process — spatial representation as action-structured rather than passively assembled — is the contemporary formulation of what the closure architecture installs: an organism that is not a camera recording a pre-existing space but a generator whose motor commands and sensory corrections compose, iteratively, the stable invariants that constitute space-for-the-organism. The allocentric frame is not found. It is built, action by action, prediction by correction, and its maintenance is the work the burn-rate pays for.
The translation between allocentric and egocentric frames — the operator whose failure §9.0 described from the inside — is implemented across the retrosplenial cortex and posterior parietal cortex, where cells jointly tuned to environmental features and to the animal's current heading perform the coordinate transformation that lets an allocentric goal be acted on by an egocentrically commanded body. The retrosplenial lesion patient, and the dense amnesic with bilateral hippocampal damage, present with disorders whose phenomenological signatures are diagnostically specific: the retrosplenial patient can describe the layout of a familiar environment but cannot navigate through it; the hippocampal amnesic can navigate through the moment's affordances but cannot assemble them into a stable map. The operator has two halves, and the halves dissociate. §9.0's patient presents with a third disorder — one in which neither half is individually broken but the signal that conjoined them has been denied, and the frames drift apart in real time because the translator has nothing to translate.
Reafference cancellation in this architecture is not a subtraction performed once. It is continuous, distributed, and paid for in metabolic currency at every moment the organism is awake. The cerebellum issues forward models that predict the sensory consequences of motor commands and subtracts those consequences from the incoming stream before the residue reaches the cortex that reads it. The vestibular nuclei integrate semicircular-canal and otolith signals with neck-proprioceptive and visual-flow information to maintain the gravitational reference across head movement, and the integration runs at latencies below ten milliseconds because anything slower would be useless for the behaviour it supports. The cortical map consumes oxygen at a rate measurable by functional imaging and terminates, within minutes of its interruption, in the scene-without-place that §9.0 described. CMRO₂ is the burn-rate currency of this stratum, and the signature of its cessation is the specific disorganisation in which each sensory channel reports correctly and the report no longer adds up to a world.
The consequence of this architecture for perceptual structure is that every perception is perspectival — indexed, necessarily and architecturally, to the moving reference point the body constitutes. The egocentric frame is not an optional phenomenological commentary the architecture could have done without; it is the minimal structural consequence of solving the reafference problem at all, because any system that stabilises distal invariants under self-motion must index its representations to a reference point that is itself in motion, and that reference point is the viewpoint. The viewpoint is real. The awareness of having a viewpoint is a different question, belonging to a different stratum, and the framework's discipline is to install the first without projecting the second. The spatial agent faces into a field. The reflexive fold by which it would also know itself to be facing awaits the installations of later chapters.
Failure modes follow directly from this architecture. The place-cell ensemble can be intact while the grid lattice is disrupted, producing a specific pattern of spatial disorientation in which landmark-based navigation is preserved and path integration collapses. The grid lattice can be intact while hippocampal replay is disrupted, producing normal moment-to-moment navigation and severe impairment in the consolidation of trajectories across sleep — the rat runs the maze today and does not remember it tomorrow. Head-direction stability can be preserved while the translation to egocentric frame is lost, producing the retrosplenial syndrome in which the map is known and cannot be walked. The reafference-cancellation apparatus can be denied its vestibular input, producing §9.0's phenomenology — the double-footed ground whose two reports cannot compose. Each of these failures is the loss of a specifically nameable operation, and the diagnostic specificity is what licences the closure claim: a real capacity breaks in the specific ways it breaks, and the ways are catalogued.
What has been installed is not an inner map. It is a governance architecture for an allocentric coordinate frame, maintained across sensory transformation by continuous reafference cancellation, compressed into re-usable invariants through replay, and translated into egocentrically commanded action through the retrosplenial and parietal operators whose specific lesion syndromes catalogue the architecture's seams. The frame is not a model of the environment in the sense in which a drawing is a model of a landscape. The frame is a stable set of firing regularities whose invariants are the invariants of distal structure, instantiated in a tissue whose CMRO₂ pays for their maintenance and whose interruption returns the organism to the reafference problem Chapter 8 handed forward. A behaviourist reading of the same animal — one that catalogued stimulus-response associations and refused to posit any internal constraint — would account for the animal's performance in stable familiar environments and fail to account for what the architecture was built to make possible: detours around obstacles never encountered, navigation toward goals through multiple distinct routes, efficient exploration that behaves as if guided by a layout the animal has constructed through its own movement. That flexibility is unintelligible under stimulus-response operators alone, and its unintelligibility is the empirical receipt for the structural claim the present chapter stakes. Flexibility of this specific kind is the scar-tissue proof that a coupling architecture has been installed, because the associative present has no operators for it.
***
§9.3 — The Ledger in the Body
A coordinate frame by itself does not move. The animal that possesses place, grid, and head-direction cells in immaculate working order, and whose retrosplenial translator is perfectly tuned, still faces a question the spatial closure cannot answer: of the trajectories this frame makes available, which ones are worth taking. The question is not abstract. It is metabolic, and it is continuous, and it is answered in every second of the animal's waking life by a machinery that runs in parallel with the spatial apparatus and fuses to it so tightly that the fusion is the embodied present itself.
The machinery is interoceptive. The organism's internal state — blood glucose, osmolarity, core temperature, tissue oxygenation, gut fill, gonadal state, immune activation, and a long list of more specific variables whose homeostatic set-points are independently maintained — is continuously sampled by visceral and humoral afferents whose targets are the nucleus of the solitary tract, the parabrachial nucleus, the hypothalamus, and, cortically, the insula. The sampling is not a read-out of a stable body. It is a prediction: the forward model of interoceptive state that a well-calibrated organism maintains is a model of where its internal variables are about to go under the current trajectory, and the discrepancy between the predicted interoceptive state and the arriving one is the precision-weighted signal from which valence is composed. Valence is not a sensation added on top of a cognitive map. It is a real-time pricing of trajectories through the map, computed continuously by a machinery whose burn-rate currency is the allostatic cost of maintaining the homeostatic set-points themselves.
The fusion of this pricing with the spatial frame is not conceptual; it is anatomical. The hippocampus receives dense projections from the amygdala and the ventral striatum; place cells' firing is modulated by the reward value of the locations they code, with preferential replay of sequences that terminated in reward during subsequent sharp-wave ripples. Grid-cell firing is modulated by task demands and by the animal's motivational state. The retrosplenial and parietal cortices — the translators of the prior section — receive inputs from the insula and the anterior cingulate that weight the allocentric-to-egocentric transformation by the affective stakes of the trajectory being commanded. The animal does not first compute a map and then price the trajectories through it. The map is maintained in a tissue whose ongoing operation is shot through with the pricing, and the pricing is inseparable from the map it prices. This is what it means for Stratum 3's embodied component to be fused with Stratum 2 rather than added to it.
The present chapter's interest in this fusion is strictly navigational. Valence here does the work of real-time trajectory selection: it determines which of the allocentrically available paths the organism will take in the next second, in the next minute, in the next hour. It does not yet do the work of phenomenal character — the felt quality of valence as experienced weight, the interoceptive ledger as something the organism reads rather than merely runs. That work requires an apparatus this chapter does not install. It requires the structural-number dissolution that Chapter 10 will conduct, with its named engagements to Seth, Chalmers, Levine, Frankish, and the somatic-marker and insular-lesion literatures. Those obligations belong where they belong. What the present chapter takes from the embodied-stream component of Stratum 3 is only this: the navigational closure is not a bare coordinate frame but a priced one, and the pricing is the interoceptive machinery doing its continuous real-time work on the trajectories the frame makes available.2
The failure modes of this fused architecture are correspondingly continuous with §9.2's. A spatial frame whose trajectories are unpriced is not a usable frame; the animal with intact place cells and an ablated amygdala navigates but does not discriminate between the paths that lead to food and the paths that lead to a predator's ambush, and discrimination failure at this layer looks like spatial failure even though the cartography is preserved. An interoceptive apparatus whose pricing is decoupled from the spatial frame — an animal whose valence is correctly computed but fails to modulate the place-cell ensemble's replay bias — fails in a different way: the map is intact, the pricing is intact, and the two no longer speak. The pharmacological and lesion literature on ventral-striatal and amygdalar disconnection syndromes catalogues these failures with enough specificity to distinguish them from the retrosplenial and hippocampal syndromes of §9.2, and the distinction is the chapter's next load.
Valence handed forward. The reader who wants the full account of what it is to feel a trajectory as lethal or as nourishing — the account in which felt weight is not epiphenomenal commentary on the pricing but the phenomenal character of the pricing itself — is owed the next chapter, where the structural-number argument carries the work. The present chapter's obligation is discharged when the embodied present has been specified as an allocentric coordinate frame maintained across reafference and priced in real time by an interoceptive ledger whose cessation signatures — disorientation and anhedonia, place-blindness and amotivation, and the specific fusions and dissociations between them — have been catalogued. With that specification on the record, the chapter can turn to the load-bearing philosophical engagement this stratum requires: the Free Energy Principle, which names much of what has just been described but names it in representational vocabulary, and from which the framework of this book must be precisely and sustainedly distinguished.
§9.4 — Attractors Without Models
The architecture §9.2 installed and §9.3 fused will be recognised, by a reader familiar with computational neuroscience in its current dominant form, as a candidate application for the Free Energy Principle. The recognition is not accidental. Karl Friston's programme, developed across the last two decades and now installed as the standard theoretical frame for a large fraction of work in predictive processing, perceptual inference, active inference, and interoceptive control, names much of what the prior two sections described: prediction-error minimisation as the operating principle of the nervous system, forward models that subtract self-generated perturbation from the sensory stream, precision-weighting of sensory channels according to their reliability, and a thermodynamic rationale — the organism as a self-maintaining far-from-equilibrium system that must bound its sensory surprise to avoid dissolution into its environment. Large portions of the empirical apparatus §9.2 deployed — reafference cancellation, the cerebellar forward models, the hippocampal–entorhinal circuit as an inference engine over spatial structure — are routinely read, in the contemporary literature, through Friston's vocabulary. The framework of this book cannot pretend the overlap does not exist. It also cannot concede it. The present section specifies, as precisely as the philosophical territory allows, where Compositional Immanence and the Free Energy Principle agree, where they diverge, why the divergence matters at this stratum in particular, and what the framework developed here predicts that the standard FEP formulation does not.
The agreements are substantial and should be stated first, because the argument that follows depends on their being granted without reservation. Prediction-error minimisation is a real mechanism, operating at every cortical and subcortical site where forward models interface with sensory returns, and its calorimetric signatures — the precision-weighting of channels under uncertainty, the attenuation of sensory gain during self-generated action, the metabolic asymmetry between expected and unexpected inputs — are among the most replicated findings in contemporary systems neuroscience. Organisms are far-from-equilibrium thermodynamic systems whose persistence depends on bounding their own dispersion into the environment; this is not a claim proprietary to FEP but one FEP articulates with unusual mathematical discipline, and the framework of this book takes it as given. The variational-inference formalism Friston and collaborators have developed offers a genuinely useful compression of a large class of neurobiological phenomena, and the framework of this book has no interest in disputing the formalism's empirical range. The disagreement is not about whether the mathematics describes the data. The disagreement is about what the mathematics describes the data as being about.
The divergence is ontological, not empirical, and it is precise. In the standard FEP formulation, the nervous system infers the hidden causes of its sensory returns by maintaining a generative model of the environment — a probabilistic representation of the external world's unobservable states — and minimising the free-energy bound on the surprise of the returns under that model. The generative model is, in the canonical writings, an internal representation of the external world's causal structure, and the brain's operations are operations on this representation: inference, model-updating, action selected to confirm the model's predictions. The Markov blanket — the statistical boundary that separates the organism's internal states from its external ones — does the work of defining whose representations these are and what they are representations of. This is a representationalist ontology. It may be a sophisticated one; it may be sufficiently hedged by its proponents that its representationalism is occasionally denied or qualified; but the load-bearing inferential vocabulary — hidden causes, generative model, posterior beliefs, model evidence — is representational through and through, and the programme's explanatory power depends on its being so. Andy Clark's extended and embodied versions of predictive processing are more ecumenical about where the representations live and how tightly they are bound to the skull, but they do not dispense with the representational relation; they distribute it. The representation is still the relation between a vehicle inside the organism (or inside the organism-plus-tool) and a content about the world outside.
The present framework rejects the representational ontology at the stratum this chapter installs. Place cells do not represent locations. They are a tissue-level attractor whose firing regularities preserve their structural relations to distal structure across the sensory transformations locomotion imposes. Head-direction cells do not represent compass bearings. They are a ring attractor whose bump tracks head rotation because the circuit's coupling topology forces it to. Grid cells do not represent a hexagonal lattice over space. They are the attractor-generated firing pattern a medial entorhinal circuit exhibits under self-motion drive, with the lattice's geometry determined by the circuit's own dynamics rather than by a model it is running. The governance these architectures perform is not inference over a hidden external state. It is the ongoing maintenance, at continuous metabolic cost, of a coupled dynamical system whose stable configurations — whose attractors, in the strict dynamical-systems sense — are the invariants of allocentric space. The distinction is not cosmetic. An attractor is a feature of a dynamical system's phase portrait; a representation is a relation between a vehicle and a content. The first requires only that the system have the coupling topology it has; the second requires a content-fixing relation whose ontological status has been the central unresolved problem of philosophy of mind for forty years. Compositional Immanence does not owe the framework a solution to that problem because it does not incur the debt.
The Markov-blanket-as-boundary-condition plays the role, in Friston's architecture, that Witness/Canon/Gluing plays in the architecture installed at Chapter 6 and deployed since. The analogy is suggestive and the difference is decisive. A Markov blanket is a statistical condition — the conditional independence of the organism's internal states from the external ones given a set of blanket states — whose satisfaction defines what counts as the organism for inferential purposes. Witness/Canon/Gluing is a closure architecture whose satisfaction defines what counts as a governed pattern: the pattern must be distributed across independent supports (publication), compressed into stable invariants (compression), and the two operations must be constitutively coupled such that each sustains the other. The Markov blanket tells the theorist which variables to put on which side of the inferential problem. The closure architecture tells the theorist whether a candidate governance has actually installed itself as a real, thermodynamically priced stratum of organisation, and it does so without requiring that the governance be about anything. The cost of the framework's closure architecture is that it must pay, at every stratum, a burn-rate currency and name a signature of cessation. The cost of the Markov-blanket formalism is that it must pay, at every stratum, an account of what the blanket's internal states are representations of and how the representational relation is fixed. Each cost is substantial. Only one of them is unpaid in contemporary philosophy of mind.
Why does the divergence matter at Stratum 2 specifically, rather than at the abstract level of metaphysics? Because the allocentric coordinate frame this chapter installed has three empirical properties that a representationalist ontology strains to accommodate and an attractor ontology predicts directly. First, the frame is stable under sensory ambiguity in a way that inference should not license: place cells maintain their firing fields in total darkness during path integration, when the sensory evidence for location is the organism's own motor efference copy and a vestibular signal, and no posterior over external location computed from these inputs alone should be narrow enough to produce the firing precision the cells actually exhibit. The attractor account predicts the precision: the cells are locked into a coupled dynamical configuration whose stability is a property of the circuit's recurrent topology, not of the evidence currently feeding it. The inferential account must posit priors strong enough to carry the stability across evidential thinning, and those priors then require their own account of where they came from and why they take the form they do — an account that, in the standard FEP literature, is typically deferred to evolution in a gesture that does no explanatory work at the stratum under examination. Second, the frame exhibits scale-specific lattice geometry — the hexagonal grid with its characteristic spacing ratio across dorsal–ventral modules — that is derivable from attractor dynamics in recurrent networks with appropriate coupling kernels, and that requires, under the inferential reading, a story about why the brain's posterior over spatial structure would converge on precisely this geometry rather than the continuous Gaussian posteriors more natural to variational inference. The attractor account predicts the geometry; the inferential account accommodates it. Third, the frame fails in the specific ways §9.0 and §9.2 described — scene-without-place under vestibular loss, place-without-layout under retrosplenial lesion, layout-without-consolidation under replay disruption — and these failures are predicted by which components of an attractor architecture have been disrupted. Under the inferential reading, the same failures must be explained as specific degradations of specific posteriors, which requires the theorist to posit, for each failure, exactly which distribution over which hidden cause has been compromised. The explanatory cost is not trivial, and the fit with the clinical phenomenology is, at every turn, tighter for the attractor account than for the inferential one.
Andy Clark's position, the more moderate representationalism of Surfing Uncertainty and its successors, is worth distinguishing from Friston's strict-inferential stance because Clark's framework shares with Compositional Immanence the emphasis on embodiment, the rejection of armchair-cognitivist computationalism, and the insistence that prediction is continuous with action rather than an antecedent cognitive process that action then implements. The shared ground is substantial. Both frameworks accept that perceiving is an active engagement with the environment rather than a passive reception of stimulus; both frameworks reject the sharp separation between cognitive and motor systems that classical computational architectures presupposed; both frameworks treat the organism as constitutively bound to its ecological niche through continuous sensorimotor coupling rather than as a disembodied inference engine that happens to be housed in a body. The difference, despite the shared ground, remains representational. Clark's predictive brain is still running a generative model; the model is distributed, embodied, action-involving, ecologically embedded, extended where ecological circumstance licences extension — but it is a model, a vehicle whose content is the environment's causal structure and whose function is to reduce prediction error with respect to that content. Compositional Immanence has no such vehicle and no such content. What Clark's programme describes as a distributed generative model, the framework of this book describes as a coupled dynamical system whose attractors are the invariants that the representationalist vocabulary interprets as the model's posteriors. The empirical predictions converge in most laboratory regimes — the two frameworks will agree about what the animal does, about what its cells fire in response to, about what its lesion will produce. The philosophical commitments diverge in every regime, and the divergence becomes consequential the moment the framework is asked to account for the kinds of failure that do not fit the inferential frame: the attractor-hijacking failures Chapter 8 catalogued at the bioelectric stratum, the coupling-specific failure signatures §9.5 will install at the allocentric stratum, and the structural-number phenomena Chapter 10 will take up at the affective. The more ecumenical Clark is about where the model lives, the more pressure the word "model" is carrying, and at each pressure-point the question is whether the theoretical work is being done by the model's representational content or by the coupling dynamics that the representational vocabulary is describing. The present framework's wager is that the dynamics are doing the work and the representation is a reading of them.
What does Compositional Immanence predict that the standard FEP formulation does not? The specific failure-mode inventory attached to attractor architectures rather than to inference processes. An inference engine fails by producing the wrong posterior — by miscomputing the likelihood, by deploying a mis-specified prior, by weighting channels against their actual reliability. An attractor architecture fails in a different taxonomy: the attractor basin can be disrupted (the hippocampal amnesic's graceful but diagnostic collapse); the attractor can be hijacked by a signal exploiting its control topology (the Toxoplasma-style failure Chapter 8 catalogued, which has no natural analogue in the inferential frame because inference-hijacking is not the same structural phenomenon as attractor-hijacking); the publication supporting the attractor can be decoupled from the compression that sustains it (the Witness/Canon failures this chapter's architecture is built to diagnose); and, most importantly, the burn-rate currency sustaining the attractor can fail while the circuit's topology remains intact — producing the metabolic signatures of cessation, hypoxic and hypoglycaemic in this chapter's case, that the inferential account treats as boundary conditions on the inference rather than as constitutive of the closure. These distinctions are empirical; they are testable; and at each point where CI and FEP make divergent predictions, the laboratory regime that discriminates between them is at least specifiable, even where the experiment has not yet been run.
The philosophical cost of the attractor ontology, to be named honestly, is that it denies the framework a certain kind of integration with the long tradition of philosophy of mind that has taken representation as its central unit of analysis. The framework cannot help itself to content-involving vocabulary without betraying its own commitments, and the self-discipline this requires will be visible at every subsequent chapter. The benefit is that the framework also declines the debts that representational vocabulary incurs — the teleosemantic problem of content fixation, the disjunction problem, the indeterminacy-of-reference problem, the hard problem of intentionality — none of which arise at the framework's stratum because the framework has not installed the relation whose obscurity generates them. Place cells do not mean locations; they are the tissue-level attractor of the locations' invariants. The two formulations are not equivalent, and the difference between them is the difference between a framework that has absorbed ninety years of unresolved philosophy of mind and a framework that has declined to incur its obligations.
Friston's programme is not wrong. It is a precise and powerful mathematical framework for a class of phenomena whose neural implementation exhibits inference-like structure under the formalism's chosen level of description. What it is not — and what the present argument is — is a metaphysical account of what governance at a cognitive stratum is. FEP describes what the brain does under one interpretation of its dynamics. Compositional Immanence specifies what the dynamics have to be for governance to be installed at all, and the specification does not require that the dynamics be about anything beyond the coupling topology that sustains them. Both accounts can coexist as readings of the same tissue. Only one of them is the account this book will use, and at every subsequent stratum the non-representational commitment made here will be what keeps the framework coherent under load. Chapter 10 will test it first, at the affective closure where the representationalist temptation is sharpest.
§9.5 — The Scar-Tissue Inventory
The architecture of §§9.2–9.3 is real by the criterion the framework has used since Chapter 8: it breaks in specific ways that are diagnostically distinguishable from one another, each way corresponding to the loss of a specifically nameable operation, and the full set of characteristic failures maps the architecture's structure without remainder. The present section catalogues these failures, and catalogues them in the order that exposes the architecture's internal composition. The catalogue is not exhaustive; it is diagnostic. Its purpose is not to list every pathology the clinical literature has named but to demonstrate that the closures this chapter installed are real capacities whose real breakages follow the structural contours of the capacities themselves.
The first family of failures corresponds to the substrate that sustains the coordinate frame at its metabolic cost. Hippocampal and entorhinal tissue consume oxygen and glucose at rates measurable by standard imaging; deprivation of either, by ischaemic stroke in the territory of the posterior cerebral artery, by transient global amnesia's still-unresolved vascular aetiology, by the focal metabolic compromise that characterises certain forms of herpes simplex encephalitis, produces the syndromes of place-blindness and disorientation whose acute phase is indistinguishable from §9.0's patient in its lived phenomenology and distinguishable from it in its cellular mechanism. The tissue can be anatomically intact and functionally silent under metabolic deprivation; it can be structurally compromised and functionally preserved if the lesion respects the circuit's redundancy; and the space between these two failure paths is where the chapter's diagnostic vocabulary does its work. The burn-rate currency is CMRO₂ and the signature of cessation is the specific disorganisation §9.0 described: a scene that arrives correctly and does not compose into place.
The second family corresponds to the distributed publication of the attractor. Bilateral hippocampal damage — the clinical syndrome whose canonical case is H.M. and whose rarer variants include the transient ischaemic episodes that have been followed with longitudinal imaging — produces a disorder in which moment-to-moment navigation is preserved and the assembly of moments into a stable layout is not. The patient can walk from the bed to the bathroom because the retrosplenial and parietal translators are intact and the body-centred affordances are available; the patient cannot, returning from the bathroom, reconstruct the trajectory as a path through a known environment, because the ensemble whose job was to publish the pattern across independent supports has been denied the cellular mass that constitutes the publication. The failure is not of compression — the animal or patient can recognise the destination when it arrives — but of the distributed substrate on which compression could operate. The Witness has been reduced below the threshold its own coherence requires.
The third family corresponds to the compression of the publication into re-usable invariants. Targeted disruption of sharp-wave ripples during post-task rest, in the laboratory preparations whose technique is now standard, leaves the rat's moment-to-moment navigation through the maze intact and degrades, specifically and measurably, the subsequent consolidation of the traversed trajectory into a re-usable compression of the space. The rat runs the maze today and runs it again tomorrow as though it were a new maze; the ensemble that fired during traversal is still the ensemble that fires during traversal; the operation by which the ensemble's firing becomes a compressed invariant, available to later planning without the behavioural cost of re-traversal, is what has been disrupted. The Canon has been decoupled from the Witness, and the coupling's failure is the failure of memory's usefulness.
The fourth family corresponds to the translator between the allocentric frame and the egocentric body. Retrosplenial lesions produce the disorder in which the layout of the environment is known — the patient can draw the floor plan of the house, name the rooms, identify the photograph of the front door — and the layout cannot be walked; the translator that converts an allocentric goal into an egocentrically commanded body has been denied. The syndrome's phenomenology is instructive: the patient does not feel disoriented in the vestibular sense of §9.0, because the vestibular integration is intact and the body's relation to gravity is unbroken; the patient feels that the environment, though perfectly known, is not accessible through walking, because the operator that would have converted its known structure into a sequence of commanded steps is no longer performing that conversion.
A sixth family, sitting between the fourth and the fifth and distinguishable from both, deserves naming because its clinical profile is specific and its mechanism is instructive. Unilateral spatial neglect — the syndrome that follows, most commonly, right-parietal stroke — presents as a failure not of the coordinate frame as a whole but of the frame's publication across one half of the field. The patient copies the right half of the clock and leaves the left half blank; eats the food from the right half of the plate and leaves the left half untouched; describes the right half of a remembered plaza and omits the left. The sensory machinery on the neglected side is intact — the eye sees, the ear hears, the skin feels — and the parts of the coordinate architecture that would compress information from that side are intact also. What has been lost is the publication to those compressors: the spatial structure of the neglected hemifield is no longer being distributed to the downstream architecture that would read it, and the downstream architecture operates on what it receives, which is now a spatially truncated input. The syndrome is neither a sensory deficit nor a memory deficit nor a motor deficit. It is a selective failure of distributed publication, asymmetric by lesion location, and its specificity — the precise hemifield involvement, the density-gradient of severity with distance from midline, the dissociation from primary sensory processing — is what identifies it as a Witness-side failure at partial coverage rather than as any of the more familiar categories into which clinical taxonomy might have placed it. Within the architecture §9.2 installed, neglect is the diagnostic signature of an asymmetric Witness whose redundancy has been halved and whose Canon is now compressing what the remaining publication supplies.
And the fifth family corresponds to the coupling that reconciles the two frames across self-generated motion in real time. This is §9.0's family. The vestibular apparatus can be destroyed while the hippocampal and cortical architectures remain intact; the patient possesses head-direction cells whose firing is correct, place cells whose fields are preserved, retrosplenial tissue whose translation is undisrupted, and nonetheless cannot stand up in the morning, because the third signal — the one without which vision's horizon and proprioception's axis cannot compose into a single gravitationally referenced scene — has been denied its source. The failure is not of any single component of §9.2's architecture; it is of the continuous reafference-cancellation machinery whose job was to integrate the components in real time, and whose cessation collapses the integration without breaking any of the parts. This is the diagnostically most subtle failure in the inventory, because each channel the examining neurologist tests in isolation reports correctly, and the failure is visible only when the composition that the channels were supposed to sustain is asked to do its work.
These five families are not independent. Their boundaries are set by the architecture's internal seams, and most clinically encountered patients present with mixed presentations whose specific profile traces the lesion's respect for or violation of these seams. The catalogue is not a taxonomy of discrete diseases. It is a map of the scar-tissue patterns that the architecture's real operations leave behind when they fail, and the map's function is to demonstrate that the architecture has the operations this chapter has claimed it has — because only a real capacity can break in these specific ways.
One observation, before the contemporary case. The inventory contains no failure mode corresponding to a representation having the wrong content. No entry in the catalogue takes the form "the patient's internal model of space is systematically in error about the external spatial world." The closest candidate — the case of confabulated layouts in certain confabulating amnesic patients — is properly located not at this stratum but at Stratum 4, where simulation without Witness-verified records produces confident-but-unverified narrative. At the present stratum, the failures are failures of coupling, publication, compression, translation, and integration — operations on the architecture's own dynamics, not on its representations of an external world. This absence is the attractor ontology of §9.4 doing its diagnostic work in the inventory. The failure modes CI predicts are the failures CI's ontology permits; the failures an inferential ontology would predict — posterior-miscomputation failures, prior-misspecification failures, likelihood-misweighting failures — are either not observed at this stratum or, where observed, are cleanly reclassified as failures of the architecture's dynamics rather than of its inferences.2
***
§9.6 — The Laboratory That Fits on a Face
The contemporary pathology that reads against this inventory with the tightest empirical fit is not a disease. It is a consumer technology, deployed at scale across millions of users, whose characteristic failure signature is a reafference conflict of sufficient specificity and reproducibility to function as the chapter's laboratory anchor in the way Chapter 8's Levin-laboratory findings anchored the bioelectric stratum. Virtual-reality head-mounted displays present the visual system with a continuously rendered scene whose geometry is updated in response to the user's head movements. The updating runs through a pipeline — inertial measurement, motion-to-photon latency, rendering, display refresh — whose total delay can be driven, with available hardware, below twenty milliseconds, and whose degradation, under network load or rendering complexity or inertial drift, can push the latency past the threshold at which the visual consequence of a head turn arrives measurably after the vestibular signal that the head has turned. When this happens the user does not typically report seeing a delay. The user reports, with remarkable consistency across populations, the symptom cluster that the clinical literature has named cybersickness: nausea, pallor, cold sweat, disorientation, postural instability, and in severe cases the complete subjective collapse of the scene's stability. The pipeline has produced, under controlled and scalable conditions, the live signature of §9.0's patient.1
The empirical case for reading cybersickness as reafference-integration failure rather than as some generic consequence of unfamiliar perception is strong enough to carry the diagnostic work this section requires. Three lines of evidence converge. First, the symptom profile is dose-responsive in latency: the greater the motion-to-photon delay, the more rapid and severe the onset, with the dose–response curve tracking the latency axis in a way that the alternative explanations (novelty of the stimulus, visual conflict with known physics, excessive field of view) do not predict. Second, the adaptation trajectory — the reduction in symptom severity across repeated exposures — follows the temporal structure characteristic of vestibular recalibration rather than of habituation to a novel stimulus; users who adapt to a high-latency system exhibit measurable after-effects in the real world consistent with a recalibration of the reafference-cancellation machinery rather than with a perceptual habituation. Third, individual differences in susceptibility correlate with measures of vestibular function, interoceptive sensitivity, and the specific postural-correction strategies users deploy during motion; the high-susceptibility profile matches, with a specificity the alternative accounts do not achieve, the profile that would be predicted if the symptom were the subjective signature of a reafference-integration apparatus failing to complete its real-time integration.
The three lines of evidence deserve one further disambiguation, because the alternative readings of cybersickness have been persistent in the literature and the framework's diagnostic commitments require that the reafference reading be distinguished from them on empirical rather than stipulative grounds. The "sensory conflict" reading — that cybersickness arises from a discrepancy between visual and vestibular signals that the brain misreads as a toxin-induced hallucination and reflexively emetically reverses — predicts that the symptom should track the magnitude of the conflict rather than the timing of the visual return, and should not be reducible by latency optimisation alone. It is. Below a threshold of roughly twenty milliseconds motion-to-photon latency on contemporary headsets, cybersickness rates fall toward the population baseline regardless of the kinematic content of the scene, and above the threshold they rise with latency more steeply than with content-driven conflict. The "postural instability" reading — that cybersickness is the subjective signature of the postural control system's failure to find a stable balance point under ambiguous visual flow — predicts susceptibility profiles that correlate with baseline postural sway and with the configuration of the support surface. They correlate, but the correlations are weaker than the latency correlation, and individual subjects exhibit latency-specific symptom onsets at stance conditions their own postural control handles perfectly in the absence of the VR system. The reafference-integration reading accommodates both of these observations as secondary consequences of the primary latency-driven failure and predicts the dose–response curve that the alternative readings, by themselves, do not. Cybersickness is the specific subjective signature of a reafference-integration apparatus denied the temporal envelope it requires, and the framework's diagnostic commitments are satisfied when the reading is the tightest fit to the data rather than merely the most internally consistent.
The clinical corroboration from vestibular-loss cases deserves one paragraph of additional density, because the first-person literature here has a specificity the general clinical vocabulary tends to obscure. The subjective reports collected across the bilateral-vestibulopathy cohorts describe, with striking consistency across linguistic and cultural background, a phenomenology that matches §9.0's patient almost exactly: the scene arrives correctly and refuses to compose; darkness does not restore stability but abolishes it; the gravitational reference that ordinary walking depends on feels like a missing agreement among signals rather than like any single absent sensation. The specificity of the phenomenology is what makes the literature's clinical value exceed its numerical modesty; bilateral vestibulopathy is rare, but each case reports the failure in terms whose convergence implicates a single architectural component whose loss has the reported signature. What the VR literature gives in reproducibility and scale, the clinical literature gives in durability and depth, and the two together map the fifth family of §9.5's inventory — now conjoined with the sixth family of Expansion F — with enough redundancy that the closure claim is not exposed to a single empirical contingency.
Within the inventory of §9.5, cybersickness maps to the fifth family with clinical precision. None of §9.2's component architectures has been anatomically compromised. Head-direction stability is intact. Place and grid cells continue to fire as they did before the headset was put on. The retrosplenial translator is undisrupted. What has been disrupted is the temporal coherence of the reafference integration — the coupling that reconciles the frames across self-generated motion in real time, and whose ten-millisecond latencies are the performance envelope the architecture was built to. Push the visual return past that envelope by any significant margin, and the integration collapses; the user experiences, for as long as the condition persists, a synthetic version of the phenomenology §9.0 described as the enduring consequence of peripheral vestibular destruction. The scene is visible and correct and will not hold together as a scene. The postural-correction cost rises immediately — the user sways more, adjusts stance more frequently, breathes faster — and the rise is measurable, in the laboratory, as a direct increase in the metabolic price of maintaining standing balance under a sensory stream whose components no longer compose.
The thermodynamic price of correction under this condition is the diagnostic receipt that VR cybersickness is not a mere sensory curiosity but the signature, at the individual-user scale, of the same architecture whose canonical closure §9.2 installed. The experimental preparations that have measured the cost — surface electromyography on postural stabilisers, indirect calorimetry on metabolic rate during VR exposure, heart-rate-variability measures on autonomic load — show a systematic elevation whose magnitude tracks latency. What the patient of §9.0 pays for the rest of her life, the VR user pays for the duration of the session, and the payment takes the same currency: additional metabolic expenditure to reconstruct, through conscious correction and sustained attention, what the integration machinery ordinarily delivered for free. The scale is smaller; the structure is identical. The closure that §9.2 installed is not a laboratory abstraction; it is a live variable in consumer electronics, and it prices its own operation in grams of glucose.
Vestibular-loss cases function here as clinical corroboration, and the corroboration is strong because the mechanism is unambiguous and the phenomenology is recorded in a literature — bilateral vestibulopathy studies, gentamicin-ototoxicity follow-ups, the clinical descriptions collected in the now-classic JC and Ian Waterman–adjacent deafferentation literature — whose first-person reports converge on the double-footed ground §9.0 narrated. The convergence is the empirical anchor. The VR literature gives the instrumented, reproducible, dose-responsive signature; the vestibular-loss literature gives the phenomenological depth and the durable clinical profile; and the two literatures read against each other map the fifth family of §9.5's inventory with enough redundancy that the closure claim is no longer exposed to a single empirical contingency.
A note on the structural significance of the VR case, earned here rather than deferred. The virtual-reality headset is, structurally, a partial prosthesis for one component of §9.2's architecture: the visual-flow signal that the reafference integration reads as evidence about head and body motion through the environment. It is a cheap prosthesis, a decoupled prosthesis, a prosthesis whose input is generated by a computational process that is not itself the environment and has no intrinsic coupling to gravitational or bodily reality. The prosthesis works — when its latency is below the integration envelope, when its rendering respects the physics the architecture expects, when the user's other channels are cooperating — because the integration machinery is permissive about the source of the visual-flow signal and strict about its timing. When the prosthesis fails, it fails in the specific way the architecture's fifth family fails. This observation is a structural IOU; its cashing belongs to Chapter 13, where digital-substrate prosthetics are the primary subject and the question is whether a prosthesis installed at a higher stratum — at symbolic and simulative rather than at sensorimotor closure — can be evaluated by the same diagnostic criteria. The present chapter owes only the observation, and the observation is now on the record.
***
§9.7 — The Ledger Cannot Price What Has Not Been Tried
The architecture installed in §9.2 and fused with valence in §9.3 solves the embodied-present coordination problem to a degree the prior bioelectric closure could not approach. The organism occupies an allocentric coordinate frame, prices the trajectories through it in real time by an interoceptive ledger whose currency is allostatic cost, and selects among the available paths by a machinery fast enough and cheap enough to scale to the coordination demands of a vertebrate ecology. What the architecture cannot do — what its success makes structurally unavoidable as the next crisis — is evaluate trajectories it has never traversed. The ledger prices by experience. The path that leads to food is priced highly because the organism has been down it and the return has been registered; the path that leads to a predator's ambush is priced low if the organism survived the encounter and priced at infinity if it did not. The pricing works perfectly, and it works too late, because a pricing that requires traversal cannot tell the organism in advance that a trajectory it is about to take for the first time will be the last thing it does.
This is the lethal-cost crisis. It has the same structural shape as the reafference crisis Chapter 8 handed forward to the present chapter, and it is generated by the same mechanism: the prior closure's success is precisely what makes the next limitation decisive. An organism that could not navigate had no trajectories to price and therefore no pricing crisis; an organism that navigates and prices its trajectories by experience is now in a position where some trajectories cannot be afforded even once, and the architecture that priced by traversal has no operator for evaluating paths before commitment. Trial-and-error learning — which is what valence-by-experience is — is an adequate policy-evaluation mechanism only in domains where errors are cheap and opportunities are abundant. In the niches that produced the vertebrate radiations and eventually the mammalian ones, neither condition holds. Predation under ambush, resource procurement under seasonal scarcity, intraspecific conflict under reputational stakes — each of these is a domain in which the first mistake erases the learner and all its accumulated cost functions, and the organism that can only price by experience is the organism that does not get to make the mistake twice.
The embodied-present closure cannot discharge this crisis. The apparatus §9.3 installed runs in real time, on the organism's current trajectory, priced by the current interoceptive ledger, and its temporal horizon is the forward model's lookahead — milliseconds to a few seconds, depending on the regime. It cannot evaluate a trajectory that has not yet been initiated, because its operators are tied to the sensorimotor stream it is currently embedded in; and it cannot evaluate a trajectory that terminates in the organism's destruction, because the terminating event is the event that would have priced it. Two limitations, one structural demand: the organism must acquire the capacity to evaluate counterfactual trajectories — paths not currently being traversed, paths that may never be traversed — by running them through some apparatus other than the sensorimotor one, and pricing them by some ledger that can import valence from the embodied stream without requiring the embodied stream's actual commitment. The apparatus is the offline-simulation architecture Chapter 11 will install; the decoupling from the embodied stream that it requires is the cognitive staircase's most consequential structural seam.2
Two structural IOUs are handed forward at this chapter's close. The first names the debt Chapter 10 will discharge: the embodied ledger §9.3 priced trajectories in real time but did not install the felt weight of those prices as phenomenal character, which requires the structural-number dissolution the next chapter conducts. What the ledger does at Stratum 2's fusion has been described. What the ledger is when the fusion extends into the affective closure proper is Chapter 10's business. The second IOU is Chapter 11's. The lethal-cost crisis this section has named requires the installation of an apparatus whose control variables run over model-space rather than over the sensorimotor stream — an offline simulation architecture coupled to the embodied-present closure through a decoupling gate whose installation is the next fold of the staircase. These two handoffs are the chapter's debt outstanding. The self-reflexive beat that closes the chapter will not discharge them; it will audit the chapter's own operation at the stratum it has just installed, and it will ask, in the format Chapter 8's §8.10 established and the Revision Plan specifies for the Ch. 9 beat, what happens when the navigation architecture this chapter theorised becomes the substrate on which the reader's own allocentric mapping is performed.2
§9.8 — The Map That Is Not the Territory Is Also Not Free
The argument of this chapter is itself an act of allocentric mapping. It has asked the reader to occupy a coordinate frame — the architecture of place cells, grid cells, head-direction cells, hippocampal replay, reafference cancellation, interoceptive pricing — and to move through it as a structured region whose features remain in fixed relations to one another across the reader's traversal. The reader has been invited, as this chapter was written, to navigate a space that is not a space, using an apparatus that the chapter itself has just theorised as the mechanism by which navigation becomes possible. The recursion is not decorative. It is the condition under which the chapter's claims have any purchase at all, because a reader who has not built the allocentric frame the argument describes has not followed the argument and is reading a sequence of sentences that do not compose into an argument, in the same structural sense in which §9.0's patient was reading a room that did not compose into a place.
The substrate on which the reader performs this mapping is not the hippocampal–entorhinal circuit the chapter described. It is, in the first instance, a set of pages or a screen — an external symbolic artefact whose letters are stable across the reader's movements through them and whose compositional regularities the reader's own apparatus is doing the integration work on. The book is a prosthesis for an allocentric mapping operation whose native substrate — the cognitive architecture §9.2 theorised — is performing the integration across the symbolic surface the book provides. This is not metaphor. The phenomenal-self-model, in Metzinger's sense, is being asked to hold a map of a map: the reader's own navigational apparatus is tracking, across the page, the theoretical navigational apparatus the page describes, and the stability of the second depends on the stability of the first. The prosthesis is cheap — the printed word's burn-rate is trivial against the cortical cost of the reading that processes it — and its cheapness is what makes the delegation structurally possible at all.
The book is therefore not a primary cognitive tool in the sense §9.2's architecture is primary. It is a Stratum 6 artefact being read by a Stratum 2 apparatus in the service of the reader's own higher-stratum integration — an arrangement whose thermodynamic economy this chapter's framework is positioned to audit. Three predictions follow, and they follow with enough structural specificity to be testable against the book's own reception.
First: the chapter's diagnostic precision will survive passage through summarisation by another reader — human or machine — to the extent that the summariser reconstructs the allocentric frame rather than merely re-transmitting the egocentric surface. A summary that lists the chapter's empirical anchors without reconstructing the architecture that composed them into a single governance structure will have preserved the book's Witness and decoupled it from its Canon, in the precise sense §9.5's third family named. The prediction is that such summaries will read, to readers who possess the frame, as structurally incoherent in exactly the way the replay-disrupted rat's maze-running reads as structurally incoherent to the experimenter: each moment correct, the composition lost. LLM summaries of the chapter will, on this prediction, exhibit the signature of Stratum 4 confabulation at the substrate level, whose full mechanisation Chapter 13 will take up.1
Second: the cost of the chapter's mapping will be unequally paid across readers in a way that tracks the prior installation of the framework's apparatus from Chapters 5 through 8. A reader who has built the Stabilisation Engine's vocabulary from Chapter 6 and the Stratification Engine's deployment from Chapter 7 will integrate the present chapter at a metabolic cost comparable to the integration costs of the previous chapters; a reader who has not will pay the full reconstruction cost for every inherited term — Witness, Canon, Gluing, Mediation, burn-rate currency — and will pay it at exactly the points where this chapter relied on the prior installation to carry the structural work. The prediction is that reports of this chapter as "too compressed" or "assuming too much" will correlate precisely with the skipping of the Engines, and reports of it as appropriately dense will correlate with their completion. The book's internal thermodynamics is visible in the distribution of its reading costs.
Third, and most structurally consequential: the more effectively the book installs the framework in a reader's apparatus, the less the book itself is required for the framework's subsequent operation. A reader in whom the allocentric frame has been built can diagnose failure modes in novel cases — in clinical presentations the book does not describe, in technological architectures the book does not name, in institutional pathologies Chapter 12 has not yet arrived to mechanise — without the book's continued consultation. The book's success condition is its own obsolescence in the reader whose apparatus it has successfully reconfigured. This is the opposite of the success condition for which most symbolic artefacts are optimised in the contemporary attention economy, whose Replicators are tuned to sustained consultation rather than to installed capacity, and the divergence between these two success conditions is the structural signature of a book that intends normative installation rather than engagement maximisation. Whether this chapter — or the five that flank it — meets its intended success condition is not a question the book can answer from inside itself. The question can only be answered by the downstream presence or absence, in readers, of a capacity the book claims to have installed. The audit, if it occurs, will be performed by the framework this chapter has just theorised, operating on the symbolic artefact the book has just become.
The chapter closes here, on its own debt outstanding. Chapter 10 takes up what it has handed forward.