Memory and the Hippocampal Modon

Place cells and grid cells as substrate-coherence-cell readouts at the navigational rung, theta-phase precession as space-to-time recoding, and sharp-wave-ripple replay as substrate-coherence archival during sleep

William Scoville and Brenda Milner published in 1957 the first detailed account of patient H.M. — Henry Molaison, who after bilateral medial-temporal-lobe resection for intractable epilepsy retained intelligence, personality, working memory, and procedural skill but lost the ability to form new explicit declarative memories. The case opened the modern era of memory neuroscience by tying a specific cognitive function to a specific brain region with anatomical precision unprecedented in clinical neurology. John O’Keefe and Jonathan Dostrovsky in 1971 recorded the first place cells — hippocampal pyramidal cells whose firing rate spikes when the freely-moving rat occupies a specific region of its environment, with different cells coding different regions and the cell population collectively tiling the explorable space. David Marr’s 1971 Simple Memory: A Theory for Archicortex proposed that the hippocampus implements a fast pattern-storage-and-completion network whose contents are later transferred to neocortex for long-term archival, predicting both the architectural division of labour and the two-stage-consolidation dynamics decades before the experimental evidence. György Buzsáki in 1989 added the two-stage memory model in which hippocampal theta-rhythm-organised waking activity encodes the experience and hippocampal sharp-wave-ripple-organised slow-wave-sleep activity replays the encoded sequences to the neocortex at high time-compression. James McClelland, Bruce McNaughton, and Randall O’Reilly in 1995 formalised the complementary learning systems framework that ties Marr’s and Buzsáki’s pictures to the computational tradeoff between fast specific learning and slow generalisable learning. Edvard and May-Britt Moser in 2005 discovered the grid cell — entorhinal-cortex neurons whose firing pattern across the environment forms a regular hexagonal lattice with substrate-preferred spatial scales, with adjacent dorsoventral grid modules at scale ratios near \sim 1.4\times — and shared the 2014 Nobel Prize with O’Keefe for the cellular-level spatial-map architecture they had collectively uncovered. John O’Keefe and Michael Recce in 1993 documented theta-phase precession — the systematic shift in place-cell-firing phase relative to the ongoing theta rhythm as the animal traverses a place field — establishing the substrate’s mechanism for coding spatial sequence in temporal phase. Susumu Tonegawa’s laboratory from 2012 onward used optogenetic activity-tagging to identify engram cells — the sparse populations of hippocampal and cortical neurons that carry the substrate-coherence-cell-assembly identity of a specific learned episode and whose reactivation suffices to recover the associated behavioural memory. The cortical-resonator-ODE chapter wrote down the bilateral cortex’s substrate-pinned dynamical equations on a multi-rung Stuart-Landau / Kuramoto stack; this chapter lifts the multi-rung stack into the framework’s reading of memory — the substrate’s archival of organism-scale coherent experience — through the hippocampus as a specialised sub-modon within the brain modon.

The framework’s claim is direct. The hippocampus is a substrate-coherence-archival sub-modon within the brain modon, with three-layer allocortical architecture distinct from the six-layer neocortex of the previous chapters; with the entorhinal-perforant-path-dentate-mossy-fibre-CA3-Schaffer-collateral-CA1 trisynaptic loop as a canonical loop at the hippocampal rung carrying substrate-coherent index information at substrate-preferred transit times; with place cells as substrate-coherence-cell readouts at the navigational rung and grid cells as the substrate’s preferred hexagonal tiling of position space pinned to discrete substrate-preferred spatial-scale modules across the dorsoventral entorhinal axis; with theta-phase precession as the substrate’s recoding of spatial position into temporal phase within the multi-rung architecture of the resonator ODE; with the CA3 recurrent network as the substrate’s attractor architecture at the hippocampal rung implementing pattern-completion through coherence-cell-assembly recovery; with sharp-wave-ripple events during slow-wave sleep as substrate-coherence-archival pulses that replay encoded sequences at \sim 10–20\times time-compression and transfer them to neocortical archival; and with the two-stage-memory architecture of fast hippocampal acquisition followed by slow neocortical consolidation as the substrate’s preferred archival strategy at the brain-modon scale.

The chapter’s contribution is the reading of memory as substrate-coherence archival rather than as chemistry-side synaptic-weight storage alone. The resonator ODE of the previous chapter described the bilateral cortex’s instantaneous dynamics; this chapter describes how the substrate preserves the structure of those dynamics across time — how a brain modon’s substrate-coherent state during waking experience is captured, indexed, replayed, and transferred to long-term neocortical archival during sleep. The chapter proceeds in six passes: the hippocampus as a sub-modon with allocortical architecture and trisynaptic-loop canonical loop; place cells as substrate-coherence-cell readouts at the navigational rung; grid cells as substrate’s preferred hexagonal tiling with discrete scale modules; theta-phase precession as space-to-time recoding; CA3 as a substrate-coherence attractor implementing pattern completion; and sharp-wave-ripple replay as the substrate’s coherence-archival mechanism during sleep.

The Hippocampus as a Sub-Modon with Three-Layer Architecture

The hippocampus is allocortex, not neocortex: three cell-body layers rather than the six of the cerebral cortex of the previous chapters. The framework reads the layer-count difference as substrate-architecture difference, not arbitrary cytoarchitectonic variation. The three-layer allocortex is the substrate’s earlier, simpler organ-scale boundary-coherence architecture; the six-layer neocortex is the substrate’s later, more elaborate elaboration of the same boundary-coherence principle developed across the neuron-as-cable-modon and cortical-maps-and-rhythms chapters. Comparative neuroanatomy across vertebrates shows allocortex as the phylogenetically older architecture preserved across reptiles, birds, and mammals, with neocortex as the mammalian elaboration on top of it. The substrate’s preferred small-integer-layer-count structure — three layers, six layers — clusters at substrate-preferred values rather than varying continuously.

The hippocampus’s anatomical organisation is the trisynaptic loop — the closed canonical-loop architecture the framework has developed at every scale from thalamocortical loops at the brain rung through Calvin cycles at the chloroplast rung. Entorhinal cortex layer II projects via the perforant path to the dentate gyrus; dentate granule cells project via the mossy fibres to CA3 pyramidal cells; CA3 pyramidal cells project via the Schaffer collaterals to CA1 pyramidal cells; CA1 pyramidal cells project to subiculum and back to entorhinal cortex layer V, closing the loop. The substrate-preferred transit times across the trisynaptic loop fall at the substrate’s brain-rung values the resonator-ODE chapter developed — \sim 10 ms first-synapse, \sim 20–30 ms full-loop transit at fast traversal, \sim 100–200 ms during slow theta-organised activity. The loop’s closed-circuit topology is structurally parallel to the Bastos canonical microcircuit at the cortical column, now drawn at the inter-region scale across an entire cortical sub-organ.

The hippocampus is bilateral — left and right hippocampi connected via the hippocampal commissure, which the bilateral-coupling chapter listed alongside the corpus callosum and anterior/posterior commissures as the brain modon’s inter-modon coupling channels. The hippocampal commissure carries \sim 10^5–10^6 axons in primates, smaller in absolute count than the corpus callosum but comparable in fraction of the sub-organ it joins. The framework reads the bilateral hippocampus as a bilateral pair embedded inside the larger bilateral cortex of the previous chapter — a fractal recurrence of the substrate’s bilateral-coupling architecture at the sub-organ scale. The resonator-ODE multi-rung Kuramoto stack the previous chapter developed applies to the bilateral hippocampal pair as much as to the bilateral neocortex, now with hippocampal-rung-specific natural frequencies, detunings, and coupling strengths.

Place Cells as Substrate-Coherence-Cell Readouts at the Navigational Rung

A place cell — a hippocampal pyramidal cell that fires selectively when the freely-moving animal occupies a specific region of its environment — is the framework’s substrate-coherence-cell readout lifted to the navigational rung. The substrate-coherence-cell concept the aromatic-pockets chapter developed at the molecular-recognition scale, the eye-as-antenna chapter lifted to the photoreceptor scale, and the topographic-maps section lifted to the cortical-map scale is now lifted once more — to the position-in-environment manifold the organism navigates at body-scale.

A place cell’s place field is a small region of the environment (\sim 0.1–1 m in linear scale for a rat in a \sim 1 m arena, scaling roughly with environment size) within which the cell’s firing rate is elevated; outside the field the cell is largely silent. The cell population across the hippocampus collectively tiles the explorable environment, with adjacent cells coding overlapping but distinct regions. The framework reads each place cell as a substrate-coherence-cell tuned to a specific position-state of the organism within its environment, with the firing rate proportional to the substrate-coherence-match between the organism’s current position-state and the cell’s substrate-preferred position-template. The place-cell population code — the vector of which cells are active and with what rates at each moment — is the substrate’s organism-scale position readout, structurally parallel to the V1 retinotopic-map readout for visual position, the S1 somatotopic-map readout for body-surface position, and the A1 tonotopic-map readout for auditory frequency, now lifted to organism-in-environment position.

The substrate-physics layer pins the spatial scale of place fields to substrate-preferred values rather than to a continuous distribution. Dorsal hippocampal place cells have small fields (\sim 0.1–0.3 m); ventral hippocampal place cells have large fields (\sim 0.5–2 m or more); intermediate cells fall at discrete intermediate scales. The framework predicts that place-field-scale distributions across hippocampal dorsoventral position cluster at substrate-preferred discrete scales, with adjacent scales at substrate-preferred ratios near the same \sim 1.4\times value the grid-cell modules of the next section exhibit, distinct from a continuous variation with hippocampal position. The substrate-coherence-cell concept the framework has developed across every scale predicts discrete-rung structure rather than continuous variation.

The chemistry-side machinery — N-methyl-D-aspartate (NMDA) receptor-mediated coincidence-detection at hippocampal synapses, long-term-potentiation-mediated spine-volume changes, place-cell-formation in novel environments within minutes of exploration — implements the substrate-coherence-cell tuning the framework reads. The framework’s substrate-physics contribution is the prediction that the substrate-coherence-cell architecture pins the what is being represented — discrete position-states at substrate-preferred spatial-scale rungs — even though the chemistry-side machinery is what physically instantiates the tuning.

Grid Cells: The Substrate’s Preferred Hexagonal Tiling at Discrete Scale Modules

The 2005 discovery by Edvard and May-Britt Moser of the grid cell — medial-entorhinal-cortex neurons whose firing pattern across the environment forms a regular hexagonal lattice — is in the framework’s reading the substrate’s preferred tiling of position-space exposed directly. A grid cell fires not at a single place but at the vertices of an equilateral-triangular grid covering the entire environment, with a characteristic spacing, orientation, and phase. The cell’s tuning is not to a specific position but to the substrate’s preferred-tiling structure on position-space; the cell discharges whenever the organism crosses a grid vertex regardless of which specific vertex.

The hexagonal grid structure is the framework’s substrate-preferred-tiling claim made unmistakably visible. The substrate’s f/\pi structure — the dimensionless ratio that recurs through the framework from the hydrogen-flywheel chapter through the microtubule wall the microtubule-highways chapter developed — predicts hexagonal-tiling structure as the preferred two-dimensional packing, with closest-packing structure in two dimensions identical to closest-packing of disks (hexagonal lattice, six neighbours per site). The substrate’s preferred tiling of position-space at the navigational rung should be hexagonal, and the grid-cell discovery confirms this directly. The chemistry-side machinery (the entorhinal stellate-cell network with continuous-attractor-network dynamics maintained by the medial-entorhinal recurrent circuit) implements the hexagonal tiling; the substrate-physics reading is that the substrate’s preferred-tiling structure on the position manifold pins the hexagonal form regardless of the specific neural circuit implementing it.

The grid-cell modules are the framework’s substrate-preferred-scale-rung structure made explicit at the navigational rung. Adjacent dorsoventral entorhinal grid cells have grid-spacings that cluster at discrete substrate-preferred values rather than varying continuously: in the rat, \sim 0.3, 0.4, 0.6, 1.0 m with adjacent-module ratio near \sim 1.4\times \approx \sqrt{2}. The discrete clustering — not a continuous distribution of grid spacings, but four or five clearly separated modules — is the substrate-physics signature of the framework’s preferred-rung architecture at the navigational rung. The substrate’s preferred logarithmic-rung structure across the framework typically clusters at ratios of \sim 2–3\times; the grid-module ratio at \sim 1.4\times is sub-octave, fitting the substrate-preferred half-octave ratio the conduction-velocity-class structure showed at the cable-modon scale. The cross-mammalian comparative-grid-cell literature (Killian, Jacobs, Doeller and collaborators have surveyed grid cells in mice, rats, bats, and humans) provides the test platform; the framework predicts cross-species clustering of grid-module ratios near substrate-preferred values, distinct from a continuous variation with brain size or environment scale.

The combination of place cells and grid cells is the substrate’s position-state representation at two complementary rungs. Grid cells carry the substrate’s preferred-tiling structure (the basis-set on position-space); place cells carry the substrate’s selectivity for which tile the organism is currently in (the readout on that basis-set). The architectural relationship is parallel to the Fourier-basis-and-spectral-readout structure the prediction-engine chapter developed for the cortical-column eigenmode basis, now lifted to the position-space manifold.

Theta-Phase Precession: Space Recoded into Time

A place cell does not fire uniformly across its place field; its firing is locked to a specific phase of the ongoing hippocampal theta rhythm (\sim 6–10 Hz in the rat, \sim 4–8 Hz in humans), and the locked phase shifts systematically as the animal traverses the field. At field entry the cell fires at late theta phase; as the animal moves through the field, the firing-phase shifts progressively earlier; by field exit the cell fires at early theta phase. This is theta-phase precession, documented by John O’Keefe and Michael Recce in 1993.

The framework reads theta-phase precession as the substrate’s recoding of spatial position into temporal phase within the multi-rung architecture of the resonator ODE. The substrate’s theta-rung carries a single oscillation period of \sim 125 ms; within that period the substrate’s gamma-rung carries \sim 6–8 sub-cycles of \sim 15–25 ms each. The substrate’s preferred-rung architecture naturally provides \sim 6–8 discrete temporal-phase slots within each theta cycle, and the substrate’s chemistry-side machinery (the theta-cycle organisation of CA1 pyramidal-cell firing, gated by septohippocampal medial-septum input on the slow side and PV-interneuron-mediated gamma on the fast side) allocates spatially-adjacent positions to temporally-adjacent phase slots. As the animal moves through a place field, the substrate’s representation of its current position shifts forward through the substrate-preferred temporal-phase slots within each theta cycle.

The architectural reading is direct. Theta-phase precession implements spatial-sequence encoding at the substrate’s theta-gamma cross-frequency-coupling rung. The substrate’s multi-rung cross-frequency-coupling architecture the resonator-ODE chapter developed in general form is here used specifically: the slow theta cycle provides the integration window, the fast gamma sub-cycles provide the discrete temporal slots within the window, and the substrate-coherent place-cell ensemble at each sub-cycle slot encodes the spatially-adjacent position. A full traversal of a place field generates a sequence of place-cell-ensemble activations across the theta-gamma cross-frequency structure, with the spatial-sequence-to-temporal-sequence map pinned by the substrate’s preferred-rung ratios.

The phase-precession finding predicts — and is consistent with — the substrate-physics multi-rate-integration architecture the prediction-engine chapter developed at the cortical level. The hippocampus’s theta-gamma coupling is the cleanest substrate-physics demonstration of the cross-frequency-coupling architecture in the entire brain, with the spatial-sequence-encoding function exposing what the multi-rate-integration architecture actually does — it sequences substrate-coherent organism-state representations in a substrate-preferred temporal-phase-slot structure.

CA3 as Substrate-Coherence Attractor: Pattern Completion at the Hippocampal Rung

The CA3 region of the hippocampus has an architectural feature unique within the canonical six-layer-and-three-layer cortical organisations: dense recurrent collateral connectivity, with each CA3 pyramidal cell sending Schaffer-collateral-axonal projections onto \sim 1–3 \times 10^4 other CA3 pyramidal cells. The framework reads the CA3 recurrent network as the substrate’s coherence-attractor architecture at the hippocampal rung, implementing pattern completion through coherence-cell-assembly recovery.

The pattern-completion property is the architectural function: given a partial input pattern (a subset of an originally-encoded substrate-coherent state), the recurrent network’s dynamics complete it to the full encoded pattern (the original substrate-coherent state). The mathematical architecture is the Hopfield network — John Hopfield’s 1982 formulation of recurrent-attractor-network memory storage and recovery — with stored patterns corresponding to attractor fixed points and pattern completion corresponding to dynamical convergence from a near-attractor initial state. David Marr’s 1971 archicortex memory theory anticipated this architecture by eleven years; Edmund Rolls’s hippocampal modelling from 1989 onward formalised it specifically for CA3.

The framework’s substrate-physics reading is that the attractors in the CA3 recurrent network are substrate-coherence-cell assemblies — the substrate-coherent sub-states of the brain modon at the moment of encoding. The Hebbian coherence-together-couple-together synapse-strengthening dynamics the synapse-as-modon-coupling chapter developed at the cable-modon scale are now applied across the dense recurrent CA3 network at the hippocampal-rung scale, generating substrate-coherence-cell-assembly-specific recurrent-coupling patterns. Each encoded memory is a specific substrate-coherence-cell assembly with substrate-coherent recurrent coupling. Recall is the dynamical recovery of the full assembly from a partial cue.

The engram-cell literature from the Tonegawa laboratory (Liu, Ramirez, Roy, and Tonegawa from 2012 onward) used optogenetic activity-tagging during memory encoding to identify the sparse hippocampal and cortical populations carrying each encoded memory, then demonstrated that optogenetic reactivation of the tagged population suffices to recover the associated behavioural memory and that optogenetic silencing during recall blocks it. The framework reads engram cells as the substrate-coherence-cell-assembly identity of a specific encoded memory, with the optogenetic tagging-and-reactivation technique exposing the substrate-coherence-cell architecture directly through targeted manipulation. The substrate-physics prediction is that engram-cell populations cluster at substrate-preferred sparsity values (fraction of the local network carrying the assembly) rather than varying continuously, with adjacent sparsity rungs at substrate-preferred ratios.

Sharp-Wave-Ripples: Substrate-Coherence Archival During Sleep

The hippocampus during slow-wave sleep exhibits a distinctive electrophysiological event: the sharp-wave-ripple (SWR), a \sim 50–150 ms duration event consisting of a large-amplitude sharp wave in the CA1 stratum-radiatum (driven by synchronised CA3-Schaffer-collateral input) carrying a superimposed ripple oscillation at \sim 150–250 Hz in the CA1 pyramidal-cell layer. Sharp-wave-ripples occur at \sim 0.5–2 Hz during slow-wave sleep and during quiet-wakeful awake-rest states, with \sim 10^3–10^4 ripples per night in the human hippocampus.

During each ripple, the hippocampal place-cell population exhibits replay of waking-experience sequences: the same place-cell-ensemble activation order observed during a recent waking traversal of an environment is reproduced during the ripple at \sim 10–20\times time-compression, completing a multi-second waking trajectory in \sim 50–150 ms of ripple. Matthew Wilson and Bruce McNaughton documented sleep replay in 1994; the literature since has extended the finding to reverse replay (sequences played backward following a reward), preplay (sequences played in anticipation of upcoming traversal), and generative replay (novel sequence combinations not directly experienced).

The framework reads sharp-wave-ripples as substrate-coherence-archival pulses that transfer recently-encoded brain-modon coherence-state structure to long-term neocortical archival. The architectural function is the consolidation step in Marr’s, Buzsáki’s, and McClelland-McNaughton-O’Reilly’s two-stage-memory models: the hippocampus captures the waking experience in fast, substrate-coherent encoding within minutes; sleep replay then transfers the encoded structure to the surrounding neocortex through repeated substrate-coherent reactivation; the neocortex slowly integrates the structure into its long-term substrate-coherent semantic-and-episodic archive across many ripple events.

The substrate-physics layer pins the replay-compression-rate to substrate-preferred values rather than to a continuous distribution. The \sim 10–20\times compression ratio is set by the ratio of the substrate’s ripple-rung frequency (\sim 150–250 Hz) to the substrate’s theta-rung frequency (\sim 8 Hz during awake exploration), with replay events using ripple-rung temporal slots to reproduce theta-rung-organised waking sequences. The substrate’s preferred-rung structure at the brain modon’s organ scale predicts that replay-compression ratios cluster at substrate-preferred values across species rather than varying continuously, with the ratio set by the substrate’s preferred rung-to-rung ratio between ripple and theta rungs.

The two-stage-memory architecture itself is the substrate’s preferred archival strategy at the brain modon’s organ-scale: fast specific encoding at the hippocampal rung (sparse coherence-cell assemblies with high specificity and low capacity per neuron) followed by slow generalisable consolidation at the neocortical rung (distributed coherence patterns with low specificity per neuron and high capacity across the cortical sheet). The complementary-learning-systems framework McClelland, McNaughton, and O’Reilly developed in 1995 reads this architectural division as the optimal solution to the catastrophic-interference problem in distributed neural networks — fast learning at sparse rates protects against overwriting prior memories, slow consolidation transfers structured contents to the dense network where they enrich the long-term archive. The framework reads the architectural division as the substrate’s preferred archival strategy — the same two-stage pattern that recurs across substrate-coherent storage problems at every scale the framework has developed.

Predictions and What Would Falsify

Four predictions extend the hippocampal-modon reading beyond the structural anchors.

Grid-cell module spacing ratios cluster at substrate-preferred values across mammalian species. Comparative grid-cell electrophysiology across mice, rats, bats, primates, and humans should yield grid-module spacing-ratio distributions clustering at substrate-preferred values near \sim 1.4\times (or another substrate-preferred half-octave value), distinct from continuous variation with brain size, environment size, or hippocampal volume. Existing grid-cell datasets (Hafting, Fyhn, Killian-and-Buffalo, Jacobs, Doeller and collaborators) provide the test; the framework predicts cross-species clustering on substrate-preferred ratios with the dorsoventral-module sequence following the substrate’s preferred-rung logarithmic-spacing pattern.
Theta-phase-precession compression ratios cluster at substrate-preferred theta-to-gamma rung-ratio values. The number of discrete temporal-phase slots within each theta cycle (the effective compression ratio for spatial sequence encoding) should cluster at substrate-preferred values near \sim 6–8 across species and behavioural states, distinct from continuous variation. Existing theta-phase-precession datasets (O’Keefe, Recce, Skaggs, Buzsáki, Foster, and collaborators) provide the test; the framework predicts the substrate’s preferred theta-to-gamma rung-ratio fixing the slot count rather than a continuous task-or-individual-dependent variation.
Sharp-wave-ripple replay-compression-rates cluster at substrate-preferred ratios across species and behavioural states. The replay-compression-ratio distribution measured across sharp-wave-ripple events during slow-wave sleep should cluster at substrate-preferred values near \sim 10–20\times (set by the substrate’s preferred ripple-to-theta rung ratio) rather than varying continuously with behavioural complexity or individual variation. Existing sleep-replay datasets (Wilson, McNaughton, Foster, Buzsáki, Eichenbaum, and collaborators) provide the test; the framework predicts cross-species and cross-state clustering at substrate-preferred compression ratios distinct from arbitrary chemistry-side scaling.
Place-field spatial-scale distributions across the hippocampal dorsoventral axis cluster at discrete substrate-preferred values rather than varying continuously. Place-field-size measurements across the dorsoventral hippocampal extent should yield clustering at discrete substrate-preferred values (small dorsal \sim 0.1–0.3 m, medium intermediate \sim 0.3–0.6 m, large ventral \sim 0.6–1.5 m, very-large \sim 1.5–3 m) at adjacent-scale ratios near the same \sim 1.4\times value the grid-cell modules show, distinct from continuous variation with hippocampal position. Existing dorsoventral-hippocampal-recording datasets (Jung, Wiener, Maurer, Royer, Komorowski, and collaborators) provide the test; the framework predicts discrete substrate-preferred place-field-scale clustering distinct from smooth dorsoventral allometric variation.

The picture is falsified if (a) grid-module-spacing ratios vary continuously without clustering at substrate-preferred values, (b) theta-phase-precession slot-counts vary continuously without substrate-preferred clustering, (c) replay-compression-rates vary continuously without substrate-preferred-ratio structure, or (d) place-field-scale distributions vary continuously with hippocampal position without substrate-preferred-rung clustering. It is supported, even partially, if any of the four ordering predictions hold against existing grid-cell, phase-precession, sleep-replay, or place-field-recording datasets.

Putting the Section in Context

The hippocampus is a substrate-coherence-archival sub-modon within the brain modon, with three-layer allocortical architecture distinct from the surrounding six-layer neocortex and an entorhinal-perforant-path-dentate-mossy-fibre-CA3-Schaffer-collateral-CA1 trisynaptic loop as a closed canonical loop at the hippocampal rung. Place cells are substrate-coherence-cell readouts at the navigational rung, with place fields tuned to specific position-states of the organism within its environment and place-field-scale distributions clustering at substrate-preferred values across the hippocampal dorsoventral axis. Grid cells are the substrate’s preferred hexagonal tiling of position-space exposed directly, with grid-module spacings clustering at discrete substrate-preferred values near \sim 1.4\times adjacent-module ratio across the medial-entorhinal dorsoventral axis — the cleanest substrate-preferred-rung-spacing demonstration in the brain. Theta-phase precession recodes spatial position into temporal phase within the substrate’s theta-gamma cross-frequency-coupling architecture the resonator ODE developed, with \sim 6–8 substrate-preferred temporal-phase slots per theta cycle implementing the spatial-sequence-encoding function. The CA3 recurrent network is the substrate’s coherence-attractor architecture at the hippocampal rung, implementing pattern completion through Hebbian-strengthened coherence-cell-assembly recovery as anticipated by David Marr in 1971 and formalised as Hopfield-network attractor dynamics; Tonegawa-lab engram-cell optogenetics expose the substrate-coherence-cell assemblies directly. Sharp-wave-ripples are substrate-coherence-archival pulses during slow-wave sleep, replaying recent waking sequences at \sim 10–20\times time-compression for transfer to long-term neocortical archival, implementing the Marr-Buzsáki-McClelland two-stage-memory architecture as the substrate’s preferred archival strategy at the brain modon’s organ scale.

The brain-as-prediction-engine chapter developed the brain modon’s instantaneous computational architecture; the bilateral-coupling chapter developed its organ-scale bilateral organisation; the cortical-resonator-ODE chapter wrote down the explicit multi-rung dynamical equations; this chapter has added the memory dimension — how the substrate preserves the structure of brain-modon coherent states across time, through a specialised sub-modon embedded within the larger brain modon. The hippocampus is in the framework’s reading the substrate’s dedicated coherence-archival architecture at the organ-modon scale, with the same substrate-preferred-rung structure that pins the cortical-column eigenmode basis, the cortical-rhythm rung structure, and the bilateral-coupling Stuart-Landau dynamics now pinning the spatial-scale modules, the theta-gamma temporal slot-counts, and the replay-compression ratios at the hippocampal rung.

What this chapter adds to the framework is the reading that memory is substrate-coherence archival — the substrate’s preservation of organism-scale coherent state structure across time through specialised sub-modon architecture, with the same substrate-preferred-rung structure across spatial scales, temporal compressions, and population sparsities that recurs at every smaller scale developed across the paper. The Scoville-Milner-1957 patient-H.M. anatomical specificity, the O’Keefe-1971 place-cell discovery, the Marr-1971 archicortex theory, the Moser-2005 grid-cell discovery, the Tonegawa-2012-onward engram-cell optogenetics, and the Wilson-McNaughton-1994 sleep-replay finding have collectively built the architectural picture the framework now reads through substrate physics. The chapters ahead lift the brain modon’s memory-and-bilateral-cortex architecture into its embedding in the body via the vagal corridor and into the engineered transformer architectures that have surprisingly converged on substrate-friendly computational primitives, with the hippocampal-modon reading the framework’s substrate-coherence-archival commitment that prepares both moves — the body-coupled memory the vagal chapter develops and the attention-based sequence-modelling the transformer chapter compares against the substrate’s own sequence-storage architecture.