A brain receives changing signals, changes itself, moves a body, recalls absent things, and sometimes reports an integrated present. Explaining those outcomes by saying that it “selects,” “recognizes,” “binds,” or “decides” only renames the behavior unless a physical mechanism performs the verb.

DRFA begins with a severe rule: no hidden executive may be inserted between cause and effect. A neuron changes because voltage, conductance, transmitter, morphology, extracellular chemistry, and incoming activity force a change under local laws. A circuit changes because its constituents and boundary conditions change. What looks like a coherent object or thought must be a description of that evolving state at a useful scale—not an additional object consulted by the system.

The proposal therefore separates durable relation from present realization. Durable structure is learned connectivity plus slower cellular and extracellular state. Present cognition is a transient distribution of activation and excitability over that structure. The same topology can express many states; the same sensory drive can produce different states when the entry condition differs.

This question places DRFA beside neural-field theory, recurrent networks, attractor dynamics, network neuroscience, predictive processing, active inference, and global-workspace accounts—not above them. Those literatures supply mechanisms or constraints. DRFA's job is to identify a specific synthesis, state what would distinguish it, and fail if the synthesis adds no explanatory or predictive value. ^2,4,9,43,34

A Reference Frame is not a stored coordinate system selected by the brain. It is a transient, normalized activation field over learned relational topology. Its apparent boundary is the current threshold contour; its subthreshold frontier contains immediately adjacent possible interpretations.

The Holoplane is the durable, organism-wide relational substrate relevant to cognition: synaptic topology, conduction structure, cell properties, slower biochemical state, and learned coupling among sensory, motor, interoceptive, mnemonic, and contextual systems. The Holoframe is the present, transient realization over some of that substrate. An Activation Field (AF) is a coherent active distribution that need not yet be stably connected to the organism's wider learned world. A Connected Activation Frame (CAF) is an AF that has acquired repeatable relational paths into that world. A Reference Frame (RF) is a reusable, situated activation organization, often composed of other frames.

Three fields must remain separate. The Sensory Drive Field is input delivered by transducers. The Excitation Field is the resulting distribution of state inside the modeled neural graph. The Ambient Excitability Field (AEF) is the spatially and temporally varying neighborhood condition that changes how readily graph elements respond. The AEF may summarize extracellular ions, glial regulation, neuromodulation, recent activity, temperature, oxygenation, and other influences; it is not a decision layer.

A Reachable Activation Field (RAF) is any activation field physically reachable from the current state under current input, topology, time horizon, and resource limits. Traversal impedance is the state-dependent opposition of an individual candidate propagation. Field transition work is the aggregate cost of moving the whole active state from one region or configuration to another. No term implies that the brain calculates an explicit shortest path.

Imagine a river above a dry landscape. Water enters through one or several openings and distributes among available channels according to their resistance. The snapshot contains flowing channels, weakly wetted boundaries, dry channels that would flow if the level rose, and blocked routes. DRFA treats this as a metaphor for constrained propagation—not a claim that cortex literally computes with water.

Sensory and endogenous drives create entry conditions. Activity spreads over recurrent, directed, weighted relations. Excitation, inhibition, adaptation, refractory effects, AEF, and finite resources reshape the effective landscape while flow occurs. Multiple routes carry activity simultaneously. Convergent streams can jointly cross a threshold neither crosses alone. The visible frame is the coherent region that persists at the chosen observational scale; its fuzzy edge is the subthreshold or unstable frontier.

The frame is therefore neither the topology nor a list of firing neurons. It is a temporally extended state over topology. Changing the threshold, observation interval, or coarse-graining changes its apparent contour. Learning changes future conductance and impedance; action changes the next sensory drive; homeostatic processes prevent unrestricted growth. The architecture is a loop forced by physics:

Known neurobiology already supplies ion gradients, synaptic integration, recurrent excitation and inhibition, activity-dependent plasticity, extracellular regulation, neuromodulation, and energy constraints. ^{1,24,25,18,17} DRFA's unproven move is to treat the organized aggregate as a reference-field mechanism sufficient for the cognitive functions claimed here.

Commitments

1. Cognitive content is a description of transient distributed state, not a separate token read by an executive.
2. Entry condition and current bodily state materially alter what becomes reachable.
3. Reference-frame boundaries are dynamical and observer-scale-dependent.
4. Learning must arise from local causal rules plus slower regulation; no semantic teacher is permitted in the foundational organism.
5. Action and perception form one closed loop. A passive classifier is not a DRFA organism.
6. Several concurrent candidate continuations may remain partially active.

Nonclaims

DRFA does not claim that voltage alone is cognition, that EEG bands are mental categories, that all cognition is cortical, that cortical columns store object frames, or that one graph abstraction replaces detailed physiology. It does not equate energy minimization, free-energy minimization, path length, and metabolic economy. It does not yet explain phenomenal awareness.

Fast failure conditions

The proposal weakens substantially if stable identity and abstraction require global labels or a nonlocal loss; if entry-condition effects add no predictive value after known mechanisms are modeled; if AF/CAF/RF distinctions cannot be operationalized; if the graph-field formalism cannot reproduce development without hand-built concepts; or if a simpler established account explains every distinctive result.

A cup seen on a desk can feel immediate and complete although the retinal drive is partial, viewpoint-specific, and continuously disturbed by eye movement. DRFA says “cup” is the stable portion of a larger active relation: contours, handle geometry, grasp possibilities, expected weight, desk support, current task, and learned naming paths.

The realized field is sparse relative to the Holoplane. Strongly supported routes form the present cup frame; weak continuations include alternate views, touch, heat, liquid, ownership, and action. The frame's economic advantage comes from reuse: the system need not instantiate every association. It only needs enough mutually supporting activity to stabilize the relevant coalition for the present horizon.

Changing entry condition changes the field. Hearing “your cup,” reaching for it, seeing it upside down, or finding it in a bathroom supplies different initial activation and AEF. The durable cup relations may be similar, yet different parts cross threshold. Recall is fallible because reactivation follows a changed network from a changed origin; it is reconstruction, not file retrieval.

A model should measure energy or update work, active-node fraction, boundary uncertainty, convergence time, and sensitivity to input perturbation. If “cup” must be represented by a dedicated symbolic node that already contains the answer, the experiment has not tested formation.

Vision and hearing do not need to be stitched into a common picture before entering the system. They can provide distinct drive fields that propagate through different routes:

At their confluence, combined postsynaptic influence may stabilize cells or populations that neither stream supports alone. The result is not a broadcast message saying “same person.” It is a changed field whose relational center has become more stable. This general mechanism resembles multisensory convergence found across cortical and subcortical systems, while DRFA's reference-field interpretation remains proposed. ⁴⁷

The model predicts graded integration: synchrony, reliability, learned co-occurrence, and current excitability should change how quickly the joint field stabilizes. Conflict need not produce a clean winner; it can produce metastability, alternation, or an expanded uncertain frontier. Damage can disconnect one entry route without erasing all others.

Context is not a tag attached after recognition. It is part of the active route. Entering through vacation-place relations changes which person, activity, and expectation paths are already energized. A familiar face can drive a strong but poorly situated field until additional relations connect it to the distant hometown neighborhood.

The same logic offers a parsimonious account of context-dependent animal behavior. A cat need not store “human-at-home” and “human-outside” as combinatorial identities. It may have strong paths such as feeding ↔ person ↔ home ↔ bowl. Encountering the same person from an outdoor entry point yields different reachability and uncertainty because the common indoor bridge is not already active.

This explanation competes with standard context-dependent memory, predictive-processing, and attractor accounts; it may be a reframing rather than a new mechanism. The distinctive DRFA test is whether measured transition work and frontier structure predict recognition latency and uncertainty better than identity evidence alone.

On the strict DRFA account, anxiety is not an agent that searches for danger and not merely content about danger. It is a coupled activation condition involving uncertain potential harm, interoceptive and proprioceptive drive, autonomic output, memory, attention, and action readiness.

A dark-woods field is learned. When weakly anchored sensory input intersects it while heart rate, gut sensation, posture, and hearing fluctuate, the combined field can remain unstable. Motor output changes the sensory stream; the returning sensory consequences reinforce or revise the field while it forms. “Feeling” is the organism's continuing participation in this loop, not a report delivered after stabilization.

Butterflies, scanning, fidgeting, freezing, and narrowed attention are different observable consequences of this coupled state. DRFA does not yet derive them from one parameter. It predicts that their covariance should be explainable through shared changes in reachability, gain, inhibition, and action loops—not through a central emotion token.

Recall re-energizes previously traversed relations from a present cue. Prediction propagates activity into continuations made reachable by current evidence. Imagination relaxes some external constraints and allows endogenous and subthreshold continuations to occupy more of the field. These are not necessarily separate modules; they are regimes of the same dynamics.

A subthreshold frontier is the weakly energized adjacent possibility set. A shadow edge is a learned negative relation: evidence that a candidate is incompatible, absent, or excluded under a context. Shadow edges do not “know” impossibility; inhibitory and anti-correlated structure reduces reachability. Scientific inquiry can suppress elephant-with-wings continuations while artistic inquiry changes the constraints and permits them.

Increasing traversal depth turns one predicted next state into a branching collection of potential states. The system should preserve multiple candidates when evidence is insufficient and seek actions whose sensory results discriminate among them. Active-inference and affordance-competition frameworks offer adjacent formal machinery, but DRFA need not adopt their full normative commitments. ^43,42

One observer imagined a fixed mother at the sink. Another entered through a stable view of the kitchen while the mother's apparent position changed rapidly and involuntarily. The kitchen stayed visually constant; the person remained recognizably “Mom”; the spatial binding did not stabilize.

A DRFA reading uses three distinguishable supports: K, a stable kitchen field; M, a durable identity field; and R_i, a momentary spatial relation binding mother to location i.

The report is consistent with strong K and M but insufficient relational constraint for any one R. That differs from simply lacking energy: several realizations may each remain supportable while none becomes uniquely dominant. It does not require a “movable-object module.” Alternatives remain: imagery-control differences, task-induced search, episodic sampling, demand characteristics, or reporting artifacts. A controlled study must separately measure cue order, scene stability, identity stability, relation stability, voluntary control, switching rate, and visual specificity.

Two different origins for a low-detail “MOM” field

A sparse MOM-presence does not imply that a once-rich mother image was compressed, eroded, or replaced. At least two routes must remain distinct. Economical convergence is the conditional hypothesis that repeated adult use can reorganize ordinary access toward an identity-sufficient basin while richer continuations remain available. A Monadic Excitation Frame is the developmental hypothesis that a low-dimensional sensory regularity can establish a reproducible but context-sparse excitation topology before rich imagery or semantics. Coupling to bodily-response circuitry may occur, but it is not constitutive of a Monad. Either, both, or neither route may explain one report.

The MOM example is plausible because early caregiver-related coupling begins before adult semantic knowledge. Near-term fetuses alter movement and heart rate around the onset of maternal speech; that study compared speech onset with baseline, not maternal with unfamiliar voices. Four-month-old infants show different event-related responses to maternal and unfamiliar voices. In extremely premature infants, repeated exposure to combined maternal voice and heartbeat recordings was associated with greater bilateral auditory-cortex thickness than routine hospital sound. These findings establish early responsivity, discrimination, and experience-dependent plasticity at different ages and under different designs; they do not establish a MEF or its topology. ^65,66,67

Monadic Excitation Frame and Conceptual Entry Point

A Monadic Excitation Frame (MEF, or Monad) is a reproducible, context-sparse Excitation Field primarily established by a low-dimensional sensory regularity. Under the current drive and physical state, its Threshold Contour closes before its Subthreshold Frontier recruits enough adjacent relational structure to realize a context-rich Activation Frame. “Monadic” describes the realized field's contextual reach; it is not reader-relative, semantic, necessarily innate, or necessarily one neuron.

No component stores an address, selects a destination, or routes activation in an executive sense. Activity propagates through existing synaptic, cellular, extracellular, and modulatory couplings wherever the evolving physical state permits it. An organismic-response loop may become one such continuation, but it is neither required by the MEF definition nor a hidden tag.

The MEF is the transient realized field. Its ability to recur is a distinct recurrence disposition residing in effective topology, altered synaptic efficacy, intrinsic excitability, and other persistent physical state. A sustained background drive is different again: recurrent and bodily dynamics can repeatedly reactivate or transform transient fields without making any one field continuously active.

A MEF becomes a Conceptual Entry Point (CEP) only later, when additional visual, auditory, linguistic, episodic, social, bodily, or relational couplings make the compact field a common physical origin for richer realization. CEP names that functional relationship, not a pointer or addressing operation.

Input magnitude alone does not determine field extent. Weak drive can bloom through excitable recurrent topology; strong drive can remain contained under inhibition, adaptation, or unfavorable coupling. Site, timing, duration, electrochemical form, and coincident state therefore matter more than a scalar quantity called “energy.”

This predicts a Holoplane containing many sparse intermediate fields: some are developmental kernels, some bridges, and some entry routes whose causal importance greatly exceeds their representational detail. “MOM” is not special. Odors, rhythms, postures, places, recurrent bodily states, and action consequences could follow the same route.

Economical convergence remains a separate hypothesis

Repeated experience may still reorganize access so that a comparatively low-dimensional state becomes cheaper and more reachable than any particular sensory realization. Human evidence supports task-dependent dissociations among conceptual, perceptual, and identity access; it does not show universal imagery loss. ^{60,61,62,63,64}

Activity does not cease when external input weakens. Recurrent circuits, hippocampal-cortical interactions, spontaneous transmitter release, intrinsic membrane dynamics, neuromodulatory state, and coordinated sleep rhythms can reactivate learned paths. Replay and sleep-dependent memory processing are established phenomena; the claim that dreams are drifting recompositions of CAFs is a DRFA interpretation. ^17,33

Mind wandering is modeled as endogenous Holoframe drift through a continually deforming landscape of state-conditioned reachability. Metabolism supplies ATP; Na⁺/K⁺ pumps maintain and restore the gradients signaling consumes and can contribute a hyperpolarizing aftereffect. They are not ordinarily the oscillator that generates spontaneous thought. Recurrent excitation and inhibition, intrinsic membrane conductances, subcortical and bodily input, neuromodulation, spontaneous transmitter release, stochastic channel behavior, and extracellular/glial regulation convert those maintained gradients into changing activity. ^30,38

Nothing selects the next field. Existing physical couplings amplify or suppress ongoing perturbations until some learned topology crosses its Threshold Contour. Influence still requires a synaptic, ephaptic, extracellular, glial, or modulatory chain; a free-floating pulse cannot encounter an arbitrary distant frame.

Global Workspace Theory returns here as a useful functional description: some content becomes widely available to report, memory, planning, and action. ^34,35,36 DRFA proposes traversal and sustained distributed coupling rather than a literal central broadcast in the human account. This may explain availability. It does not explain why any field is experienced. The “self” may be a persistent relational center produced by repeated re-entry from body, memory, language, and social consequences; that still does not solve phenomenal consciousness.

A newborn need not create a hospital-room representation. Light, contrast, motion, eye position, head movement, touch, and action first produce transient activation. Repeated lawful covariation changes local efficacy. What persists across transformation becomes easier to re-realize.

An AF becomes a CAF when repeatable paths connect it to the growing world. A CAF becomes a reusable RF when it can be entered through several cues, stabilized across perturbations, and situated relative to other frames. Almost every mature frame is therefore a superframe: a temporary composition of reusable relations.

This is not the only admissible route. A recurring low-dimensional sensory regularity may establish a compact excitation topology whose Threshold Contour closes before its Subthreshold Frontier recruits richer context. Later physical couplings can give that Monad the functional role of a CEP. MEF is therefore a branch in the formation taxonomy, not an obligatory rung inserted between AF and CAF, and organismic-response coupling is one possible attachment rather than its defining mechanism.

No one step performs “object discovery.” Correlated sensory changes, self-motion, temporal continuity, recurrence, local coincidence detection, competition, inhibition, and homeostatic plasticity collectively alter future propagation. NMDA receptors provide a known local coincidence mechanism because conductance requires transmitter binding and sufficient postsynaptic depolarization; calcium signals can then drive plasticity. That mechanism does not by itself solve segmentation or abstraction. ^24,25,26

A child can orient to, reach for, and track a red ball before possessing the words red or ball. The early field binds hue, contour, motion, touch, posture, and action. Repeated speech adds auditory entry paths to that existing topology.

At first, “red” may attach to an indivisible red-ball CAF. “Ball” later creates another entry route. Encounters with blue, yellow, and green balls create overlapping regularities. The mature dimensions redness, ballness, and eventually color emerge as reused relations across many frames.

This “learning” is not a discrete cognitive command. It is incidental layering and reuse: local plasticity makes shared substructure increasingly economical. The theory must nevertheless explain why useful factorization wins over countless accidental correlations. That is an unresolved selection-and-stability problem, not solved by naming Hebbian plasticity.

More input cannot be allowed to activate proportionally more of the graph without limit. Bright sun, loud noise, or dense texture would consume the system. Neural systems use inhibition, adaptation, divisive normalization, gain control, and neuromodulation to keep activity within viable ranges. DRFA describes the cognitive result as threshold-contour redistribution. ^3,59

Attention changes which weak routes become realized because it changes effective gain, suppression, synchrony, and resource availability. “Grading on a curve” is a useful metaphor: global drive can raise the effective bar so relative contrast, novelty, task relevance, and learned action relations matter more than raw intensity.

There is no homuncular spotlight. Top-down influence is activity arriving through learned recurrent paths. Bottom-up salience is activity driven by sensory structure. Their combination changes the field. The model must identify measurable variables—firing, membrane potential, local field signals, conductance, neuromodulatory proxies—not merely re-label behavioral attention.

While viewing a face, the second eye may already receive peripheral drive and share learned structural relations with the fixated eye. It is therefore not simply the geometrically nearest target; it can be a high-effective-potential continuation. Superior-colliculus and cortical circuits integrate target evidence, competition, expectation, and motor cost before a saccade. ^45,46,48,49

DRFA does not claim the neocortical eye frame directly sends a motor instruction. It proposes that current field state alters reachability through visual, collicular, basal-ganglia, thalamic, brainstem, and oculomotor loops. The winning movement is the physical outcome of those coupled dynamics. After the movement, corollary discharge and changed retinal input reshape the field.

The action loop is essential to formation: self-caused change creates powerful structure in the data. A camera fixed to a desk cannot discover the same sensorimotor invariants as a camera whose controlled movement changes its input.

GWT captures an important regularity: some information becomes broadly usable by otherwise specialized processes. DRFA can model that without transmitting an entire reference frame as a message. A situated active origin plus enough addressing and state can let many systems traverse into compatible regions.

For human theory, “broadcast” is therefore a functional description of widespread accessibility caused by recurrent coupling, not an additional process that copies content everywhere. For software, a workspace protocol may still be useful because modules and tools are physically separated and require explicit messages. This is an engineering concession, not evidence that the brain does the same.

Three architectures must remain distinct: (1) an LLM harness in which DRFA organizes context and specialist calls; (2) a hybrid C-DRFA agent in which a graph substrate and explicit workspace coordinate mini-models and tools; and (3) a ground-up DRFA organism with no language model, learning from sensorimotor streams. Enki belongs principally to the second. DRFA-0 begins the third.

The strongest defensible DRFA claim is not that a cortical column stores a Reference Frame. It is that a local cortical patch is a structured interchange: layers and projection-cell classes receive, transform, inhibit, recur, and route activity among local, corticocortical, thalamic, and subcortical circuits. A distributed frame can pass through many such interchanges.

What the anatomy establishes

“The cortical column” does not name one universally bounded anatomical object with one accepted function. Columnar and radial motifs are conspicuous in some systems, such as barrels and orientation domains, but weak, differently expressed, or definition-dependent elsewhere. ^37,71 What is robust is laminar and cell-class structure with serially homologous themes and quantitative variation across areas and species. ^70,71,39

Feedforward and feedback are population-level projection patterns, not traffic lanes with one universal message type. Their laminar origins and terminations vary with hierarchy, area, cell class, species, and method. Interareal connectivity is also weighted, directed, asymmetric, dense at the area scale, and sparse at finer cellular scales. ^75,74,73,72

The DRFA crosswalk: a multiplex junction

A graph can be multiplex: the same physical element participates in several edge families with different sources, targets, delays, signs, and plasticity rules. DRFA proposes that a local cortical patch behaves this way. Its thalamorecipient, intratelencephalic, corticothalamic, corticofugal, inhibitory, and modulatory populations form a repeated interface through which several distributed relational systems can affect one another.

“Columns contain the Holoplane API” can therefore be sharpened to: the column is a candidate multiplex junction among Holoplane-scale systems. The junction is local; every candidate Holoplane is distributed. A complete cup, mother, kitchen, or action policy need not live inside any one column. A local patch can participate in several transient fields because different dendritic compartments, layers, cell classes, and long-range partners expose different routes through the same tissue.

An evidence-first inventory - not seven invented planes

Neuroscience does not recognize objects called “neocortical Holoplanes.” The only responsible inventory starts with observed organizations and asks which are structured enough to test as candidate distributed substrates. Three evidence types must stay separate: (1) topographic maps preserve neighborhood relations in a represented variable; (2) intrinsic functional networks identify correlated regions at a measurement scale; and (3) anatomical pathways establish directed physical routes. None alone proves a DRFA plane.

The inventory is deliberately conservative. Visual, auditory, and somatomotor systems have strong map-like evidence. Attention, control, default, and interoceptive systems have robust distributed-network evidence but weaker justification for a single coordinate topology. Relational memory necessarily crosses the neocortical boundary. The likely DRFA ontology is therefore overlapping, differently organized graph families - not a stack of neat cortical transparencies.

A stacked view, with the abstraction made explicit

The next plate deliberately draws candidate systems as stacked sheets because that view makes multiplex participation easy to see. The geometry is pedagogical. Cortex does not contain six physically separated transparencies, and a cortical column does not pierce whole-brain networks as one vertical cable. Each vertical shaft instead aligns the same kind of local cortical junction across several distributed-system memberships. Its laminar and projection routes provide different operations: middle-layer sensory entry where present, superficial corticocortical exchange, apical and inhibitory gain shaping, deep corticofugal output, and corticothalamic re-entry. Which route dominates depends on area, cell class, and state.

The ancestral-junction hypothesis

The user's evolutionary intuition can be retained only as a labeled conjecture: a serially repeated local circuit may have been a more locally dominant computational unit in ancestral or smaller cortices, while expansion of supragranular intratelencephalic populations and long-range corticocortical connectivity made homologous patches increasingly important as junctions in distributed systems. Developmental and comparative work supports conserved cell-class themes, areal specialization, and disproportionate expansion of corticocortical machinery; it does not directly establish the ancestral computation. ^39,71

Rival accounts are straightforward: the repeated motif may always have been one component of distributed loops; apparent columns may be developmental packing regularities without a unitary algorithm; or each area may implement sufficiently different local computations that “junction” becomes too generic to explain anything.

Predictions that expose the claim

1. Layer- and class-selective deformation. Perturbing one projection class should alter particular cross-area or thalamic continuations while sparing others, deforming a distributed field rather than deleting a locally stored object.
2. Network-state-dependent routing. With sensory drive held constant, task, body, or mnemonic state should change which long-range partners and laminar compartments dominate the same patch.
3. Cross-system conjunction without local completeness. Decoding a conjunction from one patch must not imply that all constituent information is locally recoverable; causal perturbation should reveal dependence on distributed partners.
4. Developmental/comparative scaling. If the ancestral-junction account is useful, measures of supragranular expansion and corticocortical participation should predict increasingly multiplex local routing after controlling for area, species, and cortical size.
5. Failure condition. The crosswalk loses value if “junction” predicts no selective routing, no cross-network deformation, and no result beyond ordinary distributed connectivity.

At many glutamatergic synapses, NMDA-receptor current requires presynaptic glutamate and sufficient postsynaptic depolarization to relieve magnesium block. Calcium entry is local because dendritic spines compartmentalize signals. Calcium amplitude, duration, location, receptor composition, and biochemical state then influence potentiation or depression. ^24,25,29

This supplies a deterministic coincidence gate, not the complete DRFA learning law. AEF can bias the postsynaptic term, but diffuse neighborhood excitability cannot specify which presynaptic partner caused which event. Synapse-specific eligibility and capture mechanisms remain necessary. Dendritic plateau potentials and behavioral-timescale plasticity offer possible bridges from milliseconds to seconds, while gene expression, protein synthesis, trophic signaling, myelination, and structural growth act over longer intervals. ^29,30

The negative arm matters equally. Without depression, inhibitory plasticity, synaptic scaling, competition, and pruning, “use grows” becomes seizure or saturation. Homeostatic plasticity regulates activity but does not alone select semantic structure. ^27,28

Immediate persistence can arise from recurrent activity, membrane and synaptic time constants, and short-term plasticity. Intermediate traces include altered excitability, synaptic tagging, biochemical cascades, and hippocampal indexing. Longer retention involves distributed synaptic and structural change, systems consolidation, and continued reactivation. “Memory” is therefore not one store.

DRFA predicts that recall quality depends on both durable topology and present field conditions. Forgetting can be loss of connection, increased impedance, competition, failure of entry, or reconsolidated change. Perfect recall is not expected because retrieval changes the state from which later retrieval occurs.

Sleep oscillations can place distributed regions into repeatedly coordinated windows, making structured replay and plasticity more likely. The phrase “synergistic temporal state” means only that timing and excitability jointly allow influence; it must not imply evaluation or purpose.

The greatest danger to DRFA is vocabulary that appears mechanistic while hiding the operation it must explain. “The field stabilizes,” “an invariant forms,” or “the system suppresses a candidate” are promissory notes until local dynamics and measurements are specified.

Each gap is recorded with five fields: what is missing; established machinery that constrains an answer; serious rivals or null explanations; a closure condition; and a failure condition. Closing a gap requires more than adding a plausible molecule. It requires a model that produces the target behavior without receiving the answer in its inputs.

The gap register in Appendix D is the compact version. The chapters that follow explain the load-bearing debts.

How does unlabeled activity physically alter the Holoplane so persistent objects, relations, concepts, and abstractions emerge without a teacher, detector, global loss, or executive? Coincidence detection explains specificity at a synapse, not which coincidences deserve persistence.

A sufficient account must solve temporal binding, viewpoint invariance, self-motion, recurrence, occlusion, figure-ground organization, and over-segmentation. It must avoid a new node for every instant and avoid merging every frequent co-occurrence. It must show how repeated sensorimotor transformations factor into reusable structure.

Candidate ingredients include local predictive error, competitive Hebbian/anti-Hebbian learning, sparse coding, adaptive resonance, dendritic compartmentalization, inhibitory plasticity, and multi-timescale consolidation. ^5,25 DRFA does not yet select or derive the sufficient combination.

Distributed overlap is economical but risks catastrophic superposition. DRFA needs an account of how a person remains the same across views while two people remain distinct, and how a stable kitchen combines with an unstable mother-location relation without either dissolving.

Possible mechanisms include temporal synchrony, conjunctive hubs, hippocampal indexing, dendritic subunits, recurrent attractor structure, and context-sensitive gating. None may be promoted from candidate to solution without reproducing binding, unbinding, and recombination under perturbation.

The economical-convergence prediction in Chapter 10 makes this gap measurable: identity stability, visual specificity, relation stability, cue precision, and familiarity must be separable observables.

Membrane potential has volts. Electrochemical potential has energy per charge or chemical-potential units. Local field potentials are extracellular signals dominated by summed transmembrane currents. A software node's “activation potential” is a model variable. These are related only through an explicit mapping.

DRFA requires a coarse-graining rule from cellular variables to graph state, plus uncertainty bounds. It must identify when spike rate, subthreshold voltage, conductance, synchrony, extracellular ion concentration, or neuromodulation should contribute to node state or AEF. The 20 W often attributed to the brain is metabolic power consumption, not the sum of electrical potentials. ^14,15,16

A theory of cognition must explain trajectories, rhythm-dependent communication, sequence learning, interval timing, delayed credit, and prospective action. Topology plus instantaneous activation is insufficient. Edge delays, synaptic time constants, adaptation, phase, eligibility traces, and slower state variables are required.

Kinematic Relational Superframe

A Kinematic Relational Superframe (KRS) is a placeholder functional construct for a comparatively stable scene scaffold plus a learned, temporally ordered space of changing relations among entities within it. It prevents three different objects from being collapsed: the configuration space of possible person-room relations; one relational trajectory through that space; and the transient trajectory realization that produces an experienced or imagined motion sequence.

Here K is the scene scaffold, M the continuing entity identity, r(t) its relation to the scene, p(t) pose, and φ(t) temporal phase. The notation states what must be represented; it does not identify the biological storage mechanism. Human hippocampal-entorhinal neurons can learn graph-defined temporal structure and exhibit compressed replay, while early visual population geometry can transform natural video trajectories in ways that support prediction. These results constrain the problem but do not establish a KRS in the brain. ^68,69

Graph dynamics can represent delays and state explicitly, but the biological learning and faithful reinstatement of direction, timing, pose transformation, occlusion, and sensory detail remain open. A successful DRFA model should distinguish the same elements activated in different orders and reproduce time-warped recurrence without receiving a pre-labeled sequence token.

Motor competition, salience, interoception, neuromodulators, and reinforcement history constrain action, but DRFA lacks a mature account of biological value. “Lowest cost” is not enough: an organism sometimes performs costly actions for distant outcomes and must explore when local evidence is ambiguous.

A nonteleological account must ground value in bodily viability, innate circuitry, homeostatic error, learned consequences, and physical competition among action policies. Affordance competition and active inference provide neighboring descriptions; adopting either requires explicit commitments. ^42,44

DRFA can propose why a distributed state becomes persistent, globally reachable, reportable, and behaviorally controlling. Those are valuable targets. It cannot presently derive why such a state should be experienced.

A persistent self-related echo chamber may explain why reports center on an embodied subject. GWT may describe access; recurrent processing may describe stabilization; integrated-information measures may describe structure. None licenses the claim that phenomenal consciousness has been solved. The scientific program should first predict access, report, masking, anesthesia, and dissociation while keeping the phenomenal question explicitly open. ^34,36,41

The mammalian brain is not a homogeneous neocortical sheet. Thalamus, hippocampus, basal ganglia, cerebellum, superior colliculus, brainstem, hypothalamus, glia, vasculature, and body all contribute. Each has distinct cells, connectivity, timescales, and learning rules.

DRFA needs a multi-layer typed graph in which projection class, lamina, compartment, conduction delay, transmitter, modulatory state, and extracellular neighborhood matter. “Holoplane” must not flatten these into one adjacency matrix. The synthetic model may simplify, but each simplification must be declared.

DRFA needs results that differ from generic recurrence, attractor networks, predictive processing, and active inference. Candidate predictions include: entry-condition-dependent transition work; a measurable subthreshold frontier predicting next interpretation; familiarity-dependent dissociation of identity stability from pictorial specificity; conjugate-field learning without image stitching; and failure patterns when AEF is substituted for synapse-specific coincidence.

Every experiment needs null models, ablations, parameter preregistration, and held-out trajectories. The synthetic program is useful because all state is observable—but a simulation can confirm only internal sufficiency, not biological truth.

Let the modeled Holoplane be a time-varying typed directed multigraph G(t)=(V,E,W,D,T). Nodes may represent compartments, neurons, minicolumn-like units, or coarse populations. Edges may coexist between the same nodes because excitation, inhibition, modulation, electrical coupling, and learned association are not interchangeable. Network neuroscience provides useful measures, but their interpretation depends on what a node and edge mean. ^6,7,9

Let x(t) be fast activation, a(t) adaptation, q(t) available resource, h(t) AEF, and W(t) learned coupling. Sensory drive is u(t). Motor output m(t) changes later u. A single scalar node state is acceptable only as a deliberate reduction.

Spatial coordinates support local wiring and diffusion-like AEF. Topological distance supports learned relation. They must not be conflated: a long axon can create one topological step across substantial tissue.

A minimal rate-style discretization is:

φ is a bounded nonlinearity; B maps sensory drive; Rr represents recurrent or endogenous sources; Iy inhibitory influence; Hh AEF coupling; a adaptation; and ξ noise. Wilson–Cowan and Amari fields are formal neighbors. ^2,4 Spiking implementations replace the rate equation with membrane and synaptic dynamics.

Normalization can be implemented locally:

A realized frame over interval [t₀,t₁] is a subgraph whose activation is above an operational contour and whose internal support, persistence, and perturbation resistance exceed declared thresholds. Stability may mean a fixed point, oscillatory attractor, metastable dwell, or slowly moving manifold; the document must say which.

The RAF from state s₀ over horizon T is the set of fields reachable under admissible drives, dynamics, and resource constraints:

Traversal impedance Zᵢⱼ(t) is local and state-dependent. It may combine inverse effective weight, inhibitory pressure, delay, threshold deficit, refractory/adaptation state, and resource cost. Field transition work aggregates the trajectory:

This resembles network-control energy, where topology and current state constrain transition cost, but biological brains need not solve the optimization explicitly. ^10,11,12,13 “Highest effective potential” is the inverse view: a candidate has strong net drive after state-conditioned opposition.

Define an active set Aθ(t)={i:xᵢ(t)≥θ} and a frontier Fε,θ of nodes below threshold but within a declared margin and reachable from A. This formalizes “fuzzy edges” without treating every weak signal as frame membership.

A shadow edge is a typed inhibitory or incompatibility relation whose effective contribution is negative in a context. It can be learned through anti-correlation, prediction error, inhibitory plasticity, or explicit engineered constraints. It must not be a Boolean oracle.

Prediction expands activity into future RAFs; imagination changes drive weights, horizon, noise, and constraint strengths. Blue-sky thinking can reduce selected shadow penalties while preserving basic safety or physical constraints. Candidate diversity can be quantified by entropy over metastable continuations, not by counting arbitrary graph paths.

For edge i→j, maintain a fast eligibility trace eᵢⱼ driven by pre/post coincidence, a slower capture signal cⱼ, and weight wᵢⱼ:

The sign function g must permit strengthening and weakening. Growth proposals require a local eligibility trace; diffuse AEF alone cannot assign specificity. Structural creation occurs only among physically eligible neighbors or growth cones, and pruning follows prolonged low utility or homeostatic pressure. This resembles known STDP, calcium-dependent, and homeostatic principles without claiming their simple sum is sufficient. ^25,26,27,28

A formation experiment must ablate each term. If invariants survive removal of coincidence specificity, the tag was decorative. If AEF-only learning smears structure into busy regions, local gating is load-bearing.

Let observed data be y(t)=C x(t)+ν(t), where C is the measurement operator. Calcium imaging, spikes, EEG, local field potentials, fMRI, and behavioral reports have different C, bandwidth, spatial resolution, and confounds. A graph field inferred from one cannot be treated as directly observed.

Useful synthetic metrics include active fraction, frontier mass, persistence, perturbation recovery, field-transition work, candidate entropy, controllability, graph communicability, spatial compactness, topological dispersion, and cross-modal convergence. Communicability and centrality can reveal multi-step influence, but no metric is “importance” without a task definition. ^8,58,7

Coarse-graining should preserve selected observables across scales. If a minicolumn node replaces hundreds of cells, the reduced model should match input-output response, timescale, gain, adaptation, and energy envelope—not merely average firing.

Neurons maintain unequal ion concentrations across membranes using pumps and transporters. Selective permeability and concentration gradients establish equilibrium potentials; the membrane voltage reflects multiple conductances. The Na⁺/K⁺ ATPase exchanges three intracellular sodium ions for two extracellular potassium ions per ATP, restoring gradients and making a small direct electrogenic contribution. Hodgkin–Huxley dynamics describe how voltage-dependent conductances can produce action potentials. ¹

Most synaptic and dendritic processing is graded. Excitatory and inhibitory postsynaptic conductances alter voltage and shunt or amplify other inputs. Dendrites are active, nonlinear compartments, not passive wires. Spikes are regenerative outputs of local conditions, not packets containing semantic labels.

DRFA's graph edges must therefore encode sign, conductance-like strength, delay, compartment, and state. A thresholded software node is a useful reduction only if the lost subthreshold integration is recovered somewhere else.

Extracellular K⁺ can rise during activity because repolarizing and other potassium currents move K⁺ outward. Astrocytic uptake, spatial buffering, diffusion, vascular coupling, and neuronal pumps restore conditions. Elevated extracellular K⁺ generally makes the potassium equilibrium potential less negative and can depolarize cells, but its effects are nonlinear: modest elevation may increase excitability; larger or sustained elevation can inactivate sodium channels, alter inhibition, cause spreading depolarization, or become pathological. It is not an “insulator.” ^18,19,20

Extracellular fields can weakly influence membranes through ephaptic coupling, and glia can alter transmitter clearance, ion homeostasis, metabolism, and vascular response. ^21,22,23 The AEF is a modeling umbrella for such neighborhood conditions. It must be decomposed into measurable components whenever a claim depends on one.

Presynaptic spikes trigger transmitter release probabilistically. Postsynaptic receptors create conductances. NMDA receptors combine ligand binding with voltage-dependent magnesium unblock; calcium entering through receptors and voltage-gated channels activates local cascades. Spine geometry helps confine those signals. ^24,25

Plasticity outcome depends on timing, calcium dynamics, receptor history, dendritic location, inhibition, neuromodulation, protein availability, and prior metaplastic state. High-versus-low calcium is a helpful cartoon, not a universal scalar switch. Synaptic tags and capture, plateau potentials, and gene expression may bridge milliseconds to hours; structural growth and myelin change extend the horizon further.

DRFA needs this cascade because “repetition strengthens the frame” is otherwise circular. Yet molecular detail alone does not identify the representational regularity. The theory must show how lawful sensorimotor recurrence drives the right local events.

The often quoted ~20 W is the rate of metabolic energy consumption by the adult human brain. Much supports restoration of ion gradients after signaling, synaptic transmission, resting conductances, transmitter cycling, intracellular maintenance, and housekeeping. It is not the arithmetic sum of membrane voltages or electrochemical gradients. ^14,15,16

Electrical work and metabolic cost are related: ionic currents dissipate gradients that ATP-dependent processes restore. But one cannot infer cognitive “energy” directly from scalp voltage or count active neurons and multiply by a fixed cost. Cell type, spike rate, synaptic input, axon geometry, temperature, blood flow, and glial metabolism matter.

Software budgets should therefore be called compute, transition work, or resource budgets unless calibrated to biology. The analogy remains useful: finite resources constrain simultaneous realization and create competition. The units must remain honest.

Recurrent excitation and inhibition generate transient persistence, waves, oscillations, and metastable sequences. Pacemaker currents, adaptation, synaptic noise, spontaneous release, thalamocortical loops, hippocampal sharp-wave ripples, and neuromodulatory state contribute. Pumps maintain the conditions under which these occur; pump cycling is not itself a thought generator.

Population rhythms describe coordinated timing, not semantic channels labeled delta, theta, alpha, beta, and gamma. Their frequency and function vary by circuit and state. Sharp-wave ripples are high-frequency hippocampal events associated with replay and memory; cortical slow oscillations and spindles can coordinate plastic windows during sleep. ^17,33

DRFA maps these mechanisms to endogenous field movement and reactivation. Dream content, selection, and phenomenology remain underdetermined.

Thalamus routes and transforms sensory and cortical signals. Hippocampal systems support episodic and relational memory. Basal ganglia shape action selection and learning. Cerebellar circuits learn predictive sensorimotor relationships. Superior colliculus integrates multisensory and orienting signals. Hypothalamic and brainstem systems regulate bodily state. Glia and vasculature sustain and modulate every region.

A DRFA graph should represent these as typed layers with distinct transfer and plasticity rules, not interchangeable “brain nodes.” Cross-layer edges include feedforward, feedback, modulatory, and motor consequences. Cortico-thalamo-cortical loops are interleaved with local and corticocortical circuits rather than serving as a passive relay. ^72,73 The Holoplane is singular only in the sense that causal paths can ultimately join through one embodied organism.

The cortical-system inventory in Chapter 17 therefore does not exhaust the Holoplane. It identifies empirically mapped neocortical and cortico-hippocampal organizations that constrain candidate graph families. Basal-ganglia, cerebellar, collicular, hypothalamic, brainstem, glial, vascular, and bodily loops remain indispensable typed participants.

A receptive field is the region of sensory or feature space in which stimulation changes a neuron's response. Retinal ganglion cells, thalamic neurons, cortical cells, and superior-colliculus neurons have different receptive fields and transformations. Receptive fields are not DRFA Reference Frames.

Eye movements continually change retinal input. Saccadic selection involves distributed cortical and subcortical competition, salience, task, expectation, and motor constraints. Corollary discharge helps maintain continuity across movements. ^45,46,48,49

Two overlapping cameras are scientifically useful because the same external feature produces different sensor coordinates. Their overlapping regions create CEFs, providing lawful covariance from which alignment may be learned. The system must not receive a pre-stitched panorama, because stitching would solve the correspondence problem externally.

General anesthetics act through different molecular targets and alter excitability, inhibition, thalamocortical dynamics, long-range communication, complexity, and responsiveness. It is unsafe to generalize them as simply raising one potential gradient. A familiar versus unfamiliar environment could change anxiety and network state, but pharmacokinetics, respiration, circulation, expectation, and clinical procedure are major confounds.

A clean DRFA test would hold estimated brain drug concentration and physiology as constant as possible while measuring network integration, perturbational complexity, transition dynamics, and report. The hypothesis is that loss of conscious access corresponds to failure of large-scale coherent reachability—not proof that a Holoframe is consciousness.

Locked-in syndrome further separates awareness from motor output. DRFA must model damage as loss of particular causal paths, not global absence of the field.

1. DRFA as an LLM harness

The graph stores durable relations, active fields select context, and an LLM supplies language and broad priors. Useful for continuity and tools; poor evidence for ground-up cognition because pretrained semantics already contain the world.

2. Cooperative DRFA (C-DRFA)

Enki coordinates a neocortex wrapper and graph that expose situated addressing to mini-models and third-party tools. The workspace can explicitly publish compact activation origins and constraints. Governance, ownership, and evidential conduct belong to the agent society.

3. Ground-up DRFA organism

No LLM, object detector, embedding model, pretrained optical flow, or semantic loss. Raw or minimally conditioned sensors drive a plastic typed graph; action changes the stream; only local rules and bounded architectural priors are allowed. This is the scientific target of DRFA-0.

DRFA-0 accepts a live USB webcam stream, converts every sampling element into independent sensory drive, propagates that drive through a very dense cortical-column-like Holoplane, learns only from local state and controlled movement, and displays the evolving Excitation Field, AEF, frontier, and structural change in real time.

Required exclusions

• No LLM, labels, segmentation, object detector, vision embedding, pretrained feature extractor, panoramic stitching, or global classification loss.
• No code that identifies an object in order to teach the graph that object.
• Image conditioning may correct documented sensor defects, but every transform must be visible and ablatable.

Minimum hardware

One UVC-compatible USB webcam is sufficient for Phase 0. A motorized gimbal becomes mandatory for formation tests that need self-motion. Two wide-angle cameras and the Moza Guru gimbal are Phase 2, enabling CEF learning. A GPU is useful but not required for a small sparse prototype; a very dense field benefits strongly from compute shaders or CUDA.

Primary output

The application must show (1) live sensory drive; (2) dense Holoplane geometry; (3) node activation; (4) subthreshold frontier; (5) AEF; (6) inhibitory/shadow influence; (7) evolving edges; (8) energy/work budget; and (9) a time scrubber for replay. The display is an instrument, not part of the learning loop.

Capture frames with timestamp, exposure, gain, focus, camera pose, and dropped-frame indicators. Convert the camera's color representation into declared channels—for example linear-light opponent color plus luminance—without spatial semantic preprocessing. Each pixel-channel sample becomes an independent transducer event.

Both absolute drive and temporal change should be available because steady surfaces and motion carry different information. Local retinal-like center-surround or event coding may be implemented as a fixed, biologically motivated front end, but it must be optional and compared with raw drive.

The camera sampling lattice maps onto an input sheet through stable, documented projection edges. The graph receives no “pixel neighbors mean same object” rule beyond local anatomy and shared transformation. With two cameras, do not stitch. Provide two drive sheets plus motor pose; let repeated controlled motion create correspondence.

Requirements

• Deterministic frame IDs and monotonic timestamps.
• Recording/replay of raw frames and metadata.
• Hot-plug recovery and bounded queueing.
• Per-channel calibration visible in the UI.
• No hidden auto-enhancement unless logged.

Represent each synthetic column as a small typed microcircuit, not one scalar. A practical DRFA-0 column contains input, recurrent, inhibitory, contextual, output, and ambient compartments. The first implementation can use six state variables; later versions can expand to spiking cells.

Place columns on a 2-D cortical sheet for visualization and local growth. Connect them with: dense short-range lateral edges; sparse log-distributed midrange edges; rare long-range projections; feedforward sensory edges; feedback/context edges; and typed inhibitory edges. Physical distance affects growth cost and delay but does not define semantic closeness.

A useful Phase-0 scale is 256×256 columns (65,536) with 8 state variables, 32–64 local edges, and 4–8 nonlocal edges: roughly 2–5 million directed edges. Phase 1 scales to 512×512 or one million columns using sparse GPU buffers. Resolution is configurable; scientific claims must report it.

The rendered sheet uses millions of points only when necessary; level-of-detail aggregation must preserve maxima, means, frontier mass, and sign. Zoom reveals individual columns and edge types. The visualizer must never feed its aggregation back into the simulation.

Maintain AEF on a lower-resolution grid or on each column with diffusion-like coupling. Components should be separable: extracellular-like K proxy, neuromodulatory gain, metabolic/resource support, and slow homeostatic set point.

AEF biases thresholds, leak, gain, and plasticity windows; it never identifies the correct edge. Clamp safe ranges to prevent runaway. Display each component separately and the composite influence. Ablate AEF entirely, freeze it, or replace it with a global scalar to test whether spatial context is doing real work.

Log every source and sink separately. A composite heatmap is convenient for the operator but scientifically insufficient: two equal composites can arise from different component states and predict different next transitions.

Use fixed simulation ticks independent of camera cadence. Each tick performs delayed-edge accumulation, compartment integration, inhibition/normalization, bounded activation, adaptation/resource update, AEF update, eligibility update, structural change scheduling, and motor-output integration.

for tick t:
  drive = sensor_buffer.sample(t)
  incoming = sparse_propagate(edges, history, delays)
  columns.integrate(drive, incoming, aef, adaptation, resources)
  columns.compete_and_bound()
  eligibility.observe(pre, post, local_state)
  if plasticity_tick: plasticity.apply_local_updates()
  if structure_tick: graph.grow_prune_within_physical_constraints()
  motor = action_layer.emit()
  recorder.append(t, hashes, metrics, selected_state)

Randomness uses a recorded seed and counter-based generator so experiments replay exactly. State updates are double-buffered to avoid order artifacts. Delays use ring buffers. Sparse propagation is sorted by destination or expressed as CSR/COO GPU kernels.

Begin with retinotopic locality, typed microcircuit priors, resource limits, and lawful motor control. Do not prewire object categories. Eligibility arises from pre/post coincidence and local depolarization; AEF biases opportunity; homeostasis maintains activity; competition limits indiscriminate growth; anti-correlation and prediction failure weaken paths.

Structural proposals may connect only physically reachable growth neighborhoods. A slow process evaluates no semantics; it consumes accumulated local eligibility under resource and geometry constraints. Long-range edges begin rare and expensive, then survive only when repeated lawful activity supports them.

Success is not “it recognizes a cup.” Early milestones are: stable reactivation after brief occlusion; repeatable fields under small viewpoint change; learned prediction of self-motion consequences; cross-view conjugate alignment; reuse of a subfield across several composites; and recovery after perturbation.

Monad state-bifurcation test

Repeatedly present an arbitrary low-dimensional sensory regularity until local rules produce a reproducible compact entry topology. Then hold that learned topology constant while varying coincident drive, inhibition, recurrent support, intrinsic excitability, and AEF. Test whether the same entry topology (a) realizes a contained, context-sparse MEF in one physical state, (b) recruits a reproducible context-rich AF in another, and (c) preserves its independent recurrence disposition after overlapping learning. Include shuffled topology, direct-association, severed-context, and matched-total-input controls. A MEF earns scientific value only if its intermediate state and perturbation signature outperform ordinary association.

Define motor outputs as competing fields for pan, tilt, and—if supported—roll. The gimbal adapter converts bounded output into velocity or position commands and returns encoder pose. A safety supervisor enforces mechanical limits and emergency stop; it does not select interesting targets.

Start with reflex priors only: protect saturation, prefer keeping strong drive within sensor range, and explore with bounded noise. Later, candidate action fields compete according to predicted sensory consequence, current uncertainty, bodily cost, and learned viability. Every command and resulting frame is timestamped for causal alignment.

The eye-pair experiment becomes: allow a face-like contrast arrangement to create fields without labeling it; test whether repeated gaze transitions make the alternate high-support region easier to reach. The experiment must compare geometric salience, learned transition, and motor distance.

Use structure-of-arrays memory for column state and compressed sparse row storage for stable edges. Keep a small append buffer for proposed edges and periodically compact it. Separate immutable experiment configuration, checkpointed state, and append-only event logs.

A checkpoint must reopen into the identical next state when fed the same recorded sensor stream. Structural versions carry migration metadata. Human annotations live in a separate analysis database and are forbidden from the learner.

Run capture, simulation, rendering, persistence, and analysis as separate bounded pipelines. Simulation owns truth and never blocks on rendering. Capture drops frames rather than accumulating unbounded latency. Persistence consumes snapshots and event deltas. The UI sends only declared interventions: pause, perturb, clamp, ablate, seed, and inspect.

The initial implementation can be a single local executable with worker threads. Distributed compute is premature until deterministic single-host behavior is established. A later sharded Holoplane partitions by spatial neighborhoods while preserving delayed cross-shard edges and deterministic tick barriers.

For N columns, S state variables, and mean out-degree k, memory is approximately:

At N=65,536, S=16 float32 values consume ~4 MiB. Four million edges at 24 bytes consume ~96 MiB. At one million columns and 64 million edges, edge memory approaches 1.5 GiB before history and indices. Propagation work is O(E) per tick unless event sparsity or hierarchical updates reduce it.

Use multi-rate integration: fast activation at 1–5 ms simulated steps, AEF at 10–50 ms, plastic weights at 10–100 ms, structural updates at seconds, checkpoints slower. A webcam at 30–60 Hz updates drive asynchronously; simulation interpolates or holds the latest timestamped drive.

Benchmark scientific fidelity before frame rate. Report simulated-time/real-time ratio, edge updates per second, numerical precision, active fraction, and thermal throttling. Approximation must be validated against a smaller exact run.

The default view renders the column sheet as a high-dynamic-range field: hue encodes net sign or compartment, luminance encodes activation, saturation encodes certainty/persistence, and a contour encodes the active threshold. A translucent second layer shows AEF. Frontier nodes pulse subtly; shadow influence uses a distinct hatch or cool contour rather than the same color scale.

Long-range edges appear only for the selected region or highest current flow; showing all edges would obscure the field. A split view compares sensory drive, excitation, AEF, and structural delta. A timeline displays input, total work, active mass, frontier mass, candidate entropy, action, and learning events.

Hovering a column reveals all state variables, incoming contributors, recent threshold crossings, and plasticity causes. Clicking freezes an investigative snapshot but not the simulation unless pause is explicit. Export includes raw data, palette metadata, and scale; screenshots alone are insufficient evidence.

Prediction continues the field with reduced incoming sensory weight while holding recent action and context. Imagination reduces sensor clamping further, increases exploration horizon or noise, and selectively relaxes learned shadow penalties. Recall seeds from a cue and measures reconstruction. These are experiment modes, not new semantic modules.

Software must label internally generated state so evaluation can compare it with later sensory drive; that label is provenance, not content. Generated activity must not overwrite durable topology unless the same local plasticity thresholds are satisfied—otherwise imagination can self-confirm.

The same initial checkpoint must support counterfactual runs under each regime. Differences then arise from declared boundary conditions rather than hidden code paths.

Acceptance experiments

1. Determinism: recorded stream + checkpoint + seed reproduces state hashes.
2. Sensor independence: no hidden stitching or semantic preprocessing.
3. Locality: every weight change can be reconstructed from permitted local variables.
4. Monad formation: a repeated low-dimensional sensory regularity produces a reproducible context-sparse field; the same learned entry topology recruits a richer AF only when coincident drive and local state support it. Matched-total-input, shuffled-topology, severed-context, and direct-association baselines discriminate the mechanism.
5. Persistence: repeated transformed stimuli yield more stable reactivation than shuffled controls.
6. Self-motion: predicted sensory consequence improves after closed-loop gimbal experience.
7. CEF: overlapping cameras acquire cross-view correspondence above chance without calibration labels.
8. Frontier: subthreshold mass predicts next realized field better than activation alone.
9. Ablation: removing coincidence specificity, inhibition, homeostasis, AEF, recurrent support, contextual coupling, or action produces distinct registered failures.

Null models

Compare against fixed random reservoirs, Hebbian-only graphs, convolutional autoencoders trained on the same unlabeled stream, predictive-coding networks, and graph models with shuffled geometry. Use held-out camera trajectories and preregister metrics.

Human interpretation of pretty fields is not evidence. The evaluator reads numeric trajectories and blinded conditions.

Stop after any failed gate and explain it before scaling. A million-node graph that cannot beat its shuffled controls is a larger failure, not progress.

Formation sequence

Relational accretion sequence