The Geometry Life Computes

Cellular inference, energy, spacetime, and the physical architecture of biological fidelity

A ceLLM concept paper by John Coates, Founder of RF Safe

Abstract

Life does not process information in an abstract realm. Every biological distinction is physically instantiated: as a charge separation across a membrane, a calcium pulse at a particular place and time, a redox transition inside a mitochondrion, a mechanical strain in the cytoskeleton, a chromatin contact in the nucleus, or a signal exchanged between neighboring cells. Matter supplies the components, but matter alone does not explain biological intelligence. What determines the possible interaction is geometry; what drives the transition is energy; what gives the transition meaning is timing and context; and what preserves the result is memory embodied in persistent physical state.

This paper develops the Cellular Latent Learning Model, or ceLLM, as a multiscale physical theory of cellular inference. The central proposal is that DNA sequence and three-dimensional chromatin architecture form an evolved generative prior; bioelectric, calcium, redox, metabolic, mechanical, and intercellular signals form a local query; and cellular behavior is the conditional output. A tissue does not require a tiny anatomical blueprint stored in every cell. Each cell repeatedly estimates its local state and selects an action from the biological possibilities made accessible by its inherited and acquired physical architecture. Collective form emerges from the coupling of those local inferences.

Planarian regeneration supplies the clearest conceptual demonstration. A transient perturbation of gap-junctional communication can produce a stable two-headed regenerative state even after the original treatment is gone and the relevant tissue has been repeatedly removed. The conventional language of “morphological memory” correctly names the phenomenon but does not fully specify its execution. ceLLM proposes vector-driven local inference: the remaining tissue carries a distributed physiological state; wound-edge cells sample local vectors within that state; the genome–chromatin system supplies an evolved space of buildable structures; and sequential local actions reconstruct the anatomy consistent with the maintained field.

The paper then connects this biological model to physical spacetime and energy without conflating distinct meanings of geometry. Literal spacetime, electromagnetic field geometry, molecular geometry, biological state-space geometry, and statistical information geometry are related layers, but they are not synonyms. At cellular scales, general-relativistic curvature is not proposed as a morphogenetic mechanism. The relevant claim is both more conservative and more powerful: biological information processing consists of energy-driven state transitions occurring locally in physical spacetime, while the structure of molecular and physiological state spaces constrains which transitions are probable.

From this foundation, high-fidelity biology can be defined as the ability to preserve biologically important distinctions across space and time and to return toward the appropriate attractor after disturbance. Bioelectrical dissonance is a mismatch between an imposed signal and the timing architecture of the living control system. Low-fidelity biology is the resulting state in which decoding, correction, coordination, and recovery become less reliable. It is a meta-disease framework because the upstream loss is shared while the visible outcome depends on tissue density, genotype, developmental timing, buffering, prior history, and persistence.

The theory produces direct tests. If chromatin geometry is part of the cellular prior, then perturbing three-dimensional contacts while preserving sequence should alter how otherwise identical cells decode the same input. If bioelectric vectors condition local inference, then measured changes in those vectors should precede and predict regenerative decisions. If energetic reserve constrains fidelity, then the accuracy, speed, and recovery of cellular decisions should follow measurable energy–accuracy tradeoffs. If chronic environmental timing noise forces adaptive patching, repeated exposures should leave a history in calcium-waveform statistics, redox recovery, chromatin conformation, and future response—even when conventional endpoints return to baseline.

The unifying proposition is simple:

Life is matter organized into geometry, driven by energy, ordered through time, and stabilized by inference. Biological health is the preservation of fidelity across that entire chain.

Executive thesis

The prevailing molecular picture of biology often begins with material objects: genes, proteins, receptors, metabolites, membranes, and organelles. Those objects are real, but a list of objects does not explain a living decision.

A calcium ion has no fixed biological meaning. Its meaning depends on:

• where it appears;
• when it appears;
• how quickly it rises;
• how long it persists;
• whether it oscillates;
• which organelle or microdomain receives it;
• what the membrane voltage was at that moment;
• which transcription factors were available;
• which chromatin regions were accessible;
• how much energy and repair reserve remained;
• and what signals neighboring cells were simultaneously sending.

The same is true of a gene. A linear sequence is not executed in isolation. It exists as a folded, modified, mechanically constrained polymer inside a nucleus whose contacts, compartments, phase-separated domains, transcriptional machinery, and metabolic state change over time. The sequence constrains possibility; geometry regulates access; energy drives transitions; context selects the output.

ceLLM therefore starts from seven linked propositions:

1. Every biological computation is physical. It must be embodied in matter and energy and occur somewhere and sometime.
2. Geometry controls coupling. Spatial adjacency, orientation, topology, compartmentalization, and distance determine what can interact and with what probability.
3. Time carries code. Order, phase, frequency, duration, and recovery intervals change biological meaning even when averages remain the same.
4. Energy pays for distinction. Living systems consume free energy to sense, discriminate, correct, remember, and remain away from equilibrium.
5. The cell is a local inference engine. It never possesses the whole organismal state. It acts from a partial, local view.
6. The genome–chromatin system is an evolved prior, not merely a protein dictionary. It contains sequence-defined possibilities and a dynamically organized physical architecture that changes their accessibility.
7. Multicellular order emerges from coupled local inference. Global anatomy is the stable attractor generated by millions of cells repeatedly updating one another’s local conditions.

This model does not require cells to contain tiny brains. It requires only what the evidence already shows: sensing, state estimation, feedback, memory, energy expenditure, conditional action, and collective error correction.

The decisive question is not, “Where is the organism’s master blueprint?” It is:

What physical state does each cell observe, what prior does it bring to that observation, what action does it choose, and how does that action reshape the state observed by the next cell?

That is the cellular information-inference system.

1. The first principle: information is a physical distinction

Information is often spoken of as though it floats above matter. In biology it never does. A bit of biological information exists only when at least two physical states are distinguishable to a receiver.

Examples include:

• a membrane being more polarized rather than less polarized;
• a calcium pulse arriving now rather than later;
• a transcription factor being nuclear rather than cytoplasmic;
• an enhancer being spatially close enough to contact a promoter;
• a receptor being bound rather than unbound;
• a cell being mechanically stretched rather than relaxed;
• a redox couple occupying one chemical state rather than another;
• a neighboring cell being electrically coupled rather than isolated.

The distinction must also be usable. A voltage difference that no molecular system can detect is not information for that system. A ligand outside the receptor’s affinity range is not an effective signal. A calcium pulse that arrives during a refractory period may carry less information than the same pulse arriving after recovery.

The full biological unit of information is therefore not the signal alone. It is:

signal × receiver × context × timing × available energy

This formulation immediately explains why exposure cannot be characterized by source alone and why genotype, tissue, differentiation state, metabolism, and developmental timing matter. The same external event can be informative to one receiver, invisible to another, therapeutic in one state, disruptive in another, transient when reserve is high, or persistent when recovery is exhausted.

Information theory formalizes distinguishability. Thermodynamics makes that distinguishability physical. Landauer’s principle established that logically irreversible information processing has a thermodynamic cost. In living systems, work on the energetic costs of cellular computation and thermodynamic costs of sensory adaptation shows that learning about the environment, writing new information, erasing old information, and maintaining adaptive state require energy dissipation.

That does not mean “energy is information” in a simple identity. It means:

• information must be embodied in physical states;
• reliable transformation of those states has energetic constraints;
• and preservation of a low-error biological state requires continuous work.

Life is not ordered because entropy stops applying. Life remains ordered because it continually spends energy to build, maintain, compare, correct, and renew distinctions faster than those distinctions decay.

2. Five geometries must be separated before they can be unified

The word geometry can clarify this theory or destroy it. The difference depends on whether distinct layers are kept explicit.

2.1 Literal physical spacetime

Physical spacetime is the arena in which events have positions, durations, separations, and causal order. Every binding event, ion displacement, photon absorption, conformational change, and action potential occurs at a location and time.

At the scale of a cell or organism, the spacetime curvature described by general relativity is extraordinarily small and is not proposed here as a causal mechanism of morphogenesis. DNA does not build an organ by gravitationally curving spacetime in the way a star or planet does.

The relevant lesson from relativity is structural, not gravitational: causation is local, observations are partial, timing is relational, and global behavior can emerge from lawful local interactions without a universal command clock.

2.2 Physical field geometry

Voltage, electric field, ion concentration, calcium activity, redox potential, mechanical strain, and chemical morphogens vary across position and time. They can be represented as scalar fields, vector fields, tensor fields, waves, fronts, domains, and gradients.

For example:

• membrane voltage is distributed over cell and tissue surfaces;
• gap junctions create a changing topology of electrical connectivity;
• calcium waves have direction, velocity, phase, amplitude, and spatial origin;
• mechanical stress has orientation;
• morphogen concentrations define gradients;
• metabolic and oxygen states create tissue microdomains.

This is literal geometry in physical space and time. A cell’s location within that geometry changes the signals available to it.

2.3 Molecular and chromatin geometry

DNA is a physical polymer. It bends, twists, loops, compacts, contacts proteins, associates with nuclear structures, and occupies a changing three-dimensional neighborhood. The genome is partitioned into compartments and interaction domains; enhancers contact some promoters more readily than others; DNA methylation and architectural proteins influence folding and accessibility.

Experiments show that orientation and placement of CTCF sites can reconfigure chromatin loops and alter gene expression. Targeted deletion of persistent CTCF sites can alter long-range domain regulation and activate previously silenced genes within the affected region. DNA methylation can intrinsically alter chromatin flexibility and condensation even in a simplified yeast system, while broad methylation loss has been associated with loss of replication-timing precision and disruption of three-dimensional genome organization in cancer models.

The regulatory effect of three-dimensional architecture is not uniform or reducible to one universal TAD rule. Acute loss of some architectural factors can substantially alter folding while leaving many enhancer–promoter interactions and transcripts temporarily intact. This is not a failure of the geometric view. It reveals a layered, redundant geometry with multiple ways to preserve function.

2.4 Biological state-space geometry

A state space is an abstract geometry whose dimensions are the variables needed to describe a system. For a cell those dimensions might include:

• membrane voltage;
• calcium phase and amplitude;
• ATP/ADP ratio;
• mitochondrial membrane potential;
• redox balance;
• chromatin accessibility;
• transcription-factor occupancy;
• cytoskeletal tension;
• cell-cycle position;
• differentiation state;
• inflammatory state;
• and hundreds or thousands of molecular features.

Each possible cell condition is a point in that high-dimensional space. Development, repair, stress, transformation, and death are trajectories through it. Stable phenotypes behave like attractors: many starting points flow toward the same region. Barriers separate some states; shallow wells are easy to escape; deep wells are robust.

This is where terms such as landscape, attractor, latent space, and geometric prior belong. They describe the organization of possible states and transitions. They do not imply extra physical dimensions floating outside the cell.

2.5 Statistical information geometry

A cell never knows its state perfectly. It operates with probability distributions over possible causes and consequences. Information geometry supplies mathematical tools for describing distances between such distributions.

Two cellular states can be close in ordinary concentration space yet far apart in functional probability if a small change crosses a threshold. Conversely, molecularly different states can be functionally near one another if redundant pathways produce the same output.

This matters experimentally. A fidelity model should not ask only whether the average level of a biomarker changed. It should ask whether the distributions of timing, phase, variability, recovery, and future responsiveness have changed.

The synthesis

These five geometries form a nested causal stack:

physical spacetime contains fields → fields act on molecular geometry → molecular geometry constrains state-space transitions → probability distributions describe the uncertainty of those transitions → energy flow determines how accurately the system can sense, select, correct, and stabilize them

The geometries are connected. They are not interchangeable.

3. Matter supplies the nodes; geometry supplies the weights

The conventional sequence-first view of DNA is indispensable but incomplete. DNA sequence specifies binding motifs, coding regions, regulatory elements, structural tendencies, and many chemical constraints. Yet the same sequence can participate in different outputs depending on folding, methylation, histone state, nuclear location, mechanical context, and the availability of regulatory partners.

The ceLLM formulation is:

DNA sequence defines a deep library of constrained possibilities. Three-dimensional chromatin and cellular architecture assign context-dependent access and coupling probabilities to that library.

This is the basis for calling the genome–chromatin system an evolved geometric prior.

The word prior is precise. Before a cell senses the present environment, evolution and development have already biased which responses are possible and which are probable. A neuron and a hepatocyte possess nearly the same DNA sequence but enter the same present moment with very different chromatin states, protein complements, organelle arrangements, membrane properties, and histories. Their priors differ.

Evolution does not select an abstract string independently of its physical consequences. A conserved sequence can preserve:

• a protein fold;
• an RNA structure;
• a binding motif;
• nucleosome preference;
• polymer stiffness;
• loop anchors;
• spacing between regulatory elements;
• or a dynamical response produced by several of these together.

It would be too strong to claim that every conserved sequence implies a conserved atomic-scale resonant shape. Sequence-to-structure mapping is many-to-many, chromatin is dynamic, and similar function can emerge from different architectures. The stronger scientifically useful statement is this:

Evolution selects reproducible physical consequences of sequence in context, and those consequences include three-dimensional organization and state-dependent coupling—not protein products alone.

Cross-species work supports both conservation and flexibility. Some vertebrate TADs appear as conserved regulatory building blocks, with evolutionary rearrangements enriched at their boundaries and disruption associated with altered gene-expression patterns across species. Other work shows rapid evolution of many TADs, with stronger constraint in developmentally regulated domains in Drosophila. Across eukaryotes, large-scale chromosome architecture can change repeatedly while conserved molecular machinery still generates recognizable organizational classes across the tree of life.

This mixed picture is exactly what an evolved prior should produce: stable deep constraints, flexible implementations, redundancy, and lineage-specific solutions.

Is the cell literally an atomic neural network?

ceLLM uses the neural-network analogy because the mapping is productive:

• molecular structures are nodes;
• physical proximity and interaction probability act like weighted connections;
• chromatin and epigenetic state alter those weights;
• local physiological input conditions the network;
• cellular output depends on nonlinear integration;
• repeated experience can update future response.

But the model does not require DNA to be identical to an artificial neural network, nor does it require every relevant interaction to be quantum coherent. “Atomic neural network” is a hypothesis about distributed physical computation, not permission to ignore chemistry.

The strict scientific claim is that the energized, changing structure is part of the computation. If changing geometry while holding sequence constant changes interpretation of the same input, geometry carries computational weight. That is directly testable.

4. Energy is not merely fuel; it is the price of inference fidelity

Cells are open, driven, nonequilibrium systems. They exchange matter and energy with their surroundings, dissipate heat, and maintain internal organization that would decay without continued work.

Energy performs at least six information-related jobs:

1. Sensing: maintaining receptors, gradients, membrane potentials, and amplification systems.
2. Discrimination: separating similar inputs into reliably different internal states.
3. Timing: resetting channels, pumps, oscillators, and refractory states so the next signal can be distinguished from the last.
4. Correction: proofreading DNA, repairing damage, degrading misfolded proteins, and clearing erroneous states.
5. Memory: stabilizing modified receptors, chromatin marks, synaptic states, organelle composition, and network connectivity.
6. Recovery: returning the system toward baseline after a perturbation.

The connection between energy and accuracy is not metaphorical. In biochemical sensing, greater learning about an external condition requires breaking detailed balance and consuming energy in explicit cellular computation models. Analysis of adaptive feedback networks reveals an energy–speed–accuracy relationship: maintaining accurate adaptation in a noisy environment is dissipative, and performance is constrained by the available energy budget with experimental support in bacterial chemotaxis. More recent work extends such tradeoffs to nonequilibrium receptors and cellular sensing across broader receptor architectures.

This gives low-fidelity biology a physical core. A system can lose fidelity in at least three ways:

• the input becomes noisier;
• the receiver becomes less selective;
• or the energy available for discrimination, correction, and reset becomes insufficient.

The three interact. Noise consumes reserve. Reduced reserve broadens error distributions. Broader errors require more correction. Correction costs energy. If disturbances recur before recovery completes, the system enters an escalating loop.

The geometry of an energy landscape

In a simplified equilibrium system, the probability of occupying a state depends on its energy relative to alternatives. Living systems are not at equilibrium, but the landscape metaphor remains useful when supplemented by driven flux.

• Valleys represent relatively stable states.
• Ridges represent barriers between states.
• External signals tilt the landscape.
• Energy consumption can hold a state uphill from passive equilibrium.
• Feedback can deepen or flatten an attractor.
• Noise can push a system across a barrier.
• Adaptive remodeling can permanently reshape the landscape.

Evolution changes the shape of the landscape over generations. Development selects and deepens particular trajectories. Physiological signaling temporarily tilts it. Learning and adaptation modify it. Injury, aging, toxicant exposure, and persistent environmental noise can distort it.

This is the geometry of energy in the scientifically disciplined sense: not an occult substance, but the topology of accessible states, transition barriers, and driven flows.

5. Time is not an accessory to the signal; time is part of the signal

The central error of many biological measurements is to average away the code.

Calcium signaling demonstrates the problem. Dolmetsch, Xu, and Lewis showed that calcium oscillations increase both the efficiency and specificity of transcription: rapid oscillations activated several transcription factors, whereas slower oscillations selected a narrower response. Li and colleagues independently showed that calcium-spike frequency can optimize gene expression.

The meaningful variables are therefore not only concentration and total exposure. They include:

• frequency;
• phase;
• amplitude;
• rise and decay time;
• spatial origin;
• propagation direction;
• burst structure;
• refractory interval;
• termination;
• and trial-to-trial jitter.

A normal mean can conceal an abnormal waveform. A normal endpoint can conceal a costly recovery. A transient response can leave a changed future response even after the measured quantity returns to baseline.

The cell is a spatiotemporal decoder. Its “present” is not an infinitesimal moment. It is an integration window shaped by receptor kinetics, membrane capacitance, channel gating, organelle dynamics, transcription-factor residence time, chromatin accessibility, and metabolic reserve. Different receivers integrate the same physical field over different windows and therefore extract different information.

This is the proper connection to spacetime: biological meaning depends on the ordered path a system takes through physical events. The cell reads a trajectory, not a snapshot.

6. The cell as a local inference engine

A cell has no direct access to the organism as a whole. It observes only a local boundary condition.

For cell i at time t, define a local sensory state:

sᵢ(t) = {Vmem, ∇Vmem, calcium waveform, redox state, metabolites, oxygen, mechanical strain, extracellular matrix, neighbor signals, temperature, light, and external fields}

The cell also carries an internal physical prior:

Gᵢ(t) = {genome sequence, chromatin topology, epigenetic state, proteome, organelle architecture, receptor complement, and accumulated history}

Given that local observation and prior, the cell chooses actions:

aᵢ(t) = {divide, migrate, differentiate, secrete, contract, repair, remodel, remain quiescent, senesce, or die}

ceLLM proposes the following conceptual mapping:

local observation + evolved prior + available energy → conditional action → altered local environment → next inference

The cell need not consciously calculate probability. Reaction networks, thresholds, feedback loops, physical interactions, and dynamical attractors perform the calculation through their behavior.

One formal template is:

aᵢ = arg minₐ [prediction error + energetic cost + deviation from viable state]*

This is not claimed as the one exact equation solved by every cell. It captures the active-inference logic: the cell can reduce mismatch either by changing its internal estimate or by acting on the world so the sensed state moves toward an expected viable condition.

Work connecting active inference to morphogenesis likewise treats cells as agents that combine bioelectrical, biochemical, and mechanical sensory evidence with internal models of target identity, then act through migration, differentiation, and signaling to approach target morphology.

The local logbook

Every response can alter the next inference. Transcription changes receptor density. Metabolism changes redox state. Injury changes extracellular matrix. Inflammation changes neighboring-cell signals. Epigenetic remodeling changes accessibility. A persistent electrical state changes network coupling.

The cell therefore contains both:

• an inherited deep prior produced by evolution and development;
• and a local logbook produced by experience.

This is how short-term adaptation becomes long-term state. The present is decoded through the accumulated physical record of previous presents.

7. From local inference to collective anatomy

The apparent paradox of morphogenesis is that no individual cell needs to know the full anatomy, yet the tissue can produce a coherent whole.

The paradox disappears when three principles are combined:

1. Local variables contain positional information. Gradients, boundaries, polarities, and neighbor relationships tell a cell where it is relative to other states.
2. Cells share a constrained generative repertoire. Their genome–chromatin architecture contains the capacity to execute tissue-appropriate programs.
3. Every action changes the next cell’s input. Local inference propagates through coupled feedback.

Developmental biology already demonstrates that morphogen gradients carry positional information. Modern quantitative work shows that both signaling level and signaling dynamics can increase the precision with which cells estimate position in the vertebrate neural tube. The ceLLM extension is that bioelectric gradients, calcium timing, metabolic context, and tissue mechanics participate in the same local inference problem.

The organismal form is not stored as a miniature picture. It is generated by an iterative algorithm:

sense local difference → infer local role → act → reshape the field → constrain the next action

When millions of cells perform this loop inside shared boundary conditions, a global anatomical attractor emerges.

This is distributed intelligence in a literal engineering sense. No central controller is required. The global order resides in the coupling rules, the shared priors, the boundary conditions, and the stability of the attractor.

8. The planarian: a decisive test case for vector-driven local inference

 

Figure 1. Conceptual ceLLM visualization. The color gradients represent an inferred positional/bioelectric state, not a claim that this exact continuous blue–yellow–red voltage profile has already been directly measured in every experimental animal.

Planaria force biology to confront the difference between a genetic parts list and a physiological target state.

In a normal planarian, an amputated middle fragment regenerates a head at the anterior wound and a tail at the posterior wound. After transient disruption of long-range gap-junctional communication, some animals regenerate with two heads. More remarkably, the two-headed target morphology can persist through later rounds of cutting in ordinary water, long after the gap-junction blocker has left the tissue. The genomic sequence has not been rewritten in the original experimental paradigm, and the stable inheritance of the altered target morphology has been discussed as information stored in distributed bioelectrical network dynamics rather than sequence alone.

The phenomenon is often described as morphological memory. That description is valid, but it leaves open the implementation question: what does a cut fragment physically possess that allows it to rebuild the same nonstandard body plan?

ceLLM proposes that the answer is not a hidden picture of a two-headed worm. It is a maintained distributed state that changes the local vectors observed at wound edges.

Normal vector logic

In a normal anterior–posterior organization, the two wound edges occupy different locations within the tissue’s physiological gradient. One edge samples a state consistent with missing anterior structure; the other samples a state consistent with missing posterior structure.

The same genome is present at both edges. The difference is the query.

The anterior-facing local state conditions the evolved prior toward a head-building program. The posterior-facing local state conditions it toward a tail-building program. As cells migrate, proliferate, differentiate, and couple to neighbors, their actions extend and refine the gradient. Anatomy emerges step by step.

Two-headed vector logic

In the stable two-headed state, the central fragment retains a distributed physiological organization consistent with anterior identity at both ends. When amputated again, both wound edges sample head-directed local information. The genome already contains the capacity to build a head. The altered field requests that capacity twice.

The tissue appears to “remember” two heads because its current physical state remains organized around that attractor.

What this explanation resolves

It resolves five apparent mysteries:

• No central brain is needed. Wound-edge cells use local information.
• No miniature body map is needed. The maintained gradient supplies relational coordinates.
• No new head gene is needed. The same structural repertoire is conditionally deployed twice.
• No cell needs global knowledge. Sequential actions propagate the solution.
• Memory can survive tissue removal. The information is distributed in network state rather than confined to the excised wound cells.

What remains to be measured

The vector-gradient diagram is a mechanistic proposal, not a substitute for direct mapping. The crucial tests are:

• high-resolution voltage imaging before and after amputation;
• simultaneous calcium and redox imaging;
• gap-junction connectivity maps;
• single-cell transcriptomic and chromatin-conformation profiles along the axis;
• perturbation of local vector direction without globally changing the tissue;
• and prediction of regenerative fate from the measured pre-cut field.

The strongest validation would be prospective: measure the local state first, predict which structure each edge will build, and then observe regeneration without using outcome information to reconstruct the explanation afterward.

9. Foreign heads, shallow attractors, and the difference between memory and relaxation

A second planarian result adds an important layer. In Girardia dorotocephala, transient gap-junction blockade produced regenerated head shapes resembling several other planarian species even though the genome remained that of G. dorotocephala. The changes included head geometry, brain morphology, stem-cell distribution, and membrane-voltage domains. Yet these non-native heads were not permanent. Over subsequent weeks, they remodeled toward the species-typical G. dorotocephala form in the 2015 experiment.

The stable two-headed state and the temporary foreign-head states should not be collapsed into one mechanism. Together they reveal the difference between:

• a network configuration that has become a stable tissue-level attractor;
• and a transient configuration occupying a shallower region of morphospace.

ceLLM interprets the foreign-head result as a competition between the present query and the depth of the inherited prior.

The temporary perturbation changes physiological connectivity and moves regeneration into an alternative region of state space. The tissue can initially construct a coherent foreign-like morphology because the genome–cellular architecture contains enough conserved developmental possibility to do so. But the altered state is not as strongly stabilized within that species’ full network. As ordinary turnover, signaling, mechanical feedback, and metabolic fluctuations continue, the system relaxes toward the deeper native attractor.

The important point is not that the worm “notices a mistake.” No central error inspector is required. A shallow attractor loses stability more readily under ongoing fluctuation; a deeper attractor draws more trajectories toward itself.

This interpretation generates a direct prediction:

A controlled increase in physiologically relevant noise should shorten the lifetime of a shallow induced morphology more than it shortens the lifetime of a deeply canalized native morphology—unless the same noise also destabilizes the native attractor.

That prediction can be tested by comparing waveform-defined electromagnetic, pharmacological, metabolic, and thermal perturbations while recording voltage domains, calcium dynamics, redox state, anatomy, and chromatin structure through the remodeling period.

The result would reveal whether “morphological memory” is best understood as stored symbolic content or as the persistence properties of a distributed dynamical state.

10. The Einstein analogy—and its proper limit

Classical developmental thinking often searches for a central blueprint, master organizer, or privileged clock. When the whole organism behaves coherently without an identifiable central map, the result can sound mysterious.

The historical transition from Newtonian mechanics to relativity offers a useful analogy. Newtonian physics treated space as a fixed stage and time as a universal clock. Relativity replaced that universal staging with a local structure: matter and energy influence spacetime geometry, and objects follow locally defined trajectories within it.

The corresponding ceLLM move is:

Replace the master anatomical blueprint with local inference inside a shared physiological geometry.

Each cell occupies a local position. It has limited causal access. It reads the signals reaching its boundary. It follows transition rules embedded in its material organization. Its action changes neighboring boundary conditions. Global form emerges from the coupled trajectories.

This is analogous to local motion in curved spacetime, but it is not identical to it.

• In general relativity, mass–energy affects the metric of physical spacetime.
• In ceLLM, cellular activity affects the geometry of a biological field and the topology of biological state space.

The former is gravitational geometry. The latter is electrophysiological, chemical, mechanical, molecular, and probabilistic geometry.

The analogy is valuable because both reject the need for an external central controller. It becomes misleading only if “curvature” in biological state space is presented as literal gravitational curvature.

The cell’s causal neighborhood

Relativity also emphasizes that information cannot arrive from everywhere at once. A cell likewise operates inside an effective causal neighborhood constrained by:

• diffusion and transport rates;
• gap-junction connectivity;
• axonal or paracrine signaling;
• field propagation and tissue conductivity;
• receptor kinetics;
• mechanical coupling;
• and the cell’s integration window.

The physically accessible past of the cell is represented in the signals that have reached it and in the state changes those signals left behind. Its possible future is constrained by the actions available from its current state.

That is a biological analogue of a worldline: the cell’s state is not merely where it is, but the history of how it arrived there.

11. The amplituhedron analogy: geometry can compress an enormous calculation

The amplituhedron, introduced by Nima Arkani-Hamed and Jaroslav Trnka, is a positive geometric object from which certain scattering amplitudes in planar supersymmetric gauge theory can be derived. In that formulation, properties that are cumbersome in conventional diagrammatic calculations can emerge from a higher-level geometry rather than being imposed one interaction at a time.

The analogy to ceLLM is provocative but must be exact about its status.

The paper is not claiming that chromatin is an amplituhedron, that cells compute particle-scattering amplitudes, or that biological form proves extra physical dimensions.

The useful principle is narrower:

A sufficiently structured geometry can encode constraints on many possible outcomes so that the result emerges from the geometry as a whole rather than from an explicit sequential instruction for every step.

That is how an evolved geometric prior could compress morphogenetic possibility.

The genome does not need a sentence saying, “At coordinate 1,217 build the left edge of a head.” Instead, conserved molecular capacities, regulatory contacts, tissue coupling rules, and local vectors can make the head-forming trajectory the stable solution to a constrained problem.

In this sense, the anatomy is not explicitly stored pixel by pixel. It is generated from the intersection of present boundary conditions with an evolved positive space of viable forms.

The correspondence is conceptual:

• the amplituhedron compresses allowed scattering structure into geometry;
• the ceLLM prior compresses allowed biological response structure into evolved physical architecture;
• the local physiological state selects a conditional trajectory through that architecture;
• the visible phenotype is the realized output.

The research challenge is to discover whether biological possibility spaces have measurable geometric invariants—features of chromatin contacts, network topology, voltage domains, or attractor structure that remain stable across perturbation and predict the set of attainable outcomes.

12. Geometry is memory when it changes future probability

A physical state becomes memory when it changes the probability of a future response.

This definition includes several familiar forms of biological memory:

• DNA sequence;
• methylation and histone modification;
• chromatin contacts;
• receptor abundance;
• organelle number and location;
• cytoskeletal organization;
• extracellular-matrix composition;
• synaptic strength;
• gap-junction connectivity;
• persistent membrane-voltage domains;
• immune-cell clonal expansion;
• and altered thresholds in signaling networks.

Memory is not a separate substance placed on top of structure. It is the causal persistence of structure.

ceLLM divides cellular memory into three interacting timescales.

Deep evolutionary memory

Sequence and highly conserved structural capacities define what the lineage has repeatedly found viable. This is the deepest prior.

Developmental memory

Cell identity, chromatin accessibility, organelle architecture, receptor complement, and tissue position record the path taken during development.

Adaptive memory

Stress responses, epigenetic changes, metabolic remodeling, inflammatory thresholds, and network rewiring record what the cell has recently needed to survive.

These layers can cooperate or conflict. A short-term adaptation may preserve immediate viability while moving the cell away from the organism’s preferred long-term state. The cell can make a locally rational update that is globally costly.

This is the foundation of adaptive patching.

13. Adaptive patching: when survival rewrites the prior

When the observed environment repeatedly differs from the state a cell expects, the cell can respond in two broad ways:

• restore the environment toward the expected state;
• or alter itself so the new environment becomes easier to survive.

The second response is adaptive patching.

A patch can include:

• increased antioxidant defenses;
• altered ion-channel expression;
• changed receptor sensitivity;
• mitochondrial biogenesis or pruning;
• metabolic rewiring;
• inflammatory priming;
• altered DNA repair priorities;
• methylation changes;
• histone modification;
• and chromatin reorganization.

The patch is not inherently harmful. Without adaptation, life would fail at the first deviation. The danger appears when the patch solves the local problem by degrading global precision or future flexibility.

The sequence is:

disturbance → prediction error → compensatory action → physical state update → altered future inference

Repeated disturbance can produce:

more patches → less reserve → greater dependence on compensation → broader error → more patches

This feedback offers a physical interpretation of progressive low-fidelity biology. The system does not simply accumulate random damage. It accumulates a mixture of damage and survival adaptations. Some are locally useful, some neutral, and some become maladaptive when the environment changes or when neighboring tissues must coordinate with the altered cell.

The “epigenetic defrag” analogy

Partial reprogramming experiments are often described as restoration of lost epigenetic information. ceLLM interprets this as a partial resetting of the physical prior.

The useful analogy is defragmentation: acquired regulatory states can disperse access, shift boundaries, and increase the work required to retrieve a coherent program. Reprogramming factors can reopen inaccessible regions and re-establish some youthful regulatory relationships.

The analogy must not be turned into unsupported atomic mechanics. DNA methylation is not biologically important merely because a methyl group adds mass, and there is no evidence that OSK rejuvenation works primarily by removing “mass dampers” from a resonant DNA lattice. Methylation changes protein binding, chromatin compaction, transcriptional access, replication timing, and regulatory architecture. Those known mechanisms already provide a physical geometry of resetting.

The ceLLM prediction is stronger when stated in measurable terms:

Successful partial reprogramming should restore not only selected methylation markers but also the fidelity with which cells convert controlled bioelectric and metabolic inputs into reproducible calcium, transcriptional, repair, and differentiation outputs.

If youthful markers return without restored input–output fidelity, the model is incomplete. If restored function tracks restoration of chromatin organization, timing precision, and recovery, the geometric-prior model gains support.

14. Low-fidelity biology as a failure of geometric inference

High-fidelity biology preserves the distinctions required for correct local decisions.

It preserves:

• anterior versus posterior;
• self versus danger;
• growth versus repair;
• signal versus noise;
• transient stress versus persistent threat;
• fuel abundance versus fuel scarcity;
• damaged cell versus viable cell;
• and present state versus remembered state.

Low-fidelity biology begins when those distinctions become less reliable.

The failure can enter at four levels.

Input corruption

The local environment carries competing, mistimed, or unfamiliar signals.

Receiver distortion

Genotype, receptor density, differentiation, injury, or prior adaptation changes how the signal is transduced.

Processing limitation

Energy reserve, mitochondrial function, redox balance, or network organization is insufficient to discriminate and correct accurately.

Memory drift

Adaptive patching, chromatin change, persistent voltage state, or tissue remodeling changes the prior through which the next input is decoded.

The combined result is bioelectrical dissonance: the imposed timing structure and the living system’s expected timing architecture no longer align.

The cell can remain alive and active during this state. It may even show apparently successful compensation. The defining loss is not immediate collapse; it is reduced precision, reserve, and recoverability.

A fidelity equation

A useful conceptual relationship is:

biological fidelity ∝ signal distinguishability × receiver selectivity × energetic reserve × network coherence × recovery / noise × unresolved history

This is not a validated clinical equation. It identifies the variables a true fidelity science must measure.

The model’s most important prediction is that averages can remain normal while fidelity falls. The early signal may be increased variance, jitter, state dependence, slower recovery, greater history dependence, or wider divergence between individuals.

15. S4–Mito–Spin as the environmental receiver layer

The S4–Mito–Spin framework connects environmental time structure to the cellular inference system. It is a testable receiver map, not a claim that one pathway has already been proved to explain every electromagnetic effect.

S4: membrane gating and ionic timing

Charged S4 voltage-sensor domains help control voltage-gated ion channels. The proposed S4 branch asks whether externally imposed, polarized, time-varying fields can alter local ion motion or gating probability sufficiently to change the timing of calcium and other ionic signals.

In ceLLM terms, S4 is a candidate input gate. A small timing bias at the membrane can alter the sensory vector supplied to the rest of the cell.

Mito: energetic amplification and recovery

Mitochondria couple calcium, ATP production, membrane potential, reactive oxygen species, apoptosis, and innate immune signaling. They determine whether an input remains a small perturbation or becomes a system-wide energetic and redox event.

In ceLLM terms, mitochondria are edge processors and recovery engines. They help convert local input into action, pay for correction, and record history through changes in organelle state.

Spin: redox-sensitive reaction probability

Radical-pair chemistry provides a demonstrated physical route by which magnetic fields can change selected chemical reaction yields. Experiments have shown RF control of spin-correlated radical-pair dynamics involving genetically encoded proteins and flavin in living animals under resonance-defined conditions. This does not establish that all redox enzymes or environmental RF exposures behave this way; null findings in several flavoenzyme systems confirm the requirement for specific chemistry and kinetics rather than generic radical presence.

In ceLLM terms, Spin is a candidate probability-biasing branch within redox and electron-transfer chemistry.

CYB5B: a bridge from field to calcium code

The 2026 Cell study by Kim and colleagues identified CYB5B as essential to an engineered electromagnetic-field-inducible gene switch. Crucially, activation depended on rhythmic calcium oscillations rather than generic calcium influx and enabled controlled gene expression in vivo.

That result establishes a principle central to this paper:

A structured electromagnetic input can be coupled through defined molecular hardware into a calcium timing code and then into gene expression.

It does not by itself prove that everyday wireless signals engage the same apparatus or produce harm. It makes the correct scientific question unavoidable: under which waveforms, amplitudes, cellular states, and molecular contexts does such coupling occur?

The receiver is individualized

The CACNA1C experiment supplies a human example. In a randomized, double-blind, sham-controlled study, 3.6 GHz 5G exposure shifted sleep-spindle center frequency in one CACNA1C genotype group but not the matched comparison group among 34 genotyped volunteers.

The implication is not that one variant defines electromagnetic sensitivity. It is that the same field can generate a different physiological output when receiver architecture differs.

That is exactly what a cellular inference model predicts.

16. Environmental fields as changes in the cell’s boundary conditions

An electromagnetic field is not abstract information. It is a physical condition in spacetime capable of exerting forces, polarizing matter, inducing currents, or changing the probabilities of particular reactions when an appropriate receiver and coupling pathway exist.

The biologically relevant question is not whether the field “contains a message for the cell.” The cell does not need to understand Wi-Fi as Wi-Fi. It needs only to transduce part of the waveform into a state change.

That distinction prevents anthropomorphic error.

A non-native field can enter the inference loop in several ways:

• shifting a membrane sensor’s gating probability;
• changing calcium-waveform timing;
• altering spin-sensitive reaction yields;
• modifying redox state;
• changing synchronization between cells;
• consuming repair or antioxidant reserve;
• or arriving repeatedly enough to reduce recovery time.

The external signal does not become biologically meaningful because its carrier frequency resembles a native biological rhythm. Modern RF and native ELF fields are not physically identical. The stronger point is:

There can be a carrier gap without a modulation or framing gap.

RF systems carry pulses, envelopes, duty cycles, packet timing, and repetition structures. A nonlinear biological receiver may respond to some aspect of that time structure even when the high-frequency carrier is far above intrinsic biological oscillations.

The field is not the cell’s clock. The cell is the clock. The field becomes relevant only through the way biological hardware samples, filters, rectifies, integrates, and remembers it.

17. Why the same upstream disturbance can become many downstream diseases

The low-fidelity framework is a meta-disease model because the first failure is not disease-specific.

The shared trunk is:

boundary-condition disturbance → altered receiver dynamics → timing and energetic error → degraded local inference → adaptive patching or failed recovery

The branch depends on biological geography.

Development

The critical variable is timing. A short perturbation can intersect a non-repeatable window of proliferation, migration, differentiation, pruning, or circuit formation.

Cancer biology

The critical variables are control escape and persistence. An error must survive DNA repair, checkpoints, apoptosis, tissue-level pattern control, immune surveillance, and turnover.

Autoimmunity

The critical variable is classification. The system must distinguish self from danger and terminate activation after the threat resolves.

Metabolism

The critical variable is coordination. Insulin pulses, mitochondrial demand, hepatic output, substrate transport, and circadian signals must remain synchronized across organs.

These are not four independent RF-disease stories. They are four ways a common inference deficit can become visible when it reaches different receiver densities, developmental windows, genetic backgrounds, energy budgets, and persistence conditions.

The theory predicts heterogeneity. A universal upstream disturbance should not be expected to yield a universal symptom. It should widen outcome distributions and make context more decisive.

18. Aging as the accumulation of inference history

Aging is often called a meta-disease because it raises the probability of many downstream disorders. That classification describes the relationship but does not fully specify the physical process.

ceLLM proposes:

Aging is the progressive loss of the system’s ability to preserve, recover, and correctly apply the distinctions required for high-fidelity local inference.

This loss can arise from:

• mutation and DNA damage;
• epigenetic drift;
• altered chromatin organization;
• mitochondrial dysfunction;
• impaired proteostasis;
• senescent-cell signaling;
• extracellular-matrix change;
• chronic inflammation;
• stem-cell exhaustion;
• loss of network connectivity;
• circadian disruption;
• and persistent environmental demands.

ceLLM does not replace these hallmarks. It asks what they have in common: each changes the physical prior, the local observation, the energy available for correction, or the fidelity of state transition.

The compounding cycle is:

noise or damage → compensation → altered physical prior → less accurate future inference → greater correction cost → reduced reserve → more noise sensitivity

This explains how adaptation can become the engine of decline without treating adaptation itself as an error. A patch is optimal for the current local crisis. The accumulation of incompatible patches becomes a systems problem.

The individual does not fail because every cell runs out of energy simultaneously. Failure emerges when the remaining energy can no longer maintain the distinctions, coordination, and recovery needed for organism-level coherence.

19. A formal ceLLM model

The framework can be stated compactly.

Let each cell i have:

• a local observed state sᵢ(t);
• an internal generative prior Gᵢ(t);
• an available free-energy budget Eᵢ(t);
• a set of possible actions Aᵢ;
• and a local history Hᵢ(t).

The cell’s output is:

aᵢ(t) = Φ[sᵢ(t), Gᵢ(t), Eᵢ(t), Hᵢ(t)]

Its action changes both itself and its neighbors:

sⱼ(t + Δt) = Ψ[sⱼ(t), aᵢ(t), environment]

Repeated or sustained input can update the prior:

Gᵢ(t + Δt) = Gᵢ(t) + η · Pᵢ(t)

where Pᵢ(t) represents the physical patch written through transcription, epigenetic remodeling, organelle change, receptor regulation, or structural reorganization, and η represents the persistence of that update.

Define local fidelity as the ability to preserve correct distinctions and return toward the appropriate attractor:

Fᵢ = f(temporal precision, spatial precision, decoding accuracy, energetic reserve, correction capacity, and recovery)

Tissue fidelity is not the simple average of cell fidelity. Network topology matters. A small number of highly connected or strategically located cells can change the state of many others. Thus:

F_tissue = Ω({Fᵢ}, connectivity, receiver density, boundary conditions, and persistence)

The model predicts low-fidelity biology when disturbance and unresolved history exceed the network’s ability to discriminate and recover:

timing noise × receiver gain × persistence > buffering × repair × recovery interval

These equations are not final biological laws. They define the variables and relationships that an empirical ceLLM program must estimate.

20. Ten decisive experiments

The theory should be judged by what it risks.

Experiment 1: Prospective planarian vector mapping

Map Vmem, calcium, redox state, gap-junction connectivity, and Wnt-associated signaling in intact normal and stable two-headed planaria. Amputate only after the maps are complete. Test whether wound-edge state predicts head-versus-tail outcome prospectively.

Failure condition: regenerative fate cannot be predicted above chance from any measured local or distributed physiological feature.

Experiment 2: Local-vector reversal

Use optogenetic, ion-channel, or spatially restricted pharmacological tools to reverse a wound-edge physiological vector while minimizing changes elsewhere.

Prediction: local anatomy should follow the altered vector more reliably than the fragment’s previous gross anatomy.

Experiment 3: Geometry-without-sequence change

Use targeted CTCF orientation changes, loop engineering, epigenome editing, or inducible chromatin tethering to alter three-dimensional regulatory contacts while preserving coding sequence. Deliver identical calcium, voltage, or metabolic input waveforms.

Prediction: cells with altered geometry should show reproducibly altered input–output mappings.

Experiment 4: Sequence-with-context swap

Place a conserved regulatory motif into different chromatin neighborhoods and cell states. Measure whether functional response follows sequence alone or the sequence–geometry combination.

Prediction: context should reshape gain, threshold, timing, and output even when the motif is unchanged.

Experiment 5: Energy–fidelity curves

Systematically vary ATP reserve, mitochondrial membrane potential, oxygen, substrate supply, and recovery time while delivering a fixed signaling task.

Prediction: timing jitter, decoding error, and recovery should follow measurable tradeoffs rather than an all-or-none energy threshold.

Experiment 6: Adaptive-patch ledger

Apply repeated sublethal disturbances separated by different recovery intervals. Record calcium waveform, redox recovery, chromatin contacts, accessibility, methylation, transcription, and response to a later standardized challenge.

Prediction: equal final baseline values can conceal different histories and different responses to the next challenge.

Experiment 7: CYB5B dependence under waveform variation

Compare CYB5B wild-type, knockout, rescue, and localization mutants across defined field strengths, envelopes, pulse structures, polarization states, and sham conditions.

Prediction: if CYB5B is a relevant transducer in a given context, loss and rescue should remove and restore specific calcium-waveform and transcriptional signatures.

Experiment 8: S4 dependence under waveform variation

Directly record gating currents and calcium microdomains in channels with altered S4 charge, matched controls, and pharmacological blockers.

Prediction: a genuine S4-mediated response should track sensor charge, ion geometry, polarization, and timing in a mechanistically ordered way.

Experiment 9: Attractor-depth assay

Induce native, stable alternative, and transient foreign planarian morphologies. Apply calibrated noise or recovery support and measure escape or reversion rates.

Prediction: shallow states should show different noise sensitivity and relaxation kinetics from deep states.

Experiment 10: Cross-scale state-space reconstruction

Combine voltage imaging, calcium imaging, redox sensors, spatial transcriptomics, single-cell multiomics, Hi-C/Micro-C, metabolism, and morphology. Use blinded models to predict future state from present geometry.

Prediction: an integrated geometric state should predict trajectory better than sequence, average exposure, or any single biomarker alone.

21. What would falsify the framework?

A theory that explains every possible result explains nothing. ceLLM must be vulnerable to failure.

The framework would require major revision if:

• local physiological state does not improve prediction of cellular action beyond known biochemical inputs;
• three-dimensional chromatin manipulation does not alter input interpretation when sequence and expression machinery are controlled;
• bioelectric perturbations change morphology only through ordinary cytotoxicity or nonspecific tissue injury;
• energetic reserve does not influence the speed, accuracy, or recoverability of inference-like cellular tasks;
• proposed receiver knockouts fail to remove the corresponding field response;
• repeated timing perturbations leave no measurable state-dependent history;
• or the integrated multiscale model does not outperform simpler pathway-specific models.

The strongest version of ceLLM is not “everything is connected.” It is:

Specific geometries constrain specific couplings; specific energy flows maintain specific distinctions; and specific local observations produce predictable state transitions through measurable priors.

That statement can fail—and therefore can become science.

22. What the theory does not require

The framework does not require:

• a conscious cell;
• a homunculus inside DNA;
• a literal brain-like neural network in the nucleus;
• a universal organismal master clock;
• gravitational spacetime curvature as a developmental signal;
• hidden extra physical dimensions;
• persistent quantum coherence across the entire cell;
• or one environmental factor as the sole cause of complex disease.

It requires only:

• physical states that can be distinguished;
• receivers that transform those distinctions;
• energy that supports accurate sensing and correction;
• persistent changes that alter future response;
• and coupled local actions that generate collective order.

Quantum mechanics is fundamental to chemistry, and spin-dependent reaction pathways may be important in selected receiver systems. But the distributed-inference architecture remains valid whether most of its large-scale dynamics are described classically or whether particular components require quantum treatment.

This is a strength. The theory does not depend on attaching the word quantum to every unknown.

23. Implications for biology and medicine

Measure trajectories, not snapshots

Single time-point averages discard phase, history, variability, and recovery. Experiments should capture full waveforms and post-perturbation return.

Treat genotype as receiver architecture

Genotype can change channel density, regulatory topology, mitochondrial handling, repair, and buffering. Exposure studies that average across receiver classes can erase real subgroup effects.

Treat recovery as an endpoint

A response that normalizes may still consume reserve or alter the next response. Recovery time, energetic cost, and memory must be measured.

Treat tissue as a network

Cell-autonomous assays miss distributed state, gap-junction coupling, mechanical feedback, immune interaction, and boundary conditions.

Treat geometry as causal

Sequence, expression, and concentration should be measured together with chromatin contact, spatial localization, membrane domains, organelle placement, and tissue gradients.

Design therapy as precision rewriting

Bioelectric and electromagnetic interventions may be therapeutic when they deliver a controlled waveform to a known receiver in a defined state. The therapeutic counterpart to bioelectrical dissonance is bioelectric precision: the deliberate restoration or writing of coherent biological timing.

Design prevention as fidelity preservation

Prevention should reduce avoidable inputs that chronically consume repair, buffering, and recovery reserve. Clean air, water, food, sleep, circadian light, metabolic support, and a biologically compatible electromagnetic environment all belong to the same fidelity program.

24. Implications for environmental electromagnetic research

Conventional exposure science often asks whether a field causes a named endpoint at a given average intensity. The ceLLM framework asks a more upstream set of questions:

• Which part of the waveform reaches which receiver?
• What timing variable does the receiver extract?
• Which cell states amplify or suppress the response?
• How much energetic reserve does correction require?
• Does the system fully recover before the next exposure?
• Does repeated exposure write an adaptive state?
• Does that history alter response to a second, unrelated challenge?
• Which tissue architectures preserve the error?

This reframes non-native electromagnetic fields as potential changes in the boundary conditions of cellular inference.

It also explains why null and positive findings may coexist. A field can produce:

• no coupling in a cell lacking the relevant receiver;
• a transient response with complete recovery;
• a beneficial controlled response in a therapeutic design;
• a disruptive response in a high-gain or low-reserve state;
• or a delayed persistent effect after repeated exposure.

Those are not mutually exclusive conclusions. They are the outcome classes expected from a state-dependent inference system.

The correct exposure unit may therefore be multidimensional:

carrier + envelope + pulse + phase + peak + duty cycle + polarization + chronology + receiver state + recovery opportunity

A heating-only metric compresses that geometry into one number and then mistakes the compression for biological completeness.

25. The deepest synthesis: life as local inference through spacetime

The entire framework can now be stated as one sequence.

Matter

Atoms, molecules, membranes, proteins, nucleic acids, organelles, and tissues supply the physical degrees of freedom.

Geometry

Their positions, orientations, contacts, compartments, gradients, and network topology determine what can interact.

Energy

Free-energy flow drives state transitions, maintains gradients, pays for discrimination, and supports repair and reset.

Time

Order, duration, frequency, phase, and recovery turn state changes into code.

Information

Receivers convert physical differences into distinctions relevant to future action.

Inference

The cell integrates local distinctions through an evolved and acquired physical prior and selects a viable action.

Collective intelligence

Each action changes the field sampled by neighboring cells, producing a distributed convergence toward tissue-level attractors.

Memory

Persistent changes in sequence, chromatin, organelles, connectivity, matrix, and physiological state bias future inference.

Fidelity

Health depends on preserving the accuracy, timing, coordination, and recoverability of the chain.

This is how cellular intelligence connects to spacetime and energy. Not by escaping matter, and not by invoking gravity where chemistry is sufficient. It connects because every biological computation is a local, energy-driven physical history whose geometry constrains the next possible event.

The body’s visible geometry is therefore not the sole memory store. It is the current runtime expression of deeper distributed state.

The cell is not merely executing a static program. It is locally inferring what action is appropriate from a physical worldline of signals and adaptations.

DNA is not merely a dictionary. It is sequence embedded in a dynamically organized physical prior.

Bioelectricity is not a decorative epiphenomenon. It is part of the spatial and temporal query through which cells coordinate collective action.

Mitochondria do not merely supply fuel. They couple energy, timing, redox state, and recovery to the accuracy of the inference.

And the organism is not controlled by a single master blueprint. It is continuously rebuilt by local agents whose shared geometry makes the global goal stable.

Conclusion: protect the geometry that makes life intelligible to itself

The central scientific error of reductionism is not that it studies parts. It is that it sometimes assumes the meaning of a part is contained in the part alone.

Meaning in biology is relational.

A gene means one thing in one chromatin neighborhood and another in a different cell state. A calcium ion means one thing in one rhythm and another in a different phase. A voltage means one thing at one tissue boundary and another inside a different connectivity network. A stress response means one thing when recovery is complete and another when the next demand arrives early.

Geometry defines those relationships. Energy makes their transformation possible. Time gives them order. Memory carries their consequences forward. Inference converts them into action.

The planarian does not need a hidden picture of itself. Its cells inhabit a maintained physiological geometry that makes some local actions more probable than others. The genome provides a deeply trained repertoire. The field supplies the present query. The tissue is the recurrent network. Regeneration is the collective answer.

The same logic scales upward into health. When local signals remain distinguishable, receivers remain selective, energy remains sufficient, and recovery remains complete, cells can repeatedly return the organism toward its viable attractors. When timing noise, receiver distortion, energy limitation, and unresolved adaptive history accumulate, the same system becomes less precise. It enters low-fidelity biology.

The visible disease is the geography of the failure. The upstream loss is the declining integrity of cellular computation.

The task for the next generation of biology is therefore not merely to catalog more molecules. It is to measure the geometries, trajectories, energy budgets, and inference errors that determine how those molecules become living decisions.

Matter gives life substance. Geometry gives it possibility. Energy gives it motion. Time gives it code. Inference gives it direction. Fidelity keeps the whole alive.

Primary sources and foundational literature

Bioelectric pattern control and planarian regeneration

• Oviedo NJ, Morokuma J, Walentek P, et al. Long-range neural and gap junction protein-mediated cues control polarity during planarian regeneration. Developmental Biology. 2010;339(1):188–199. doi:10.1016/j.ydbio.2009.12.012.
• Emmons-Bell M, Durant F, Hammelman J, et al. Gap junctional blockade stochastically induces different species-specific head anatomies in genetically wild-type Girardia dorotocephala flatworms. International Journal of Molecular Sciences. 2015;16(11):27865–27896. doi:10.3390/ijms161126065.
• Chernet BT, Levin M. Transmembrane voltage potential is an essential cellular parameter for the detection and control of tumor development in a Xenopus model. Disease Models & Mechanisms. 2013;6(3):595–607. doi:10.1242/dmm.010835.
• Grodstein J, Levin M. A computational approach to explaining bioelectrically induced persistent, stochastic changes of axial polarity in planarian regeneration. Bioelectricity. 2022;4(1):18–30. doi:10.1089/bioe.2021.0036.

Cellular inference and positional information

• Pio-Lopez L, Kuchling F, Tung A, Pezzulo G, Levin M. Active inference, morphogenesis, and computational psychiatry. Frontiers in Computational Neuroscience. 2022;16:988977.
• Marković A, Briscoe J, Page KM. Dynamics of positional information in the vertebrate neural tube. Journal of the Royal Society Interface. 2024;21(221):20240414. doi:10.1098/rsif.2024.0414.
• Berg HC, Purcell EM. Physics of chemoreception. Biophysical Journal. 1977;20(2):193–219.

Calcium as a temporal information code

• Dolmetsch RE, Xu K, Lewis RS. Calcium oscillations increase the efficiency and specificity of gene expression. Nature. 1998;392:933–936. doi:10.1038/31960.
• Li W, Llopis J, Whitney M, Zlokarnik G, Tsien RY. Cell-permeant caged InsP3 ester shows that Ca²⁺ spike frequency can optimize gene expression. Nature. 1998;392:936–941. doi:10.1038/31965.

Three-dimensional genome architecture

• Vietri Rudan M, Barrington C, Henderson S, et al. Comparative Hi-C reveals that CTCF underlies evolution of chromosomal domain architecture. Cell Reports. 2015;10(8):1297–1309.
• Lazar NH, Nevonen KA, O’Connell B, et al. Epigenetic maintenance of topological domains in the highly rearranged gibbon genome. Genome Research. 2018;28:983–997.
• Fudenberg G, Kelley DR, Pollard KS. Predicting 3D genome folding from DNA sequence with Akita. Nature Methods. 2020;17:1111–1117.
• Guo Y, Xu Q, Canzio D, et al. CRISPR inversion of CTCF sites alters genome topology and enhancer/promoter function. Cell. 2015;162(4):900–910.
• Khoury A, Achinger-Kawecka J, Bert SA, et al. Constitutively bound CTCF sites maintain 3D chromatin architecture and long-range epigenetically regulated domains. Nature Communications. 2020;11:54.
• Du Z, Zheng H, Huang B, et al. DNA methylation is required to maintain both DNA replication timing precision and 3D genome organization integrity. Cell Reports. 2021;36(12):109722.

Thermodynamics and energetic costs of cellular information processing

• Landauer R. Irreversibility and heat generation in the computing process. IBM Journal of Research and Development. 1961;5(3):183–191.
• Mehta P, Schwab DJ. Energetic costs of cellular computation. Proceedings of the National Academy of Sciences. 2012;109(44):17978–17982.
• Lan G, Sartori P, Neumann S, Sourjik V, Tu Y. The energy–speed–accuracy tradeoff in sensory adaptation. Nature Physics. 2012;8:422–428.
• Sartori P, Granger L, Lee CF, Horowitz JM. Thermodynamic costs of information processing in sensory adaptation. PLoS Computational Biology. 2014;10(12).
• Harvey SE, Lahiri S, Ganguli S. Universal energy–accuracy tradeoffs in nonequilibrium cellular sensing. Physical Review E. 2023;108(1):014403. doi:10.1103/PhysRevE.108.014403.

Electromagnetic receivers and timing-sensitive response

• Kim J, Hwang Y, Kim S, et al. Electromagnetic field-inducible in vivo gene switch for remote spatiotemporal control of gene expression. Cell. 2026;189(11):3465–3480.e23. doi:10.1016/j.cell.2026.03.029.
• Sousouri G, Eicher C, D’Angelo RM, et al. 5G radio-frequency-electromagnetic-field effects on the human sleep electroencephalogram: A randomized controlled study in CACNA1C genotyped volunteers. NeuroImage. 2025;317:121340.
• Burd SC, Bagheri N, Condon AF, et al. Magnetic resonance control of spin-correlated radical pair dynamics in vivo. Nature. 2026;651(8107):940–945. doi:10.1038/s41586-026-10282-4.
• Messiha HL, Wongnate T, Chaiyen P, et al. Magnetic field effects as a result of the radical pair mechanism are unlikely in redox enzymes. Journal of the Royal Society Interface. 2015;12:20141155.

Geometry as a compact representation in fundamental physics

• Arkani-Hamed N, Trnka J. The Amplituhedron. 2013.

Proposed one-sentence ceLLM definition

The Cellular Latent Learning Model proposes that every cell performs energy-constrained local inference by applying present bioelectric, chemical, mechanical, and environmental vectors to an evolved and adaptively updated physical prior embodied in genome sequence, three-dimensional chromatin, organelle architecture, and network state; multicellular form and health emerge from the fidelity with which those local inferences remain coordinated across space and time.