This is a theoretical framework developed through reasoning from first principles, pharmacological knowledge, harm reduction experience, and fragments of spinal cord injury case literature. It is not peer-reviewed. Several components are well-supported by current neuroscience; others are novel proposals that await experimental test; a few are explicitly speculative bridges. These categories are distinguished below.
Status: Novel framing, partially supported
The brain is approximately 60% lipid by dry weight. NAT proposes that this lipid-rich macro-structure is not merely metabolic scaffolding but functions as an active cognitive architecture — specifically as the high-level overseer of memory and motor program coordination.
Supported: Neuronal membrane capacitance exhibits circadian oscillations regulated by lipid metabolism, directly altering synaptic integration windows and network synchronisation. Myelin (70% lipid) regulates neural synchrony via conduction velocity. Astrocytes coordinate phase-synchronised neural wave patterns. The substrate exists.
Novel claim: NAT acts as a crystal coordinator — storing high-level memory components in lipid droplets and issuing top-down control signals to motor and memory circuits.
Gap: No direct evidence that lipid droplets store representational content (as opposed to metabolic state). This is the framework's most speculative structural claim. The mechanism connecting lipid droplet composition to cognitive content remains unspecified.
Status: Well-supported at cellular level
The fundamental unit of memory is proposed to be a functional triad: sensory neuron → interneuron → motor neuron. This is not merely a reflex arc but the primitive unit from which all memories are assembled.
Supported: Kandel's Aplysia work directly demonstrates that long-term memory storage involves structural changes across exactly this triad — sensory terminals grow or retract, motor neuron dendrites remodel, interneurons mediate the modulatory signal. Nobel Prize-level experimental validation.
Extended claim: The triad is the universal memory primitive, not just a sensorimotor special case. All memory types — episodic, semantic, procedural — are ultimately assembled from chains of these triads.
Gap: Extension beyond sensorimotor memory to episodic and semantic memory is inferred rather than demonstrated. The interneuron population is highly diverse (inhibitory subtypes play distinct roles) and the clean three-node picture may underspecify this complexity.
Status: Partially supported, novel assembly claim
Memories are not stored as fixed traces (contra the engram tradition) but are dynamically reconstructed each time from discrete sequential units called frames. These frames are assembled into coherent sequences by motor programs — the motor system doesn't just execute memories, it constitutes their sequential architecture.
Supported:
Novel claim: Motor programs generate episodic memory structure rather than being merely co-recruited with it. The causal direction is reversed from the mainstream view.
Gap: The causal direction — that motor programs assemble rather than merely accompany memory — is the hardest claim to defend and lacks direct experimental support. The apraxia evidence is suggestive but not conclusive.
Status: Well-supported
The myelinated white matter network is the physical substrate that associates distributed memory fragments into coherent memories. This is not novel — it maps precisely onto the structural connectome literature.
Supported: White matter connectivity forms the communication highways linking cortical and subcortical regions into functional networks. Myelination degree governs whether signals arrive in phase, making it a timing architecture as much as a wiring diagram. Oligodendrocytes actively support axonal energy metabolism independent of conduction velocity.
Framework contribution: Positioning the myelinated network explicitly as the associator of triad-based memory fragments gives the white matter a functional role that the connectome literature describes anatomically but doesn't frame in these terms.
Status: Speculative but mechanistically plausible
Motor program memory is not stored exclusively in the CNS. The proposal is a distributed architecture:
Supported:
Gap: No direct evidence that peripheral adipose tissue stores motor program components as opposed to metabolic state. The mechanism for encoding kinematic information in lipid droplet architecture at peripheral sites is entirely unspecified. This is the framework's most speculative peripheral claim.
Interesting prediction: Localised nerve blocks should selectively impair specific motor programs without affecting global motor cognition — this is partially testable with existing clinical tools.
Status: Novel mechanism, supporting evidence fragmentary but consistent
During sleep, the NAT issues the stored motor program downward through the triad network into the spinal column. The spinal column holds a proprioceptive record of the day's actual movements. A matching process compares the CNS program against this peripheral ground truth. On successful match, consolidation occurs via LTP and the program is stabilised. The glymphatic system then clears the metabolic cost of this process.
Supported:
Novel claim: The spinal column functions as a verification reference rather than merely a passive conduit during sleep. The matching process is the consolidation gate.
Gap: The matching/verification mechanism is inferred from the pieces above. No study has specifically examined spinal afferent activity as a verification signal during motor replay. The tetraplegic BrainGate participant — where replay occurred above a spinal lesion — represents the inverse of this prediction and is the most significant challenge: replay happened without intact spinal feedback. However, the quality or completeness of consolidation in that case was not compared to intact controls.
Status: Mostly supported, sequential dependency novel
Post-consolidation, myelin sheaths are updated to reflect the newly stabilised circuit architecture. The glymphatic system then clears debris from this remodelling process. This gives sleep a three-phase motor memory function: verify → consolidate (LTP) → update myelin → glymphatic cleanup.
Supported: All four components are individually documented. Their sequential dependency — that cleanup follows myelin update which follows consolidation — is consistent with known timing but not explicitly demonstrated as a causal chain.
Testable prediction: Interfering with glymphatic clearance specifically during the post-consolidation window (late slow-wave sleep) should degrade motor program retention even when LTP is intact. This has not been tested.
Status: Well-supported
The CNS terminates not only in the NAT (cortical/lipid architecture) but also in the cerebellum, which acts as the real-time bridge to the periphery. The cerebellum implements a 4-10 Hz oscillator that synchronises with the peripheral spinomuscular oscillator, providing the millisecond-level timing coordination that the NAT cannot supply at speed.
This gives a functional triangle: NAT (goal abstraction and oversight) ↔ Cerebellum (timing and error minimisation) ↔ Periphery (execution).
Supported: Directly confirmed by recent work showing cerebellar interposed nucleus handles moment-to-moment timing while mPFC handles adaptive switching between programs.
Status: Strongly supported, significant new evidence
Musical memory is the last form of memory lost in dementia. This is not incidental — it is direct evidence for the hardware/software distinction implicit in the rest of the framework.
Supported: Long-term musical memory is encoded in the caudal anterior cingulate and ventral pre-supplementary motor area — regions showing minimal atrophy and minimal glucose metabolism disruption in Alzheimer's compared to the rest of the brain. The temporal lobes are not essential for musical memory; motor-associated regions are. Procedural and retrograde semantic memory are relatively spared in Alzheimer's while episodic memory is impaired early.
Framework interpretation: Music is a maximally consolidated motor-program-assembled sequence — rhythmically regular, repeatedly verified across a lifetime of sleep cycles, and therefore most deeply written into myelinated motor hardware. The dementia progression sequence (episodic → semantic/language → procedural/musical → brainstem) is the reverse of the system's hardware/software gradient: newest, most lightly myelinated, most hippocampally-dependent memory is lost first; oldest, most heavily myelinated, most motor-grounded memory survives longest.
Status: Coherent reframing, consistent with known dissociations
A key refinement: the hippocampus is not where memories are stored. It is a working cache — temporary, high-bandwidth, context-sensitive — that holds incoming sensory-motor association data long enough for sleep-based verification to run. Verified content is written through to deeper, more permanent architecture (motor cortex, NAT, possibly local nodes at the entorhinal/brainstem level); unverified content is discarded or overwritten.
Supported: This explains the H.M. dissociation cleanly — hippocampal destruction erased the cache but left already-committed motor programs intact, since those had already been written through to hardware. It is also consistent with ongoing hippocampal neurogenesis continuously remodelling circuits, which would explain why retrieval cues degrade over time even for cache contents that were never properly committed.
Walking, worked through the full model:
Gap: The "integrity check" / matching mechanism remains the framework's central novel and unverified claim (see Section 6). The BrainGate tetraplegic case — where motor replay occurred above a spinal lesion, i.e. without intact spinal feedback — is a direct challenge to spinal cord involvement being necessary for replay, though it does not rule out spinal feedback affecting the quality of consolidation, which was not tested.
Status: Explanatory coherence test — passed without special cases
A model is more credible if it explains its failure modes as naturally as its successes. Common memory failures map cleanly onto three points of failure in the architecture:
| Failure pattern | Framework location | Mechanism |
|---|---|---|
| Low-value info forgotten | NAT tagging | Never flagged for verification — no salience signal, cache entry cleared unverified |
| Information overload | Cache capacity | Hippocampal cache overflows; concurrent neurogenesis-driven remodelling degrades retrieval cues before verification runs |
| Faces remembered, names forgotten | Triad strength | Face = rich multimodal sensory-motor triad; name = arbitrary symbol with no motor grounding, nothing for sleep verification to check against |
| Tip-of-the-tongue | Cache pointer loss | Underlying program/trace intact, but the cache index pointing to it has degraded or been overwritten |
| Context-dependent recall | Triad completeness | Sensory context is encoded as part of the frame; removing it removes part of the retrieval cue |
| Trauma memory (overpersistence) | NAT over-tagging | Extreme noradrenergic tagging causes repeated re-flagging for verification every sleep cycle, potentially deepening rather than resolving the trace |
| Intentional forgetting | NAT suppression | Top-down inhibitory signal (prefrontal → temporal cortex) prevents the cache entry from being submitted for verification at all |
| Interference forgetting | Cache competition | New triads sharing sensory components with old ones compete for the same cache slots; verification runs on the newer version, older pointer displaced |
This is presented as a coherence check, not as proof — failure-mode coverage is necessary but not sufficient for a theory to be correct.
Status: Strongly supported architecturally, novel synthesis
Working backward from the oldest sense (olfaction) and the oldest brain structures (brainstem), a stack of semi-autonomous local storage nodes emerges, each with confirmed local synaptic machinery, each progressively older evolutionarily, and each feeding the node above it:
| Structure | Local synaptic storage confirmed? | What it stores | Evolutionary age |
|---|---|---|---|
| Brainstem (CPGs) | Yes — central pattern generators with dedicated synaptic circuits | Breathing, swallowing, core locomotion primitives | Oldest, pre-cortical |
| Spinal cord | Yes — reflex arc synapses | Postural reflexes, withdrawal, stepping patterns | Ancient |
| Entorhinal cortex | Yes — local LTP independent of hippocampus, confirmed in lateral EC | Olfactory associations, episodic memory primitives | Early cortical |
| Hippocampus | Functions as cache, not permanent store (Section 10) | Working memory, context, novelty detection | More recent |
| Pre-SMA / motor cortex | Yes — deeply myelinated, minimally affected by Alzheimer's pathology | Overlearned motor sequences, musical memory | Cortical |
| NAT | Proposed (not confirmed) — lipid droplet storage | High-level program abstraction, oversight | Framework's core novel claim |
Why this matters: Alzheimer's disease progression strips this stack in reverse evolutionary order — newest and most software-like (episodic, hippocampal-cache-dependent) fails first; oldest and most hardware-like (brainstem CPGs for breathing and swallowing) fails last, and its failure is what kills the patient via aspiration pneumonia or autonomic failure. This is not assumed — it is the documented clinical progression pattern, and the framework's contribution is recognising it as a coherent unwinding of an evolutionary storage stack rather than a list of unrelated symptoms.
The olfactory-entorhinal link is the strongest single piece of supporting evidence in the whole framework: smell identification deficits predict 5-year cognitive decline better than episodic memory tests do, and Alzheimer's pathology (tau accumulation) appears in lateral entorhinal cortex before anywhere else. This is consistent with the entorhinal cortex acting as the gateway between the oldest sensory primitive and the hippocampal cache — when the gateway degrades, the cache starves before it visibly fails, which is exactly the observed clinical sequence.
Gap: NAT remains the only node in this stack without confirmed local synaptic/storage evidence. Every other node now has direct empirical support for local, semi-autonomous synaptic storage. This sharpens rather than resolves the framework's central open question.
| Component | Confidence | Primary Gap |
|---|---|---|
| NAT as lipid cognitive architecture | Moderate | Mechanism linking lipid droplets to cognitive content |
| Triad as memory primitive | High (sensorimotor), Moderate (universal) | Interneuron diversity underspecified |
| Frames assembled by motor programs | Moderate | Causal direction undemonstrated |
| Myelinated network as associator | High | Not novel — well-documented |
| Peripheral adipose storage | Low-Moderate | No encoding mechanism specified |
| Sleep verification loop | Moderate | Spinal verification role not directly tested |
| Myelin update in sleep | High (components), Moderate (sequence) | Sequential dependency not demonstrated |
| Cerebellum as co-terminus | High | Well-supported |
| Musical memory as hardware floor | High | Strongly evidenced; interpretation (not just data) is the framework's contribution |
| Hippocampus as cache, not store | Moderate-High | Consistent with dissociation evidence; "integrity check" mechanism still unverified |
| Failure-mode mapping | Explanatory, not predictive | Coherence achieved without special cases, but doesn't independently confirm the model |
| Evolutionary storage stack | High (each node individually), Moderate (NAT's place in it) | Every node but NAT has confirmed local synaptic storage |
None of these are in the current literature in this form. All three are experimentally approachable with existing tools.
Document compiled June 2026 from a working conversation. Framework is Andrew's original theoretical work. Literature citations available in source conversation.