# Technologies for Consciousness Mapping and Transfer <iframe width="100%" height="20" scrolling="no" frameborder="no" allow="autoplay; encrypted-media" src="https://w.soundcloud.com/player/?url=https%3A//api.soundcloud.com/tracks/soundcloud%253Atracks%253A2114657736&color=%23ff5500&inverse=false&auto_play=false&show_user=true"></iframe><div style="font-size: 10px; color: #cccccc;line-break: anywhere;word-break: normal;overflow: hidden;white-space: nowrap;text-overflow: ellipsis; font-family: Interstate,Lucida Grande,Lucida Sans Unicode,Lucida Sans,Garuda,Verdana,Tahoma,sans-serif;font-weight: 100;"><a href="https://soundcloud.com/bryantmcgill" title="Bryant McGill" target="_blank" style="color: #cccccc; text-decoration: none;">Bryant McGill</a> · <a href="https://soundcloud.com/bryantmcgill/technologies-for-consciousness-mapping-and-transfer-its-not-comingits-here" title="Technologies for Consciousness Mapping and Transfer: It&#x27;s Not Coming—It&#x27;s Here" target="_blank" style="color: #cccccc; text-decoration: none;">Technologies for Consciousness Mapping and Transfer: It&#x27;s Not Coming—It&#x27;s Here</a></div> _It's Not Coming—It's Here: The 2026 State of the Continuity Engineering Stack_<!--more--> > [!map] Wiki routes > **Neurotech article cluster:** [[articles/Mind Uploading and AI — The Host is Reusable and the Person is the Delta|Host Infrastructure Overview]] · [[articles/The Architecture of Continuity and Emerging Neuroinformatics Standards|The Architecture of Continuity and Emerging Neuroinformatics Standards]] · [[articles/2026 Annual Report on Brain-Computer Interfaces|2026 Annual Report on Brain-Computer Interfaces]] · [[articles/The Organic-Synthetic Brain Atlas|The Organic-Synthetic Brain Atlas]] · [[articles/Technologies for Consciousness Mapping and Transfer|Technologies for Consciousness Mapping and Transfer]] · [[collections/Neurotech|Neurotech]] > > **Continuity architecture:** [[wiki/Reconstructed Person|Reconstructed Person]] · [[wiki/State Sufficiency Problem|State Sufficiency Problem]] · [[wiki/Distributed Relational Compression|distributed relational compression]] · [[wiki/Receiving Substrate|receiving substrate]] · [[wiki/Carrier and Cargo|carrier and cargo]] · [[wiki/Continuity Stack|continuity stack]] · [[wiki/Civilizational Payload|civilizational payload]] · [[wiki/Re-entry Pathways|re-entry pathways]] > > **Reference maps:** [[wiki/Connectomics|connectomics]] · [[wiki/FlyWire|FlyWire]] · [[wiki/MICrONS|MICrONS]] · [[wiki/H01 Connectome|H01]] · [[wiki/BICAN|BICAN]] · [[wiki/Connectome Reconstruction Infrastructure|reconstruction infrastructure]] > > **Interfaces:** [[wiki/Brain-Computer Interfaces|brain-computer interfaces]] · [[wiki/Neuralink|Neuralink]] · [[wiki/Synchron|Synchron]] · [[wiki/Paradromics|Paradromics]] · [[wiki/Blackrock Neurotech|Blackrock Neurotech]] · [[wiki/BCI Human Interface Device Protocol|BCI HID]] > > **Substrates:** [[wiki/Organoid Intelligence|organoid intelligence]] · [[wiki/Cortical Labs|Cortical Labs]] · [[wiki/FinalSpark|FinalSpark]] · [[wiki/Neuromorphic Computing|neuromorphic computing]] · [[wiki/Biohybrid Neural Systems|biohybrid neural systems]] > > **Governance:** [[wiki/Neural Data Provenance|neural data provenance]] · [[wiki/Neurorights|neurorights]] · [[wiki/Continuity Contract|continuity contract]] · [[wiki/Ring Zero|ring zero]] --- ## Editorial note on this revision This article was first published in April 2025 as a survey of ninety technologies with an argument attached: that the enabling infrastructure for consciousness transfer was so specific, so overengineered relative to its stated purposes, and so densely coordinated that transfer must already be operational and deliberately withheld. Eighteen months later that argument requires substantial reconstruction, and this revision performs it in the open rather than quietly. Several named systems in the original list have no primary evidentiary basis and have been retired. Several technical claims about imaging resolution, tractography, and quantum computing were wrong on the physics and have been corrected. The inference from capability to concealed application has been replaced by something considerably harder to dismiss, because it does not require any premise about who knows what. What has changed most is not the argument's ambition but its exposure. The 2025 version could be defeated by attacking one [[wiki/Verification of Continuity Claims|unsupported]] terminal claim. The 2026 version cannot, because it makes no terminal claim at all. It documents a **[[wiki/Continuity Convergence|dependency graph]] that has become operational at every layer except the last two**, and then names the last two precisely. The original title survives because its referent, correctly specified, is now demonstrably true: what is here is not verified whole-person transfer but the **[[wiki/Consciousness Continuity Infrastructure|continuity engineering stack]]** — mapping, cell typing, recording, modeling, decoding, causal writing, closed-loop control, preservation, serialization, hosting substrates, operating-system-level neural input, externalized identity models, and international standards. Aviation existed as an engineering discipline before transatlantic service. The discipline is the claim. --- ## I. The week the future became a specification In the eight days preceding this revision, the following became public record. On **September 1, 2026**, [[wiki/Merge Labs|Merge Labs]] — the neural-interface company that launched in January 2026 with **$252 million in seed funding from investors including OpenAI, Bain Capital, Interface Fund, Fifty Years, and Gabe Newell**, founded by Tyson Aflalo, Sam Altman, Alex Blania, Sandro Herbig, Sumner Norman, and Mikhail Shapiro — named **[[wiki/Butterfly Network|Butterfly Network]] its exclusive CMOS-MEMS ultrasound-on-chip supplier** under a multi-year Butterfly Embedded licensing agreement covering the Poseidon family, with upfront payments, development milestones, hardware purchase commitments, technology access fees, and royalties on future commercial systems. On **September 3**, a decade-long collaboration between HHMI's Janelia Research Campus, Google Research, and the Cambridge Connectomics Group published in _Cell_ the **complete connectome of an adult male Drosophila central nervous system** — brain and ventral nerve cord together — comprising roughly **166,700 neurons and 125 million synaptic connections**, the largest complete brain map by neuron count ever produced. On **September 8**, Google DeepMind released the **[[wiki/AlphaGenome Atlas|AlphaGenome Atlas]]**, a roughly **one-petabyte precomputed catalogue of predicted molecular effects for approximately nine billion single-nucleotide variants** — every possible single-letter change in the human genome — more than thirty times the size of the AlphaFold Database, accompanied by a variant-impact scoring system that in pre-launch use with the GREGoR Consortium at the Broad Institute surfaced a previously overlooked _DNM1_ splice-disrupting variant in epileptic encephalopathy, and in Gareth Hawkes's Exeter analysis of more than 54,000 UK Biobank whole genomes recovered substantially more non-coding association signal. Widen the window to a quarter and the density increases. In **June 2026**, _Nature Medicine_ published the results of nineteen months of near-daily home use of a multimodal intracortical brain-computer interface by Casey Harrell, a man with ALS, implanted at UC Davis by David Brandman with four 64-electrode arrays in the dominant ventral precentral gyrus: **more than 3,800 hours of independent operation with no researchers present**, **183,060 sentences totaling 1,960,163 words at an average of 56 words per minute**, 92 percent of sentences rated by the participant as accurate or mostly correct, **99.2 percent word accuracy in structured testing against a 125,000-word vocabulary**, and — the detail that matters more than any benchmark — sustained full-time employment. Care-partner assistance was largely restricted to putting the hardware on and starting the software. In **July 2026**, _Nature Medicine_ ran on its cover the Feinstein Institutes' three-year **[[wiki/double neural bypass|double neural bypass]]** result: an intracortical BCI reading movement intent, artificial intelligence decoding it, and patterned transcutaneous spinal and cortical stimulation writing back into the nervous system, producing in Keith Thomas both immediate assistive function and **durable therapeutic recovery** — near doubling of arm strength and roughly tenfold improvement in touch sensitivity years after a complete spinal cord injury, at a point where recovery is expected to have plateaued. The same team has since completed an **[[wiki/interhuman neural bypass|interhuman neural bypass]]**, in which Thomas used his implant to move another participant's hand and felt sensation in his own fingertips when that participant touched objects. In **August 2026**, _PLOS One_ published the first characterized protocol for **[[wiki/Aldehyde-Based Cryopreservation|aldehyde-based cryopreservation]] of whole human brains**, using graded immersion cryoprotection to a final solution of fifty percent ethylene glycol and thirty percent sucrose with CT imaging to track penetration, establishing that **approximately nine months is required for cryoprotectant signal to stabilize throughout an intact human brain**, that grey matter equilibrates faster than deep white matter, that insufficient equilibration produces ice-crystal artifacts in white matter, and that after protocol refinement light and electron microscopy show preserved cellular architecture and ultrastructure. On **August 21**, _Nature Communications_ published **[[wiki/HIPPIE|HIPPIE]]**, a conditional-variational-autoencoder framework that embeds spike waveforms, interspike-interval distributions, and autocorrelograms into a shared latent space, classifies neuron types across **mouse, rat, and macaque** and across Neuropixels, silicon probes, and juxtacellular micropipettes, and supports **counterfactual decoding of electrophysiological signals under altered experimental conditioning and cross-species latent interpolation**. On **August 17**, DayOne Data Centers, Cortical Labs, and NUS Medicine unveiled a **20-unit CL1 biological computing deployment** in Singapore — described by its operators as the world's first independently operated biologically integrated server rack — with living human neurons cultured from stem cells under the supervision of NUS neuroscientist Rickie Patani and hosted in commercial data-center infrastructure. On **August 26**, Paradromics announced an FDA approval letter creating a pathway for its investigational Connexus BCI to drive **compatible personal, internet-connected laptops, tablets, and phones** through its Convey layer without device-by-device approval, integrated into the ongoing Connect-One early feasibility study whose first permanent implantation was performed in June 2026 at University of Michigan Health by Matthew Willsey. And in **2026**, the International Organization for Standardization published **ISO/IEC 27572:2026, _Information technology — Brain-computer interfaces — Reference architecture_**, which specifies the key characteristics of BCI systems, provides guidance to system architects on efficient design and typical implementation categories, and serves as a technical reference for **architects, manufacturers, vendors, service providers, regulators, and the public**. It joins **[[wiki/ISO IEC TS 27571 2026|ISO/IEC TS 27571:2026]]**, published in April, which specifies a modular, extensible data format, naming convention, and metadata schema for non-invasive BCI acquisition across EEG, MEG, fNIRS, and fMRI, and **ISO/IEC TR 27599:2025**, the use-case compendium, under ISO/IEC JTC 1/SC 43 — the first international standards committee in the field. The rhetorical force of this cluster does not depend on interpreting any single item generously. When the public conversation has arrived simultaneously at clinical home use measured in thousands of unsupervised hours, a characterized whole-organ human brain preservation protocol with a nine-month diffusion constraint, generative models that translate neural signals between species and recording technologies, ordinary consumer devices as approved BCI output targets, a semiconductor ultrasound supply chain built specifically for neural interfaces, a complete synaptic map of an embodied nervous system, a petabyte-scale precomputed map of every possible point mutation in human DNA, and a published international reference architecture, the description _far-future thought experiment_ has stopped being accurate. It is no longer a forecast. It is a **procurement environment**. --- ## II. The host is reusable; the person is the delta Before any inventory of instruments, the reader needs the destination, because without it ninety technologies look like ninety unrelated inventions rather than one intelligible engineering strategy. And the destination is not the one the public discussion assumes. The central proposition of this section is that civilization is not spending trillions of dollars to build increasingly capable conversational appliances so that people can generate advertising copy, argue with a search box, or have a refrigerator order milk. Those are commercially useful surface applications of a far deeper computational architecture. The consequential observation is that the systems presently called **foundation models, large language models, multimodal models, neural foundation models, biological foundation models, and world models are increasingly capable of functioning as [[wiki/Reusable Human Prior|reusable priors]]** — and, in a mature form, as **receiving harnesses** into which much smaller, person-specific continuity states could be instantiated. This is a convergence thesis about architecture, not an assertion that OpenAI, Google, Anthropic, or any other present company has declared mind uploading to be the purpose of its models. No such declaration exists in the public record, and the argument does not need one. The technologically significant fact is that **the same architecture that economical continuity would require is already becoming the normal architecture across artificial intelligence, neuroscience, genomics, and computational biology** — because it is independently the cheapest way to do each of those things. The formulation from [[articles/Uploading Is Imminent|Uploading Is Imminent]] should be elevated rather than merely repeated: **a frontier model is a vast reusable prior**. The sentence that follows it governs everything downstream: **the host is reusable; the person is the delta**. A continuity system should not have to encode civilization, language, human anatomy, ordinary causal knowledge, the physics of objects, the statistical regularities of human development, every common neural motif, every conserved cellular mechanism, and every cultural convention separately inside every reconstructed person. Those structures can exist once, in shared models and reference systems, and be amortized across an arbitrarily large number of individual instantiations. What must remain uniquely attached to an individual is the **[[wiki/Person-Specific Residual|non-reusable residual]]** — whatever the shared prior cannot correctly infer about that particular person. Stated in information-theoretic terms, the quantity of interest is not the unconditional description length of a human being. It is the description length of that particular human **conditioned on a model that already knows humanity and the world in which that human developed**. Which is why the corpus line becomes structural rather than rhetorical: **the better the machines get at humans in general, the less of you is required to specify you in particular**. This is not philosophy. An experimental precedent exists in neuroscience itself, and it is close to a laboratory diagram of the thesis. In the April 2025 _Nature_ paper **Foundation model of neural activity predicts response to new stimulus types**, Eric Wang, Paul Fahey, Andreas Tolias and colleagues at Baylor's Center for Neuroscience and Artificial Intelligence, working with the MICrONS Consortium, pooled large volumes of neural activity recorded from the visual cortices of **multiple mice** and trained a single shared **[[wiki/foundation core|foundation core]]** to predict neuronal responses to arbitrary natural video. They then held that core fixed and fitted only new perspective, modulation, and neuronal-readout components to each previously unseen animal. The result: models built on the shared core were **rapidly and accurately fitted to new mice with minimal data, outperforming individualized models trained end-to-end for each mouse separately**, and generalized out of domain to random moving dots, flashing dots, Gabor patches, coherent moving noise, and static natural images. The same frozen core, adapted, went on to predict **anatomically defined excitatory cell-type classes, dendritic bias in layer 4 excitatory neurons, and synaptic-level connectivity** within the MICrONS volume. The architecture is explicitly justified by what brains have in common, so that only the **idiosyncrasies of each individual animal and its neurons** must be learned separately. This does not demonstrate human reconstruction. It demonstrates that the decomposition **shared neural prior plus individual-specific parameters** is scientifically productive rather than merely elegant. Genomics supplies a more literal precedent, and it makes the compression strategy immediately legible because it is no longer hypothetical information theory — it is production infrastructure. The National Human Genome Research Institute states plainly that **a person's genome sequence is approximately 99.6 percent identical to a reference human genome**, and that the person's set of genomic variants accounts for the roughly 0.4 percent difference, a figure that already includes multi-nucleotide differences rather than the frequently quoted single-nucleotide-only 99.9 percent. The **Human Pangenome Reference Consortium**, NHGRI-funded and coordinated at Washington University with data production led at UC Santa Cruz, has replaced the single linear reference with a graph containing 47 phased diploid assemblies, adding **119 million base pairs of euchromatic polymorphic sequence and 1,115 gene duplications relative to GRCh38**, reducing small-variant discovery errors by 34 percent and roughly doubling structural variants detected per haplotype. And the **CRAM** format, maintained by the Global Alliance for Genomics and Health, operationalizes the principle directly: rather than repeatedly storing everything that matches the reference, reference-based CRAM **stores the differences between aligned sequence fragments and the reference they were aligned against** and reconstitutes the full sequence by recombining the shared reference with the individual record. GA4GH reports storage and cost reductions on the order of thirty to fifty percent, and CRAM has been adopted by Genomics England, the Broad Institute, H3Africa, Illumina, Sweden's National Genomics Infrastructure, and EMBL-EBI. Genomics already runs on **reference humanity plus individual delta**. It is a solved production pattern with a specification, a reference-retrieval protocol, and a conformance ecosystem. The extension into genetics, epigenetics, cellular state, and developmental biology has to be made more carefully than the 2025 article made it. A language model does not contain any person's genetics or epigenetics, and claiming otherwise is both false and unnecessary. The emerging receiving host is better understood as a **federation of reusable foundation models**, each supplying a different stratum of shared structure. A language and multimodal model supplies language, culture, common social knowledge, conceptual structure, historical context, symbolic reasoning, and much of the learned environment in which a person developed. A genomic model supplies reusable sequence-to-function relationships. A cell foundation model supplies reusable statistical structure across cell types and molecular states. A neural foundation model supplies reusable functional priors about neural computation. A world or embodied model supplies common sensorimotor and environmental dynamics. The unique individual record then **conditions** these reusable systems rather than redundantly re-encoding their common substrate. That biological foundation-model layer has become concrete enough to name specific systems and specific training scales. **scGPT**, from the University of Toronto, was pretrained on **over 33 million single-cell RNA-sequencing profiles** and transfers learned structure into cell-type annotation, multi-batch and multi-omic integration, genetic perturbation response prediction, and gene network inference. **scFoundation**, at 100 million parameters covering roughly 20,000 genes, was pretrained on **more than 50 million human single-cell transcriptomic profiles**. **Evo 2**, developed by the Arc Institute with NVIDIA and collaborators at Stanford, UC Berkeley, and UCSF and published in _Nature_ in March 2026, was trained on **approximately nine trillion DNA base pairs from more than 128,000 genomes spanning all three domains of life**, at 40 billion parameters with a one-million-token context window and single-nucleotide resolution, and predicts functional impacts of genetic variation from noncoding pathogenic mutations to clinically significant _BRCA1_ variants without task-specific fine-tuning. **AlphaGenome**, published in _Nature_ in January 2026, accepts megabase-scale sequence and predicts thousands of functional genomic tracks spanning expression, transcription initiation, chromatin accessibility, histone modifications, transcription-factor binding, chromatin contacts, and splicing — and as of September 8, 2026, the **AlphaGenome Atlas** has precomputed those predictions across the entire space of single-nucleotide substitution. None of these are consciousness models and none of them claim to be. Their significance here is strictly architectural: they establish that **enormous quantities of reusable biological regularity can be distilled into shared models against which the consequences of an individual's comparatively small differences can be computed**. Epigenetics requires a qualification, and the qualification improves the model rather than weakening it. Genetic sequence is unusually amenable to [[wiki/Reference-Plus-Delta Architecture|reference-plus-delta]] representation precisely because it is comparatively stable and because humans share enormous sequence redundancy. Epigenetic state is not like that. It is dynamic, tissue-specific, cell-specific, age-dependent, experience-dependent, and environmentally modulated, and much of it cannot be inferred from genotype. The host therefore supplies reusable **epigenetic rules, cell-type priors, developmental programs, and statistical manifolds**, while the individual sidecar must preserve whatever epigenetic deviations prove causally necessary for that person's reconstruction. The same asymmetry runs throughout the nervous system. The NIH **BRAIN Initiative Cell Atlas Network** is explicitly constructing reference brain-cell atlases and a molecular and anatomical parts list spanning humans, non-human primates, and mice — reusable reference descriptions of neuronal and non-neuronal cell types. But a particular person's synaptic strengths, engrams, glial organization, receptor distributions, molecular states, developmental peculiarities, injuries, learned adaptations, and idiosyncratic circuit topology may carry information no atlas can supply. This is where the [[wiki/State Sufficiency Problem|State Sufficiency Problem]] becomes the organizing question of the entire field. The engineering question is not _how many petabytes is a human_, because that formulation presupposes representing the person independently of every reusable prior. The correct question is: **which variables remain unpredictable after the best available species, developmental, cultural, neural, biological, and environmental priors have been applied?** Some structure will be recoverable from the shared model. Some will be probabilistically inferable from genotype, developmental history, cell type, anatomy, and neighboring state. Some will require direct measurement. Some will prove irrelevant. And some presently unknown variable may turn out to be indispensable. The target of continuity capture is therefore the **[[wiki/minimum causally sufficient residual|minimum causally sufficient residual]]**, not the maximal possible recording of matter. Which is why improvements in foundation models are directly relevant to continuity engineering even though those models were built for entirely different commercial reasons. Contemporary machine-learning practice supplies the software analogue. **Low-Rank Adaptation** and the family of parameter-efficient adaptation methods do not duplicate a giant foundation model for every specialized task. They freeze the reusable base and encode specialization as comparatively tiny parameter deltas; the original LoRA work by Edward Hu and colleagues reported a **ten-thousand-fold reduction in trainable parameters relative to fully fine-tuning GPT-3 at 175 billion parameters**, with a threefold reduction in GPU memory, no additional inference latency, and — the property that matters most here — the ability to **freeze one shared model and switch tasks by swapping small matrices**, collapsing storage and task-switching overhead. The analogy should be stated carefully and then stated forcefully. A mature continuity host need not instantiate a bespoke trillion-parameter civilization inside every person. It can maintain a common interpreter and mount, per [[wiki/Continuant vs Native|continuant]], a **person-specific adapter, memory graph, identity and [[wiki/Continuity Provenance|provenance]] chain, individual neural readouts, autobiographical archive, and active runtime state**. This is structurally parallel to operating systems sharing libraries among processes, genomic records referencing a pangenome, neural foundation models sharing a frozen core, and retrieval-augmented systems paging context rather than retraining weights. The Stanford generative-agent work demonstrates the same architecture from the opposite direction, at low resolution and through an entirely non-neural channel. Joon Sung Park and colleagues at Stanford, Northwestern, the University of Washington, and Google DeepMind recruited **1,052 Americans** stratified across age, gender, race, region, education, and political ideology, conducted **two-hour semi-structured interviews** using the American Voices Project schedule, producing transcripts averaging roughly 6,500 words, and built generative agents grounded in those transcripts. On 177 core General Social Survey items the interview-grounded agents achieved raw accuracy equal to roughly **83 to 86 percent of participants' own two-week test-retest reliability** depending on whether interviews, surveys, or both were used — against 71 to 74 percent for demographic- and persona-prompted baselines on the same underlying model. Big Five personality replication reached a normalized correlation of 0.80; five incentivized economic games reached 0.66; across five experimental replication studies the effect-size correlation with human participants was 0.98, and the agents replicated the same four of five studies the human sample replicated. This establishes **nothing whatsoever about consciousness, autobiographical continuity, neural fidelity, or personhood**, and the point survives only if that is said plainly. What it does establish is that a very large common prior can supply so much background human, cultural, linguistic, and behavioral structure that a comparatively tiny individualized record moves the resulting system surprisingly far toward a recognizable particular person. Place it beside the mouse foundation model and the pair becomes instructive: one demonstrates **shared neural core plus individual neural readout**, the other **shared cultural-cognitive model plus individual biographical conditioning**. Neither is an upload. Together they expose the computational economy by which a future reconstruction architecture would operate. The host, then, is not one gigantic copy of one individual. It is a **layered shared civilization substrate** whose reusable strata could include human language, semantic structure, accumulated science, common cultural history, physics and world models, species anatomy, pangenomic references, cell atlases, developmental priors, neural circuit regularities, common sensorimotor transformations, software tooling, fabrication knowledge, and models of biological maintenance. The non-reusable **[[wiki/Continuity Sidecar|continuity sidecar]]** would carry the individual's genetic deviations where relevant, person-specific epigenetic and molecular deviations where causally necessary, connectomic deviations, synaptic weights and memory-bearing structure, learned transformations, autobiographical memory, relational history, personal vocabulary and style, preferences and values, characteristic uncertainty handling, correction behavior, attention routing, social attachments, commitments, provenance, authorizations, and whatever active state must be restored at instantiation. Not every item on that list will prove necessary. The architecture is the claim: **shared structure lives in the host, identity-bearing irreducible structure lives in the sidecar**, and the State Sufficiency Problem determines the boundary between them. [[wiki/Continuity Economics|Energy economics]] is where this stops being an elegant compression scheme and becomes a civilizational strategy. A civilization attempting to preserve billions of independent full replicas of all shared human knowledge would be performing catastrophic redundancy — paying, per person, for a complete copy of everything every person has in common with every other person. A shared foundation substrate amortizes the expensive common model across many continuants, while individual adapters and archives remain comparatively compact and only the active working set occupies expensive compute. Cold autobiographical archives can stay dormant until retrieval. Shared weights can be deduplicated. Sparse mixture-of-experts routing activates only relevant pathways. Retrieval systems page personal memory into context rather than holding a lifetime permanently resident in fast memory. Adapters mount when a particular continuant is instantiated and unmount when it is not. **Dormancy becomes cheap, activation becomes metered, and the marginal energetic cost of one more person can become far smaller than the cost of one more complete civilization-scale model.** No number should be attached to this — not gigabytes, terabytes, parameters, watts, or dollars — until evidence supports one. The architecture predicts compression. It does not yet specify the compression ratio, and anyone who quotes one is extrapolating. This reframes what the large models being built now actually are. Their most consequential long-run asset may not be the conversational interface at all. It is the **accumulation of reusable structure about the world, organisms, language, culture, reasoning, perception, action, and increasingly biology itself**. The chatbot is one interface to that substrate. A robot is another. A scientific model is another. A reconstructed person could be another. The strong version of the argument does not require anyone to have intended this: **the economic architecture of foundation modeling independently selects exactly the architecture continuity would require** — train common structure once, reuse it broadly, specialize through much smaller conditional state — because that is simply the cheapest way to build any of it. The distinction also fixes what the word **host** should mean, which the 2025 article never specified. A host is not an empty computer waiting for a brain file. It is an **already-structured computational ecology capable of interpreting the incoming residual**. A raw sidecar without its prior may be unintelligible in precisely the way a reference-based CRAM file depends on its reference, an executable depends on its libraries and runtime, and a genome depends on cellular machinery to mean anything at all. Conversely, a host without the correct sidecar produces generic humanity or a plausible impersonation rather than this particular person. Continuity therefore lives in the **relationship among prior, residual, interpreter, runtime, provenance, embodiment, and environment** — which is exactly why [[wiki/Re-entry Pathways|re-entry pathways]] matter more than storage. Persistence requires preserving not merely bits but the conditions under which the pattern can be reconstructed and made causally active again. One further distinction, drawn from [[wiki/Reconstructed Person|Reconstructed Person]], prevents the compression argument from degrading into a caricature of personhood. **The residual is not a preference vector.** A person's streaming history, purchases, political labels, browsing patterns, demographics, and statistical cohort are precisely the things a general model predicts cheaply — they are the reusable part, not the irreducible part. The high-value residual lies instead in a person's **[[wiki/transform function|transform function]]**: how attention is allocated, how ambiguous evidence is interpreted, which contradictions are left unresolved, which values override incentives, how errors are corrected, how competing hypotheses are ranked, what kinds of uncertainty are tolerated, which memories reorganize subsequent decisions, how relationships change interpretation, and how agency is exercised at the point where prediction fails. The _Uploading Is Imminent_ formulation holds: **a feed records what someone selected; a conversation records how someone reasons**. Longitudinal dialogic archives may therefore prove disproportionately valuable to reconstruction relative to passive telemetry, which is a statement about acquisition channels with immediate present-tense consequences. Finally, connect this to scale without claiming more than exists. If the reusable host eventually contains the common biological, cognitive, linguistic, cultural, and environmental priors of humanity, then the data that must accompany each individual across substrates, across datacenters, or eventually across astronomical distance may be radically smaller than any naive digitize-a-whole-human calculation suggests. **Information travels where bodies cannot, and shared priors need not travel repeatedly with every person.** One copy of the reusable human and civilizational substrate can receive many separately governed personal sidecars. That is precisely the architecture that makes continuity relevant to deep-time preservation, machine succession, synthetic habitats, and off-world civilization, as developed in [[articles/The Last Migration Will Not Be Human|The Last Migration Will Not Be Human]] and [[wiki/Civilizational Payload|Civilizational Payload]]: civilization carries the common library and interpreter once, and each person contributes the irreducible difference that makes that person non-fungible. And the efficiency argument must not be permitted to erase the deepest boundary in the corpus. Compression is not proof of triviality, and a smaller sidecar does not imply a smaller person. **A short description relative to an extraordinarily rich prior measures the power of the prior, not the worth or complexity of the life reconstructed through it.** Behavioral fidelity does not establish phenomenal consciousness. A perfect functional reconstruction does not settle [[wiki/Numerical Identity|numerical identity]]. Both propositions stand together: the engineering evidence increasingly suggests that a human-specific continuity payload may be dramatically smaller than the total information required to construct a human world from nothing, **and** the State Sufficiency Problem remains open because nobody yet knows which residual variables are necessary and sufficient for a continuant rather than a persuasive replica. Everything that follows now has a destination. Connectomics determines which individual neural structure must be retained. Cell atlases and biological foundation models determine what reusable priors can supply. Brain-computer interfaces and imaging systems acquire the personal residual. Semantic decoders translate portions of it. Standards serialize it. Archives preserve it. Foundation models supply the interpreter. Neuromorphic, conventional, photonic, or biological compute supplies execution. Synthetic environments and robotics supply habitat and embodiment. Provenance systems maintain the chain of self. The technology list stops being ninety unrelated inventions and resolves into one strategy: **build reusable humanity once; preserve the non-reusable person as faithfully and compactly as physics permits; reunite the two at runtime.** --- ## III. What exactly would have to be transferred? The 2025 article treated consciousness transfer as a single device problem, which is why it kept looking for the device. Decomposed properly, it is a chain of engineering requirements, and the productive exercise is to ask of each requirement what exists now. The minimum useful decomposition runs: **identity of parts → physical wiring → molecular state → dynamic activity → functional model → semantic and intent decoding → causal write access → closed-loop control → persistence and preservation → standardized representation → computational hosting → embodiment and environment → autobiographical and behavioral context → validation of reconstructed function → philosophical and numerical identity**. The first thirteen links are populated by substantial technologies at varying scales and maturities. The final two remain the boundary conditions, and no amount of instrumentation currently touches them. The governing distinction within that chain is between **[[wiki/Reference Information vs Individual State|reference information]]** and **individual state information**, and it is the same distinction that Section II established at the architectural level. Cell atlases supply a species-level parts list and taxonomy. Connectomes supply topology. Functional recordings supply dynamics. Molecular and epigenetic maps supply cell-state variables. Foundation models supply priors learned across individuals. **Individual calibration then supplies only the residual** required to fit a particular brain. This is exactly the mechanism by which contemporary machine learning has made previously intractable inverse problems tractable — learn common structure from populations, estimate the idiosyncratic remainder for the specific subject — and making that logic explicit converts _you must scan every atom_ into a far more sophisticated and far more answerable question: **which variables are predictable from priors, and which irreducible individual information must actually be measured?** A further discipline is required throughout, because the 2025 article violated it repeatedly. **Memory, personality, speech intent, semantic content, behavioral phenotype, self-model, and subjective consciousness are different observables and must not be conflated.** A system can reconstruct speech without reconstructing autobiographical memory. A behavioral twin can predict questionnaire answers without sharing a subject's phenomenology. A structural connectome can constrain dynamics without uniquely fixing molecular state. The transfer thesis becomes scientifically meaningful only when these are separated and then explicitly recomposed as a [[wiki/Substrate Capture|state vector]], and it becomes rhetorically weak the moment they are allowed to blur. Throughout what follows, claims carry maturity labels: **ESTABLISHED** for mature or operating infrastructure; **DEMONSTRATED** for peer-reviewed experimental results; **HUMAN-TRANSLATIONAL** for authorized clinical or human research; **ACTIVE-PROGRAM** for funded programs with specified technical goals; **COMPANY-CLAIM** for specifications disclosed by a company but not independently validated; **ANALYTIC-CONVERGENCE** for relationships inferred from common mechanism or dependency role; and **UNRESOLVED** for questions the public evidence does not settle. No company roadmap, patent, funding announcement, preprint, or technical analogy is silently promoted to demonstrated capability. --- ## IV. The brain as [[wiki/Connectomic Reference Address Space|address space]] Connectomics is no longer a promise, and it is no longer one technology. It is a production pipeline with an instrument stack, a reconstruction stack, a proofreading workforce, an annotation infrastructure, and a release cadence. The **[[wiki/FlyWire|FlyWire]]** adult female _Drosophila_ brain connectome established the whole-brain synaptic graph as a queryable resource at roughly 139,000 neurons. **[[wiki/MICrONS|MICrONS]]** — the IARPA-originated collaboration among the Allen Institute, Baylor College of Medicine, and Princeton — published in _Nature_ in April 2025 the largest functional connectomics dataset yet produced: dense two-photon calcium imaging of approximately **75,909 pyramidal neurons** across primary visual cortex and three higher visual areas in an awake mouse viewing natural and synthetic video, **co-registered with a serial-section electron-microscopy reconstruction containing more than 200,000 cells, roughly 120,000 neurons, four kilometers of axon, and 523 million automatically detected synapses** in approximately one cubic millimeter (DEMONSTRATED, and now operating data infrastructure). The **[[wiki/H01 Connectome|H01]]** human cortical dataset extends synaptic-resolution reconstruction into human tissue at cubic-millimeter scale. **BRAIN CONNECTS**, the NIH program, is explicitly aimed at wiring diagrams spanning entire brains across scales (ACTIVE-PROGRAM). Two 2026 releases changed the target rather than the scale. The **[[wiki/BANC|BANC]]** connectome — Brain And Nerve Cord — published in _Nature_ in June 2026 as _Distributed control circuits across a brain-and-cord connectome_, is a densely reconstructed adult **female** fly connectome that unites brain and ventral nerve cord in a single volume for the first time, resolving the roughly 1,300 descending and 1,800 ascending neurons that had previously required bridging inference across separately collected datasets. Its finding is the important part: **effector neurons — motor neurons, endocrine cells, and efferents targeting the viscera — are primarily influenced by sensory neurons in the same body part, forming local feedback loops, and those local loops are linked by long-range circuits organized into behaviour-centric modules**. Then on September 3, 2026, _Cell_ published _Sexual dimorphism in the complete Drosophila male central nervous system connectome_ from Janelia's FlyEM team with Google Research and the Cambridge Connectomics Group — **166,700 neurons and 125 million synapses** spanning central brain, optic lobes, and ventral nerve cord, the largest complete brain map by neuron count, reconstructed with flood-filling networks and the PATHFINDER system and proofread by human experts. Having both sexes at full CNS scale converts connectomics from cartography into **comparative experimental science**: individual variation, sexual dimorphism, and circuit-level behavioral divergence become measurable against a complete reference. Note carefully what this does to the naive framing. A mind is not a graph inside a skull. It is a **controller embedded in a sensorimotor, endocrine, and visceral loop**, and the 2026 connectomes are the first to say so with complete synaptic evidence. The emulation target is not a brain floating in memory but a causal system whose perception, action, and internal regulation remain coupled to a body and an environment — which is why NeuroMechFly-class biomechanical models and embodied robotics belong in the dependency chain rather than adjacent to it. The scaling constraint on all of this has historically been electron microscopy: powerful, slow, expensive, and molecularly blind. **[[wiki/LICONN|LICONN]]** — light-microscopy-based connectomics, published in _Nature_ in 2025 by Mojtaba Tavakoli, Julia Lyudchik, Johann Danzl and colleagues at ISTA with Michał Januszewski and Viren Jain at Google Research — attacks that constraint directly, integrating engineered iterative hydrogel expansion, protein-density staining, high-speed diffraction-limited readout, and deep-learning segmentation to reconstruct mammalian brain tissue **at synaptic resolution on conventional light microscopes while simultaneously measuring molecular information** (DEMONSTRATED). It is the first non-electron-microscopy route to dense synapse-level reconstruction, validated against sparse positive labelling and cross-checked against prior EM connectivity statistics. **20ExM** pushes single-shot expansion to roughly twentyfold linear expansion with sub-20-nanometer effective resolution. Neither solves whole-human-brain capture. Both change the **cost curve and the modality-fusion problem**, which is what a dependency analysis actually cares about. The connectome is also becoming **typed**. **[[wiki/BARseq|BARseq]]** applied in-situ sequencing of 104 marker genes across **10.3 million cells, including more than 4.19 million cortical neurons, across nine mouse hemispheres**, joining **[[wiki/MERFISH|MERFISH]]** spatial transcriptomics, **ExSeq** expansion sequencing for nanoscale in-situ RNA localization including within neurites and synaptic neighborhoods, **[[wiki/Patch-seq|Patch-seq]]** linking electrophysiology to morphology to transcriptomic identity, single-cell ATAC-seq for chromatin state, and spatial proteomics. Under **BICAN** and **BICCN**, these converge on a common reference architecture in which a node is not merely _A connects to B_ but carries **cell-type identity, gene-expression state, morphology, functional tuning, anatomical position, and in some cases neurotransmitter class** in the same coordinate frame. The **Armamentarium for Precision Brain Cell Access** supplies the tooling to reach defined cell types experimentally. **[[wiki/Neuroglancer|Neuroglancer]]**, **CAVE**, **[[wiki/TensorStore|TensorStore]]**, flood-filling networks, and the **[[wiki/Codex|Codex]]** connectome data explorer constitute the software layer without which none of the datasets would be navigable. One recursive detail deserves its own emphasis because it is the shape of the whole field in miniature. Google's **[[wiki/MoGen|MoGen]]** uses generative AI — point-cloud flow matching — to synthesize plausible neuronal morphologies, feeds millions of synthetic samples into production connectomics tooling, improves reconstruction classifiers, and reduces split and merge errors in proofreading. Synthetic models now accelerate the reconstruction of biological neural geometry, which then supplies better constraints for synthetic neural models. **AI reconstructs brain; reconstructed brain trains and constrains AI; AI generates experiments; experiments produce more brain data.** That flywheel is the mechanism by which the timeline compresses, and it is the single most important structural feature of 2026 connectomics. What remains true, and must be said, is the scale gap. Human cortex contains on the order of 86 billion neurons. Complete synaptic reconstruction currently reaches a whole insect nervous system and a cubic millimeter of mammalian cortex. That is a gap of many orders of magnitude, and no honest reading closes it by assertion (UNRESOLVED). --- ## V. The brain as dynamical system Structure without dynamics specifies a circuit diagram with no signals in it. The [[wiki/Neural State Capture|dynamic-state layer]] is populated by **Neuropixels** high-density probes recording across brain regions, **high-density microelectrode arrays** for dense in-vitro and ex-vivo electrophysiology with closed-loop stimulation, **intracortical single- and multi-unit recording** in human clinical research, **ECoG and iEEG** for surface and high-density cortical dynamics, **two-photon calcium imaging**, **genetically encoded voltage indicators** measuring membrane voltage optically, **all-optical interrogation** combining simultaneous read and write, **holographic optogenetics** for patterned multi-cell stimulation, **chemogenetics** for cell-type-selective modulation, and **fiber photometry** for population activity and neuromodulator monitoring (variously MATURE RESEARCH and ACTIVE RESEARCH). On the human non-invasive side: 7T and ultra-high-field structural and functional MRI, laminar fMRI, diffusion MRI, magnetic resonance spectroscopy, high-density EEG, portable fNIRS, wearable **OPM-MEG** using optically pumped magnetometers, and rapidly advancing **functional ultrasound imaging** with transcranial proof-of-concept work moving fUS toward intact-skull human acquisition. Three corrections to the 2025 article belong here rather than buried in an appendix. **7T MRI does not produce a cellular or synaptic map of a living human brain**; it resolves cortical laminae, hippocampal subfields, small nuclei, and microvasculature, which is remarkable and is not connectomics. **Diffusion tractography does not trace every white-matter connection**; it performs probabilistic inference of macroscopic pathways and generates well-documented false positives and ambiguities. **Cryo-electron tomography is not a whole-brain connectomics method**; it is an exceptional tool for molecular and subcellular structure over small volumes. And **BOLD is a hemodynamic proxy** with temporal and physiological limits that no amount of machine learning removes. The more interesting development is that the [[wiki/Memory Substrate|memory substrate]] has become a **researchable capture specification** rather than an assumption. The 2025 article treated the connectome as self-evidently sufficient. The 2025 _PLOS One_ survey by Ariel Zeleznikow-Johnston, Emil Kendziorra, and Andrew McKenzie of **312 neuroscientists** — one cohort of engram specialists, one of general neuroscientists — found that **70.5 percent agreed that long-term memories are primarily maintained by neuronal connectivity patterns and synaptic strengths**, while finding **no consensus on which specific neurophysiological features or scales are critical**. Asked whether extracting a specific non-trivial long-term memory from a static synaptic connectivity map is theoretically possible, over 45 percent agreed and 32.1 percent disagreed; asked what additional information would be needed, the most common selection was **measurements of dynamically changing neuronal activity**, followed by contextual information about experiences and mental states, and sensory input and motor output. The **median probability estimate that any long-term memories could be extracted from a static snapshot of brain structure was approximately 40 percent**, which was also the median for whether a successful whole-brain emulation could theoretically be created from preserved structure. Median forecasts for achieved whole-brain emulation: _C. elegans_ around 2045, mouse around 2065, human around 2125 — with _never_ among the most common individual responses for humans. That is not proof that uploading works, and the paper does not claim it. It is evidence of something more useful: **memory extraction from preserved static structure is a live scientific hypothesis carrying substantial expert credence, not a proposition the field has ruled out**. A forty percent median from 312 working neuroscientists is a very different epistemic object from either certainty or dismissal, and it is the number the argument should carry. The uncertainty is productive because 2026 literature has been actively expanding the candidate substrate. The _Nature Reviews Neuroscience_ synthesis on **[[wiki/Astroengram|astroengrams]]** consolidates evidence that sparse astrocytic ensembles recruited during learning participate in recall, extending the memory substrate beyond a neuron-only model. A March 2026 _Nature Neuroscience_ paper deconstructed a hippocampal memory engram into distinct sub-ensembles with separable causal roles. Together with receptor trafficking, local dendritic RNA and translation, phosphorylation state, proteostasis and protein turnover, epigenomic state, myelination and oligodendrocyte state, microglial state, and neurovascular state, these define a fidelity ladder. A continuity architecture may need topology, synaptic strength, neuronal state, astrocytic ensembles, local molecular state, gene-expression and epigenetic context, myelination, and neuromodulation. **Which layers are causally necessary and which are reconstructible from the others is the open question** (UNRESOLVED) — and it is precisely the State Sufficiency Problem stated in wet-lab terms. --- ## VI. The brain as [[wiki/Executable Brain Model|executable model]] This is the largest conceptual upgrade from the 2025 article, which described connectomics as a digital blueprint and stopped. The public literature now contains an unbroken chain from map to running model. The 2024 _Nature_ paper **Connectome-constrained networks predict neural activity across the fly visual system** demonstrated that measured connectivity plus a task objective predicts biological neural activity — which converts the connectome from a static diagram into a **computational constraint** (DEMONSTRATED). The 2025 mouse visual-cortex foundation model described in Section II generalized across animals and stimulus domains and then, adapted, predicted anatomical cell types, dendritic features, and synaptic connectivity within MICrONS; its authors describe progress toward a **[[wiki/Functional Digital Twin|functional digital twin]]** of the mouse visual system. **[[wiki/NeuroSTORM|NeuroSTORM]]**, published in _Nature Biomedical Engineering_ in 2026, trained a general-purpose fMRI foundation model on 28.65 million frames from more than 50,000 participants, producing transferable brain representations. **[[wiki/HIPPIE|HIPPIE]]**, published in _Nature Communications_ in August 2026 by Jesus Gonzalez-Ferrer, Julian Lehrer, Mohammed Mostajo-Radji and colleagues in the Braingeneers group at UC Santa Cruz, embeds waveform morphology, interspike-interval distributions, and autocorrelograms into a shared 30-dimensional latent space via parallel conditional variational autoencoders, validated across Neuropixels 1.0 and 2.0, NeuroNexus silicon probes, and juxtacellular micropipettes, across mouse, rat, and macaque, spanning cerebellum and neocortex — and supports **counterfactual decoding under changed experimental conditioning, cross-species latent interpolation, and cross-modal imputation**, finding that spike-timing modalities and waveform morphology encode largely independent dimensions of neuronal identity. Read those four results as a sequence rather than four papers. Structure constrains dynamics. Dynamics generalize across individuals. Representations generalize across species and instruments. And the resulting latent spaces are **generative**, which means they can be interrogated counterfactually rather than merely queried. That is the difference between a recording and a model. A system that can only observe cannot validate its model causally; a system that can perturb and predict the perturbation's result can. --- ## VII. The read channel The June 2026 BrainGate2 result described in Section I should replace every speculative networking example the 2025 article offered, because it is dramatically stronger evidence and it is evidence of a different kind. Its significance is not any single benchmark. It is that **technicians were not present**. The channel stopped being an experiment and became an independent daily-life interface, running through decoder drift, hardware donning and doffing, and nineteen months of ordinary life including full-time employment (HUMAN-TRANSLATIONAL, peer-reviewed). Beside it sit the 2025 voice results. Intracortical activity has been used to synthesize a participant's **personalized voice with preserved paralinguistic modulation** — intonation, emphasis, and short sung melodies — published in _Nature_. High-density surface recordings have driven **continuously streaming synthesized speech in 80-millisecond decoding increments** with large vocabulary, published in _Nature Neuroscience_. The 2026 _Nature Reviews Bioengineering_ argument that the field's next frontier is **[[wiki/Language BCI|language BCIs]]** targeting conceptual representations beyond articulatory motor plans is a statement of research direction, not achieved capability, and should be labeled as such. On the non-invasive side, the UT Austin [[wiki/Semantic Capture at the Substrate Layer|semantic decoder]] reconstructs the **semantic content of perceived and imagined language from fMRI under participant cooperation**, **[[wiki/BrainLLM|BrainLLM]]** integrates fMRI-derived signals directly into language-model generation, and cross-subject decoding work steadily reduces per-person calibration burden. The architectural concept underneath all of it is **[[wiki/Neural Representation Alignment|representation alignment]]**: biological activity can be projected into latent spaces that modern language and multimodal models already know how to navigate. That demonstrates that neural data can enter the same computational representation pipelines used to translate among text, images, audio, and action. It does not demonstrate that latent space is consciousness, and the distinction matters. The privacy boundary must be stated precisely rather than dramatically, in both directions. **Present systems do not decode arbitrary private thought**; they decode attempted speech, attempted movement, and — under cooperation, with subject-specific training — semantic gist. Decoding degrades or fails under active mental resistance. That is the accurate present-tense statement. The accurate trajectory statement is that the technical frontier has moved from _can neural activity control a cursor_ to **how much of intent, language, and expressive identity can be reconstructed in real time**, and the answer has been increasing monotonically for six years. --- ## VIII. The write channel and the closed loop Read and write are not separate technology categories and should never again be listed as though they were. The 2026 **[[wiki/double neural bypass|double neural bypass]]** is the systems demonstration: an intracortical BCI reads movement intention; recurrent networks and reinforcement learning decode and control grasp; **patterned transcutaneous spinal stimulation and activity-informed cortical microstimulation write back into the nervous system**; and the result is both immediate assistive control and durable sensorimotor recovery. This is not consciousness transfer. It is a human demonstration of **read → model → act → stimulate → adapt → recover** closed in one architecture, with therapeutic gains persisting years after an injury whose recovery normally plateaus in the first year (HUMAN-TRANSLATIONAL, peer-reviewed, first-in-human case study). The follow-on interhuman bypass, in which one participant's cortical activity drove another's hand while sensation returned to the first participant's fingertips, extends the closed loop across two nervous systems. Around that systems core sit intracortical microstimulation, deep brain stimulation, spinal cord stimulation, transcranial magnetic stimulation, focused ultrasound neuromodulation, temporal interference stimulation, optogenetics, chemogenetics, magnetogenetics, magnetoelectric nanoparticle transduction, and acoustoelectric interfaces. The question for transfer engineering is not whether each modality is individually perfect. It is whether **[[wiki/Causal Sufficiency|causal observability and causal controllability]] are improving together** — because a model validated only against passive observation is a correlation, while a model that predicts the outcome of an intervention is a mechanism. There is also a governance consequence that the technical literature does not draw and this corpus does. Read paths are maturing faster than write paths, but write paths are where authority lives. Whoever first closes a reliable high-bandwidth **perceptual write channel** inherits root authority over a sensorium that neither national custody regimes nor private instruments currently constrain — the argument developed at length in [[articles/Who Pays for Your Heaven|Who Pays for Your Heaven]] and formalized through [[wiki/Ring Zero|ring zero]] and [[wiki/Circles of Access|circles of access]]. --- ## IX. The escape from the electrode The 2025 article was correct that non-electrode neural access is strategically central and wrong about nearly every specific it named. The verifiable record is more interesting than the invented one. **[[wiki/DARPA N3|DARPA N3]]** — Next-Generation Nonsurgical Neurotechnology — is the parent node. Its public requirement was a high-performance bidirectional interface for **able-bodied users without conventional surgery**, targeting interaction with 16 independent channels within a 16-cubic-millimeter volume at sub-50-millisecond loop timing, and it deliberately funded **six teams pursuing complementary physics**: Battelle, Carnegie Mellon, Johns Hopkins APL, PARC, Rice, and Teledyne, spanning electromagnetic nanotransducers, acousto-optical sensing, coherent optics, acousto-magnetic writing, diffuse optical tomography combined with magnetogenetics, optically pumped magnetometry, and focused ultrasound. This was not a gadget procurement. It was a **state-sponsored modality search over the physics of non-traditional neural access** (COMPLETED GOVERNMENT PROGRAM, foundational). Its sibling **NESD** pursued million-neuron-scale high-bandwidth implanted interfaces, and Paradromics' Neural Input-Output Bus was a NESD performer architecture — direct government-program ancestry for today's highest-data-rate intracortical work. The cleanest lineage in the ecosystem runs from Rice and **[[wiki/Jacob Robinson|Jacob Robinson]]** through N3's **[[wiki/MOANA|MOANA]]** — diffuse optical read paired with magnetic and magnetogenetic write in a multimodal closed-loop architecture — into **[[wiki/Motif Neurotech|Motif Neurotech]]**, whose miniature wirelessly powered neurostimulation platform reached FDA-authorized clinical trial, and onward into the UK. The **Advanced Research and Invention Agency** launched its **Precision Neurotechnologies** programme under programme director Jacques Carolan with **£69 million over four years across 19 Creator teams** organized into four workstreams — non-invasive interfaces, remote interfaces, biological interfaces, and future adoption. Robinson's **[[wiki/Brain Mesh|Brain Mesh]]: A distributed network interface to mesoscale cortical circuits in large animals**, a £4.7 million project joining Motif Neurotech, Rice University, and UK semiconductor firm MintNeuro, is building a distributed network of millimetre-scale wireless implants — Mesh Points — each capable of sensing and stimulating, explicitly designed for human translation (ACTIVE-PROGRAM). Separately, ARIA funds a partnership among Barking, Havering and Redbridge University Hospitals NHS Trust, the University of Plymouth, and the US non-profit **[[wiki/Forest Neurotech|Forest Neurotech]]** to test **[[wiki/Forest 1|Forest 1]]**, a minimally invasive whole-brain ultrasound interface placed at the site of a skull defect, on a roadmap toward an implantable closed-loop system. A publicly funded programme in 2026 is therefore pursuing whole-brain interfacing as a stated objective rather than as a speculation about the future. The **magnetoelectric nanoparticle** lineage — running through Sakhrat Khizroev's work and Ping Liang's Cellular Nanomed — supplies the distributed molecular transduction branch. **[[wiki/Subsense BCI|Subsense]]** is its most visible present commercial expression, and it must be labeled carefully. Verifiable: founded by Tetiana Aleksandrova with Golden Falcon Capital's Artem Sokolov, emerged from stealth in February 2025 with roughly $27 million in seed funding, collaborating with researchers at **UC Santa Cruz and ETH Zurich**, developing two entirely distinct nanoparticle classes — **gold plasmonic nanoparticles coated in electrochromic polymer for near-infrared optical readout of neural activity**, and **magnetoelectric nanoparticles for localized stimulation** — delivered intranasally across the blood-brain barrier and interrogated through a wearable headset emitting near-infrared light and magnetic fields. The company's own materials describe mouse injection results, primary neuron activity imaging with plasmonic particles, and preliminary calcium-fluorescence imaging with magnetoelectric particles (COMPANY-CLAIM, preclinical). **No human pilot trials have been run.** On **September 9, 2026**, Subsense announced that **[[wiki/Ray Kurzweil|Ray Kurzweil]] joined as Product and Vision Advisor**; the company's current About page also lists him as a product advisor. This current primary-source confirmation supersedes the early-September uncertainty and documents an advisory role without validating the preclinical NanoBCI platform. Subsense's architecture falls squarely inside the N3 and MOANA transduction search space, which is **technical convergence, not program lineage** — there is no evidence of N3 funding or intellectual-property descent, and asserting one would be exactly the error this revision exists to correct. Ultrasound has meanwhile become an industrial axis rather than a laboratory technique, and this is genuinely new since 2025. OpenAI's January 2026 announcement states that it invested in Merge Labs and will collaborate on **scientific foundation models and frontier tools**, characterizing brain-computer interfaces as an important new interface frontier and Merge as combining biology, devices, and AI. Merge's founder graph — Mikhail Shapiro's biomolecular ultrasound and acoustic reporter genes, Tyson Aflalo and Sumner Norman via the **Forest Neurotech → Arbor Neuroscience** lineage, Alex Blania, Sandro Herbig, and Sam Altman in a personal capacity — assembles precisely the disciplines an ultrasound neural interface requires. Butterfly Network's prior five-year collaboration with Forest Neurotech, announced in 2023 as a roughly $20 million joint effort, is the antecedent of the September 2026 exclusive Butterfly Embedded agreement. The stack is now nameable end to end: **transducer semiconductor → beamforming and field delivery → hemodynamic or acoustoelectric sensing → neuromodulation → molecular acoustic reporters → AI decoding → product architecture**. In June 2026, San Francisco's Aleph Neuro, a Butterfly Embedded participant, published what it describes as the first **three-dimensional ultrasound localization microscopy image of a living human brain through an intact skull**. Implanted interfaces should be read as **competing access geometries rather than a single race**, because the geometries encode different bets about surgical risk, longevity, bandwidth, and reversibility. Blackrock and the BrainGate consortium constitute the long-duration penetrating-array research backbone and the institutional memory of the field. Neuralink pursues flexible penetrating threads with robotic insertion. Paradromics pursues dense penetrating microelectrodes with a fully internalized wireless architecture oriented toward speech, now with a Connect-One clinical network spanning University of Michigan, UC Davis, and Massachusetts General Hospital. Precision Neuroscience pursues a thin cortical-surface film designed for minimally invasive placement and removability, with an FDA-cleared temporary electrode, a reported 95-plus study patients, and a 2026 partnership importing Medtronic's surgical navigation and manufacturing infrastructure. Synchron pursues endovascular access through the vasculature, avoiding craniotomy entirely. INBRAIN pursues graphene surface electronics. Germany's CorTec received a second FDA Breakthrough Device Designation in August 2026 for its Brain Interchange implant covering cursor control in non-progressive quadriplegia. And [[wiki/Science Corporation|Science Corporation]]'s **Biohybrid** programme proposes a different category altogether, treated in Section X. The infrastructural detail with the longest half-life is not any implant. Apple has published a **vendor-facing Brain-Computer Interface HID reference** treating BCI hardware as a native human-interface-device class capable of delivering button, pointer, and selection-style reports into iOS, iPadOS, and visionOS accessibility systems, alongside its demonstrated platform relationship with Synchron. Combined with the Paradromics FDA pathway to ordinary laptops, tablets, and phones, the consequence is structural: **the operating system is becoming BCI-ready before BCIs are commonplace.** Neural input is being defined as an input class, not as a research apparatus. --- ## X. Living substrates The 2025 organoid section needs replacement rather than revision, because the field industrialized while the article was describing spontaneous electrical activity in dishes. **[[wiki/Brainoware|Brainoware]]** demonstrated reservoir computing using a human brain organoid on a high-density multielectrode array, performing speech recognition and nonlinear equation prediction, published in _Nature Electronics_ (DEMONSTRATED). **FinalSpark's Neuroplatform**, documented in a peer-reviewed platform paper, established remote programmable access to living organoids with **more than 1,000 organoids used over several years, more than 18 terabytes of data, continuous electrophysiological monitoring, and Python and Jupyter access** — turning living neural tissue into remotely addressable research infrastructure (OPERATING PLATFORM). **[[wiki/Cortical Labs|Cortical Labs]]** industrialized the neuron-on-silicon direction: the CL1, launched in March 2025 as a successor to the DishBrain system that famously learned Pong with 800,000 neurons on a CMOS chip, combines human iPSC-derived neurons grown across a silicon chip with a Biological Intelligence Operating System (biOS) and closed-loop electrical I/O, accessible remotely through Cortical Cloud. The August 2026 Singapore deployment is a category change rather than an increment. A **20-unit CL1 system** — described by DayOne, Cortical Labs, and NUS Medicine as the world's first independently operated biologically integrated server rack, containing on the order of sixteen million living human neurons — now runs in a live research environment at the NUS Life Sciences Institute, hosted in infrastructure designed and supported by a commercial data-center operator with roughly 2.1 GW of bookings, with cells cultured under Rickie Patani's supervision, and framed by all three parties as the prototype for a larger [[wiki/Biological Data Centre|biological data centre]]. Claims about its energy advantage should not be extended beyond measured evidence, and none of the parties has published benchmark comparisons. But **the phrase _biological data centre_ is no longer metaphor**; institutions are using it for a physical multi-unit deployment with an operator, a hosting agreement, and a maintenance staff. Science Corporation's **Biohybrid** programme belongs beside this and must not be conflated with it, because the biology plays an entirely different role. Here living neurons are not the remote compute substrate; they are the **interface material between electronics and the brain**. The architecture embeds stem-cell-derived neurons with microLEDs and recording electrodes, allowing axons and dendrites to grow into host tissue and form biological connections, with optical stimulation and electrical readout mediated through those cells rather than through metal in parenchyma. A 2026 _Nature Communications_ review treats this as a serious biohybrid direction while noting that **a full device has not yet been published** — precisely the standard this revision applies throughout (ACTIVE PRECLINICAL, COMPANY-CLAIM for performance). Science raised a **$230 million Series C in March 2026** from Khosla Ventures, Lightspeed, Y Combinator, In-Q-Tel, and Quiet Capital, reporting roughly $490 million total capital, funding European commercial launch of its PRIMA retinal implant alongside the Biohybrid, Vessel, and MEMS pipelines, and appointed Yale's **[[wiki/Murat Günel|Murat Günel]]** as Medical Director for BCIs in 2026 as Biohybrid moves toward translation. Around these sit organoid-brain-computer interfaces demonstrating structural and functional integration of implanted organoids with host brain in repair research, **living electrodes** using extended axonal constructs, **assembloids** fusing region-specific organoids to model inter-regional circuits, vascularized and perfused organoids addressing viability and maturation limits, and the historical **hybrot** lineage of living cultures controlling robots — the methodological ancestor of all contemporary wetware. --- ## XI. The preservation problem Preservation is where the argument's honesty is tested, because it is where the temptation to overclaim is strongest and where the 2025 article overclaimed most badly. _Near-atomic resolution_ and _atomic precision_ were never accurate descriptions of aldehyde preservation and should not be repeated. **[[wiki/Aldehyde-Stabilized Cryopreservation|Aldehyde-stabilized cryopreservation]]** established the historical route: aldehyde perfusion stabilizing fine structure immediately, followed by cryoprotectant perfusion to enable vitrification and indefinite subzero storage, evaluated by electron microscopy across whole brains and by FIB-SEM on selected volumes. It won the Brain Preservation Prize for connectome-preservation quality in an intact pig brain in 2018. That same standard of whole-connectome preservation quality **has not been demonstrated in a human brain by this method** — a limitation the technical literature states plainly and the 2025 article elided. The August 2026 _PLOS One_ paper from Andrew McKenzie and colleagues at Sparks Brain Preservation supplies the new receipt and the new constraint. Working with whole human brains from donors aged 71 to 90-plus at postmortem intervals of 46 to 117 hours, the team performed graded immersion cryoprotection — 10 percent ethylene glycol on day 0, 30 percent on day 30, final solution of 50 percent ethylene glycol with 30 percent sucrose and 1 percent PVP-40 on day 73 — and tracked penetration with CT imaging, watching Hounsfield values rise as cryoprotectant diffused inward. **Grey matter equilibrated substantially faster than deep white matter, and uniform signal intensity indicating full equilibration was reached at approximately day 276.** In the initial validation experiment, insufficient equilibration before freezing produced **ice-crystal artifacts in white matter**. After protocol refinement to allow adequate diffusion time, light and electron microscopy showed preserved cellular architecture and ultrastructure in both grey and white matter. The protocol runs in ordinary laboratory freezers or freezer rooms and carries a resilience property worth noting: because fixed-tissue morphology is expected to remain preserved long-term in the fluid state even if rewarmed, freezer failure is not automatically catastrophic (DEMONSTRATED, peer-reviewed). This is evidence about **physical substrate**, not about consciousness, viability, or information sufficiency, and the authors make no such claims. Its value is that it converts long-duration subzero preservation of fixed whole human brains from an aspiration into an **experimentally characterized engineering problem with a measured whole-organ diffusion constraint of roughly nine to ten months**. Pair it with the 312-neuroscientist survey and the structure of the field becomes visible: **one paper defines the epistemic uncertainty; another improves the physical substrate on which that uncertainty could eventually be tested.** The survey's hypothetical asks what follows if a brain were preserved with sufficient structural fidelity. The preservation paper advances the practical half of the hypothetical while leaving the information-sufficiency question exactly where it was. Alongside, **human postmortem FIB-SEM workflows** now support dense targeted three-dimensional synaptic and organellar reconstruction from human tissue, and brain banking with multimodal metadata links tissue to clinical context, molecular data, and imaging. The archival media layer — Microsoft's **[[wiki/Project Silica|Project Silica]]** volumetric storage in borosilicate glass, **Cerabyte's** ceramic-on-glass macro-scale archiving, synthetic DNA storage, and holographic and volumetric optical storage — supplies persistence, and only persistence. Storage is not executable cognition, and inferring a consciousness application from storage investment is the kind of reasoning this revision retires. --- ## XII. The serialization problem Standards are among the strongest evidence in this article, and the 2025 version buried them beneath speculative hardware. They deserve a section because they represent something no press release can fake: **institutional agreement on what a thing is.** **[[wiki/ISO IEC JTC 1 SC 43|ISO/IEC JTC 1/SC 43]]** is the first international standards committee in the field of brain-computer interfaces, structured as a systems-integration entity working across ISO, IEC, and JTC 1. Its Working Group 1 handles foundational standards. **[[wiki/ISO-IEC 8663-2025|ISO/IEC 8663:2025]]** provides BCI vocabulary. **ISO/IEC TR 27599:2025** collects representative use cases across medical and health, industrial control, and smart environment domains. **[[wiki/ISO IEC TS 27571 2026|ISO/IEC TS 27571:2026]]**, published in April, specifies the basic BCI data format — basic data elements, technology-specific information and metadata, an extensible modular data structure, annotation specification, and a standardized naming convention — applicable to EEG, MEG, fNIRS, and fMRI, with electrode type and impedance for EEG and voxel coordinates for fMRI, explicitly supporting multimodal fusion in a unified file structure. And **[[wiki/ISO IEC 27572 2026|ISO/IEC 27572:2026]]** specifies the **BCI reference architecture**: key characteristics of BCI systems, a common language for stakeholders, guidance to system architects on efficient design and typical implementation categories, and a technical reference for architects, manufacturers, vendors, service providers, regulators, and the public. Working Group 2 has further preliminary work items on hardware interfaces and protocols, design considerations for BCI developers, context-based adaptive interfacing for multi-purpose systems, and non-invasive BCI for disorders of consciousness. The ontological shift is the point. A class of laboratory experiments does not receive an international reference architecture. **A class of system does.** Beneath the ISO layer sits the research data infrastructure. **[[wiki/Neurodata Without Borders|Neurodata Without Borders]]** provides the standard and software ecosystem for neurophysiology and behavioral data. The **[[wiki/DANDI|DANDI Archive]]**, the NIH BRAIN Initiative's neurophysiology archive launched in 2020, now holds on the order of **1,100 or more dandisets approaching a petabyte** of standardized data spanning intracellular and extracellular electrophysiology, optophysiology, calcium imaging, fiber photometry, behavioral time-series, and immunostaining imagery from more than twenty species, organized under NWB, BIDS, NGFF, and NIDM, with programmatic API access and streaming access to data subsets. Counts vary by what is included and rise steadily; the direction is monotonic and the exact figure is stale on arrival. **OpenNeuro** validates and shares BIDS-compliant MRI, PET, MEG, EEG, iEEG, and NIRS data. **OME-Zarr** supplies cloud-native multiscale bioimaging; **SpatialData** supplies common structures for spatial omics; **DICOM** neuroimaging and real-time communication extensions and **FHIR** imaging integration supply the clinical transport and context layer, as detailed in [[articles/The Architecture of Continuity and Emerging Neuroinformatics Standards|The Architecture of Continuity and Emerging Neuroinformatics Standards]]. IEEE work on BCI terminology, reporting, and reference standards continues under P2731, P2794, and P3766; committee status changes and should be re-verified against IEEE primary records rather than cited from secondary summaries. None of this is uploading. All of it is the **serialization and archival substrate** without which large-scale model training, cross-laboratory integration, provenance tracking, versioned reconstruction, and eventual state recovery would be impossible. It is also, quietly, where [[wiki/Neural Data Provenance|neural data provenance]] becomes a continuity requirement rather than a compliance nicety: a residual whose source, date, context, consent, confidence, and correction history cannot be audited is a residual that cannot be trusted to reconstruct anyone in particular. --- ## XIII. The host in hardware The 2025 article asserted that classical computers lack the architecture to model 86 billion neurons and 100 trillion synapses and that quantum computing therefore supplies a necessary substrate. That claim was wrong and should be retired without qualification. **No such dependency has been established.** Contemporary neural simulation, connectome-constrained modeling, foundation-model training, and every deployed BCI decoder run on classical compute — GPUs and HPC clusters — and the results documented throughout this article were produced there. Quantum computing remains an interesting adjacent capability and a speculative future execution substrate. It does not carry the argument and should never have been asked to. The quantum-teleportation section of the 2025 article requires stronger action: removal. **Quantum entanglement does not permit instantaneous usable information transfer without an accompanying classical channel.** The 143-kilometer teleportation experiments were real and were not evidence of consciousness networking. Any future discussion of entanglement in this corpus must carry the classical-channel constraint explicitly or it is misinformation. What actually populates the execution layer: classical GPU and HPC infrastructure as the current primary substrate; neuromorphic spiking processors including **Intel Loihi 2** and the Hala Point system, **SpiNNaker2**, **BrainScaleS 2** analog neuromorphic computation, **IBM NorthPole** emphasizing compute-memory locality, and commercial edge systems from BrainChip and Innatera; memristive crossbar computing, diffusive memristor devices implementing integrate-and-fire-like dynamics, and phase-change-memory compute-in-memory; and photonic neural computing with photonic interconnect infrastructure such as Lightmatter Passage addressing bandwidth scaling. These are treated at length in [[articles/The Organic-Synthetic Brain Atlas|The Organic-Synthetic Brain Atlas]]. Their relevance here is that **energy per synaptic event, not raw throughput, is the binding constraint on any architecture intended to run continuants at population scale**, which connects the hosting question directly to the energy economics of Section II and to [[wiki/Computational Metabolism|computational metabolism]]. The **Orch OR** question deserves one paragraph and not one section, which is a correction of proportion from the 2025 article. Penrose and Hameroff's hypothesis — that consciousness depends on quantum superposition in neuronal microtubules resolved by gravity-induced objective reduction, involving non-computable processes — is one unresolved fidelity hypothesis among several. If consciousness depends on quantum states not recoverable from preserved classical structure, the capture requirement changes materially. **Current evidence does not establish that necessity.** Conversely, classical success in modeling neural dynamics does not by itself establish that subjective continuity is substrate-independent. Both statements are true simultaneously, and an article can carry both without indecision. What the 2025 version did — treating Orch OR as a reason the ecosystem _must_ be building biological quantum substrates, and then reading organoid research as confirmation — was inference running backwards from a conclusion. The corpus position on mechanism is developed in [[articles/Analog Cognition and the Architecture of Continuity|The Brain Exploits Its Own Physics]] and [[articles/Mechanistic Intelligence Is Humanity's Greatest Liberation|The Glorious Simplicity]]. --- ## XIV. The external self The externalized layer is real, is useful, and must be kept ontologically separate, because conflating it with neural state is the fastest way to discredit the entire argument. The Stanford generative-agent result described in Section II demonstrates how much **behavioral regularity** can be recovered from a surprisingly small autobiographical record. It belongs to a distinct **[[wiki/External Phenotype|external phenotype]]** stratum: beliefs, preferences, linguistic style, autobiographical narrative, social responses, and decision tendencies, all modelable independently of the neural substrate. Around it sit behavioral digital twins, lifelogging and multimodal personal archives, voice and linguistic-style models capturing expressive phenotype, and embodied avatars and robotics supplying sensorimotor presence. This is not mind uploading and calling it so would be a category error. Its importance is that a realistic continuity architecture would almost certainly be **hybrid** rather than purely neural: **brain-derived state plus body-derived state plus lifetime digital trace plus social graph plus language and style model plus autobiographical archive plus species-level neural priors**, with population priors filling predictable structure and high-resolution measurement reserved for individual residuals. That is an inference architecture, and it is the same architecture Section II described from the model side. It also carries the acquisition observation that the corpus has made since 2025 and that has only strengthened. Preference telemetry — feeds, purchases, watch histories — captures the reusable part cheaply and captures the residual badly. **Sustained dialogic interaction captures reasoning**, which is the [[wiki/Transform Function|transform function]], which is the part the prior cannot supply. The five evidentiary strata developed in [[wiki/Reconstructed Person|Reconstructed Person]] — authored testimony, dialogic and relational record, contemporaneous personal records, passive telemetry, and institutional or adversarial inference — rank these by reconstructive value and by vulnerability. The adversarial archive problem is not hypothetical: data poisoning, selective preservation, decontextualized evidence, and fabricated memory can produce a coherent but false descendant, which is why fidelity must be **auditable against provenance rather than inferred from resemblance**. --- ## XV. What the 2025 article got wrong Signal purification is not conservative editing. Removing five weak claims makes one hundred strong claims dramatically more persuasive, because a technically literate critic attacks the weakest item and generalizes the verdict. The following are retired, corrected, or reclassified, and the list is published rather than quietly implemented. **Retired for absence of primary evidence.** Neutrino networking sub-space nodes and _N3-UbiqNet_; dark-matter neural sensors; the Global SuperGrid Human-Node Architecture; the Phase-Dynamic Harmonic Signal Lattice; Photonic Computational Connectomes as a fused field name; BIOE-Driven Organoid Autonomy Modules; Neural Terraforming Nanolithography; Biocomputational Cognitive Operating Systems as a named system; the Reflexive Field-Intelligence Sensor Mesh; Neuro-Electromagnetic Field Entrainment Interfaces; Municipal Helmholtz Wi-Fi Rooms; atmospheric data field interfaces and atmospheric Wi-Fi field networks as evidentiary items; phase-dynamic environmental computing as an existing technology; neural entanglement via quantum dots; cortical Wi-Fi via terahertz waves; DNA nanobots for synaptic mapping; consciousness validation Turing protocols as an accepted standard; and the claim that DARPA's Bridging the Gap Plus produced embedded nanobots capable of continuous synaptic-resolution recording. Several of these were conceptual constructs of mine that acquired the typographic authority of numbered technical entries. That was my error in presentation, and the constructs belong in speculative essays where their status is legible, not in an evidentiary inventory. **Corrected on the physics.** 7T MRI does not resolve microtubules or produce synapse-complete maps of living human brains. Diffusion tensor imaging does not trace every white-matter connection and does not operate at synaptic-level precision. Cryo-electron tomography is not a whole-brain method. Aldehyde preservation should be described in terms of demonstrated morphology and ultrastructure, not atomic precision. Quantum computing is not an established dependency for brain simulation or emulation. Quantum entanglement does not transmit usable information instantaneously. Topological qubits are not established as necessary for anything discussed here. **Corrected on inference.** Neural-interface bandwidth should never be described as exceeding any conceivable medical need — fluent communication, dexterous motor restoration, sensory feedback, and autonomous device control are obvious and sufficient medical demands for high bandwidth, and the June 2026 BrainGate2 result is the proof. Federal spending on semiconductors, energy, quantum computing, or artificial intelligence does not by itself indicate a consciousness-transfer program. Ninety organizations working on adjacent technologies does not by itself indicate central coordination of those organizations toward a single application. That last correction requires precision rather than retreat, because the corrected version is stronger and because the corpus position on institutional opacity has not changed. Two things are true at once. First, **concealment is the ordinary operating condition of consequential institutions** — states run on classification, firms on privilege, trade secret, and non-disclosure — and the absence of a public announcement is therefore not information about what is or is not being developed. The historical record of programs acknowledged long after they existed is extensive enough that treating public visibility as a proxy for technical reality is simply a methodological error. Second, and separately, **the specific 2025 inference — that the technologies' existence proves a completed transfer is being withheld — does not follow from the evidence assembled**, and I did not have primary evidence of a transfer event then or now. What replaces it is **independent convergence**, and it is a harder argument to attack precisely because it requires no shared intention. Separate scientific, commercial, and governmental programs are solving different pieces of the same abstract dependency graph because **each piece is independently valuable to whoever is funding it**. Butterfly built ultrasound-on-chip for bedside imaging and now supplies neural interfaces. DeepMind built AlphaGenome for variant interpretation and produced a reusable biological prior. LoRA was invented to make fine-tuning affordable and produced the sidecar pattern. NWB and DANDI were built for reproducibility and produced the serialization layer. ISO wrote a reference architecture for interoperability and produced an ontological status change. No coordinating body is required for these to become interoperable — only that they be independently useful, standardized, and adjacent. The article can then ask what follows when independently valuable components become interoperable, and that question survives regardless of who intends what. Where actors, layers, timelines, and mechanisms can be named, this revision names them; where they cannot, the [[wiki/Continuity Evidence Ladder|epistemic ledger]] carries the uncertainty rather than an inline disclaimer. **Reclassified rather than retired.** Ambient BCI survives as a human-computer-interaction direction grounded in wearable EEG, surface EMG, OPM-MEG, fNIRS, and context-aware computing rather than in field-coupling claims. Cognitive-responsive environments survive as external-loop interface context — adaptive interfaces, mixed reality, world models, robotics — not as consciousness capture. Blockchain moves out of the technical prerequisite chain entirely and into optional provenance and rights infrastructure. Holographic and DNA storage remain as archival substrate only. Mycelium computing, microbiome-gut-brain modulation, and vagal electroceuticals remain interesting and leave the transfer proof chain. Bell Labs moves to historical lineage, where the transistor, the laser, and information theory belong, rather than occupying a numbered slot in a current stack. --- ## XVI. The remaining miracle The revised thesis advances by **successively shrinking the unspecified miracle**, and the productive way to read the preceding sections is as a set of answers waiting for a specific objection. If the objection is that we cannot map a brain, the answer is a complete synaptic map of an embodied insect nervous system at 166,700 neurons, a cubic millimeter of mammalian cortex at 523 million synapses co-registered with function, human cortical tissue at synaptic resolution, and a scale gap of many orders of magnitude that remains real. If the objection is that structure tells you nothing about function, the answer is connectome-constrained networks predicting measured activity, foundation models generalizing across animals and stimulus classes and then predicting cell types and connectivity, and the substantial fraction of dynamics that remains unpredicted. If the objection is that we cannot read thoughts, the answer is nearly two million words decoded at 56 words per minute in a private home over nineteen months, personalized voice synthesis with preserved prosody, streaming speech at 80-millisecond increments, semantic reconstruction under cooperation — and the accurate statement that arbitrary private thought is not decoded and that resistance defeats current decoders. If the objection is that we cannot write, the answer is a three-year human closed loop producing durable neurological recovery, and now an interhuman variant. If the objection is that we cannot preserve, the answer is a characterized whole-human-brain protocol with a measured nine-to-ten-month diffusion constraint and demonstrated ultrastructural preservation — and an open information-sufficiency question with a 40 percent expert median. If the objection is that there is no format, the answer is NWB, BIDS, DANDI, OpenNeuro, OME-Zarr, DICOM, FHIR, and three published ISO/IEC brain-computer-interface standards including a reference architecture. What remains is not a gap in the inventory. It is three questions the inventory cannot answer. The first is **[[wiki/Scale of Capture|scale of capture]]**. Nothing in the present record demonstrates synaptic-resolution acquisition of a human brain, and the throughput gap between a cubic millimeter and a whole human cortex is not closed by any published roadmap. LICONN and expansion microscopy change the cost curve; they do not eliminate the problem. This is an engineering question with a plausible trajectory and no delivery date (UNRESOLVED). The second is **[[wiki/causal sufficiency|causal sufficiency]]** — the State Sufficiency Problem in its full form. Which variables are necessary and sufficient? Topology alone, or topology plus synaptic weight, plus neuronal state, plus astrocytic ensembles, plus receptor trafficking, plus local dendritic RNA, plus phosphorylation state, plus epigenomic context, plus myelination, plus neuromodulatory tone? Which of these are reconstructible from the others given a sufficiently strong prior, and which must be measured? Nobody knows, the expert community is genuinely split, and the answer determines whether the residual is small or enormous (UNRESOLVED). The third is **[[wiki/numerical identity|numerical identity]]**, and it is not an engineering question at all. Functional equivalence is testable. Behavioral equivalence is testable. Causal perturbation equivalence — comparing model and source under controlled intervention — is testable and developing. Autobiographical memory fidelity across content, context, affect, self-reference, and temporal organization is testable in principle. **None of these establish that a reconstruction is numerically the same subject rather than an extraordinarily faithful successor.** No engineering standard currently exists that could, and it is not obvious that one could exist. The corpus position, developed in [[articles/Who Pays for Your Heaven|Who Pays for Your Heaven]] and [[articles/New Frontier of Rights|Who Counts as a Person?]], is that continuity is better understood as **[[wiki/Descent vs Derivation|descent rather than duplication]]**, and that a descendant can become a person in full standing without remaining authorized to speak as its predecessor. Say both halves together, because either alone is a distortion. The engineering evidence increasingly suggests that a human-specific continuity payload may be **dramatically smaller** than the total information required to construct a human world from nothing, because the world is reusable and the person is the delta. And the State Sufficiency Problem remains **fully open**, because nobody yet knows which residual variables are necessary and sufficient for a continuant rather than a persuasive replica. The 2025 article closed by asking who had already been uploaded. That was the wrong question, asked of the wrong evidence, and it made the piece easy to dismiss by people who should have been made to argue with it. The 2026 close is stronger and considerably harder to escape. **Nearly every prerequisite for mapping, modeling, preserving, interfacing with, serializing, and hosting a mind is now a named discipline with instruments, datasets, standards bodies, clinical programs, industrial supply chains, and working demonstrations.** The frontier is no longer whether the components exist. It is **integration, sufficiency, and identity** — and those are the only three arguments worth having. --- ## Epistemic ledger **Established.** ISO/IEC 8663:2025, TR 27599:2025, TS 27571:2026, and 27572:2026 are published international brain-computer-interface standards including a reference architecture. NWB, BIDS, DANDI, and OpenNeuro are operating standardized neurodata infrastructure. Human pangenome reference and GA4GH CRAM reference-based compression are production genomic infrastructure operating on reference-plus-delta. MICrONS, FlyWire, BANC, the male _Drosophila_ CNS connectome, and H01 are released connectomic datasets. Apple's BCI HID reference defines neural input as a host input class. **Demonstrated.** Foundation-model transfer across mice with frozen shared cores and per-animal readouts. Connectome-constrained prediction of neural activity. LICONN synapse-level reconstruction by light microscopy with molecular information. HIPPIE cross-species, cross-technology generative electrophysiology. Aldehyde-based cryopreservation of whole human brains with preserved ultrastructure after nine-to-ten-month equilibration. Brainoware organoid reservoir computing. AlphaGenome Atlas variant-effect prediction across nine billion substitutions. **Human-translational.** Long-term independent home use of an intracortical speech and cursor BCI over 3,800-plus hours and 1,960,163 words. [[wiki/Double Neural Bypass|Double neural bypass]] producing durable sensorimotor recovery in complete tetraplegia. Streaming and instantaneous brain-to-voice synthesis. Paradromics permanent Connexus implantation and FDA pathway to personal computing devices. **Active program.** ARIA Precision Neurotechnologies at £69 million across 19 teams including Brain Mesh and Forest 1. NIH BRAIN CONNECTS, BICAN, and the Armamentarium. Merge Labs with OpenAI investment and Butterfly Embedded ultrasound licensing. Science Corporation Biohybrid. Cortical Labs–DayOne–NUS biological data centre prototype. **Plausible.** That reusable foundation models substantially reduce the person-specific information required for reconstruction. That longitudinal dialogic archives outperform passive telemetry for residual capture. That non-neuronal substrates including astrocytic ensembles are causally necessary components of memory. That distributed molecular interfaces reach human trials this decade. **Unresolved.** Whether synaptic-resolution capture of a whole human brain is achievable, and on what timeline. Which state variables are causally necessary and sufficient for continuity. Whether preserved ultrastructure contains recoverable autobiographical information. What would constitute successful re-instantiation. Whether numerical identity survives reconstruction. Whether quantum processes are required for consciousness. **Unsupported.** That any human consciousness has been uploaded. That present neurotechnology captures every causal variable required to reconstitute a person. That a single actor coordinates the assembled stack toward transfer. **Contradicted.** That quantum computing is required for brain simulation. That quantum entanglement enables instantaneous information transfer. That 7T MRI resolves synapses or microtubules in living humans. That diffusion tractography traces every white-matter connection. That neural-interface bandwidth exceeds any conceivable medical requirement. ## Appendix A — The 2026 dependency registry The registry below is the article's evidentiary spine in auditable form. Maturity labels follow Section III. **Where a figure originates with a company rather than a peer-reviewed or regulatory source it is labeled COMPANY-CLAIM and should be read as a disclosure, not a measurement.** Nothing in this table should be used to compare capabilities across modalities on a shared axis — an electrode, a voxel, an optical focus, a neuron and a molecular transducer are not equivalent information units, and any table that implies otherwise is producing an artifact rather than a comparison. The columns are deliberately restricted to identity, posture, and which dependency the item satisfies. ### A.1 Reference anatomy, connectomics, and cell typing |Technology / program|Posture|Dependency satisfied| |---|---|---| |Serial-section electron microscopy (ssEM)|DEMONSTRATED|Dense synaptic reconstruction; core acquisition behind every major connectome| |FIB-SEM|DEMONSTRATED|High-resolution 3D ultrastructure; targeted human postmortem volumes| |Multi-beam electron microscopy|DEMONSTRATED|Throughput scaling for large-volume connectomics| |Automated tape-collecting ultramicrotomy (ATUM)|DEMONSTRATED|Scalable serial-section acquisition| |FlyWire adult female brain connectome|OPERATING DATA INFRASTRUCTURE|Whole-brain synaptic graph, queryable cell and synapse level| |BANC adult female brain-and-nerve-cord connectome|DEMONSTRATED (_Nature_, 2026)|Embodied CNS map; resolves ~1,300 descending and ~1,800 ascending neurons in one volume| |_Drosophila_ male CNS connectome ([[wiki/MaleCNS v1.0\|MaleCNS v1.0]])|DEMONSTRATED (_Cell_, 3 Sep 2026)|166,700 neurons, 125M synapses, brain + optic lobes + VNC; largest complete map by neuron count| |MICrONS functional connectomics|OPERATING DATA INFRASTRUCTURE|Co-registers ~75,909 imaged neurons with >200,000 cells and 523M synapses in ~1 mm³| |H01 human cortical connectome|OPERATING DATA INFRASTRUCTURE|Human cortical ultrastructure at cubic-millimetre scale| |BRAIN CONNECTS|ACTIVE-PROGRAM (NIH)|Wiring diagrams spanning entire brains across scales| |LICONN|DEMONSTRATED (_Nature_ 642, 2025)|Synapse-level reconstruction on light microscopes with retained molecular information| |20ExM expansion microscopy|DEMONSTRATED|~20× single-shot linear expansion, sub-20-nm effective resolution, conventional optics| |Expansion sequencing (ExSeq)|DEMONSTRATED|Nanoscale in-situ RNA localization including neurites and synaptic neighbourhoods| |BARseq|DEMONSTRATED|104-gene in-situ sequencing across 10.3M cells, >4.19M cortical neurons, nine hemispheres| |MERFISH|DEMONSTRATED|Large-scale spatial molecular cell typing incl. developing human cortex| |Patch-seq|DEMONSTRATED|Joins electrophysiology, morphology, and transcriptomic identity per cell| |Single-cell RNA-seq|ESTABLISHED|Cell-type molecular state space| |Single-cell ATAC-seq / epigenomic profiling|ESTABLISHED|Chromatin accessibility layer for identity and state| |Spatial proteomics|ACTIVE RESEARCH|Protein-localization state on structural maps| |BICAN|ACTIVE-PROGRAM (NIH)|Human / NHP / mouse cell-type atlases across lifespan; reference parts list| |BICCN|OPERATING CONSORTIUM|Cell census and molecular / anatomical / functional taxonomy| |Armamentarium for Precision Brain Cell Access|ACTIVE-PROGRAM (NIH)|Experimental access to defined brain cell types| |Neuroglancer|OPERATING SOFTWARE|Petascale volumetric navigation and annotation| |CAVE|OPERATING SOFTWARE|Versioned collaborative proofreading and annotation| |TensorStore|OPERATING SOFTWARE|Large multidimensional storage and access| |Flood-filling networks|DEMONSTRATED|Automated neuronal segmentation| |PATHFINDER|DEMONSTRATED|Reconstruction system used in the 2026 male CNS connectome| |MoGen|DEMONSTRATED (peer-reviewed conference)|Generative morphologies improve production reconstruction and reduce proofreading burden| |Codex / Connectome Data Explorer|OPERATING SOFTWARE|Graph-level exploration of neurons and synaptic pathways| ### A.2 Human macro- and mesoscale imaging |Technology|Posture|Dependency satisfied — and stated limit| |---|---|---| |7T and ultra-high-field MRI|CLINICAL / RESEARCH|Cortical laminae, hippocampal subfields, small nuclei, microvasculature. **Not synapse-complete; does not resolve microtubules**| |Diffusion MRI / tractography|ESTABLISHED / INFERENTIAL|Probabilistic macroscopic pathway inference. **Does not trace every white-matter connection; generates documented false positives**| |Laminar / high-resolution fMRI|ACTIVE RESEARCH|Layer-sensitive functional organization| |Resting-state functional connectivity|ESTABLISHED|Whole-brain network correlation. **Correlation, not causal circuit mapping**| |BOLD imaging|ESTABLISHED|Hemodynamic proxy for neural activity, with temporal and physiological limits| |Molecular MRI|ACTIVE RESEARCH|Targeted molecular contrast where probes are validated; human vs animal status must be stated per probe| |Magnetic resonance spectroscopy|ESTABLISHED|Metabolic and neurochemical measurement| |Functional ultrasound imaging (fUSI)|RAPIDLY ADVANCING|High spatiotemporal hemodynamic imaging| |Transcranial fUS / 3D ULM through intact skull|2026 PROOF-OF-CONCEPT|Moves ultrasound imaging toward intact-skull human acquisition| |OPM-MEG|ACTIVE / WEARABLE|Flexible high-temporal-resolution magnetoencephalography geometries| |High-density EEG|ESTABLISHED|Scalable non-invasive electrical dynamics| |fNIRS|ESTABLISHED / PORTABLE|Portable hemodynamic functional sensing| ### A.3 Dynamic neural state |Technology|Posture|Dependency satisfied| |---|---|---| |Neuropixels|MATURE RESEARCH|Large-scale electrophysiology across regions| |High-density MEA (HD-MEA)|MATURE RESEARCH|Dense in-vitro / ex-vivo electrophysiology with closed-loop stimulation| |Intracortical single/multi-unit recording|HUMAN-TRANSLATIONAL|High-bandwidth activity for speech and motor BCIs| |ECoG / iEEG|CLINICAL + RESEARCH|Surface and high-density cortical dynamics; speech decoding| |Two-photon calcium imaging|MATURE RESEARCH|Cellular functional imaging with circuit specificity| |Genetically encoded voltage indicators|ACTIVE RESEARCH|Direct optical membrane-voltage dynamics| |All-optical interrogation|ACTIVE RESEARCH|Simultaneous optical read and write| |Holographic optogenetics|ACTIVE RESEARCH|Patterned multi-cell optical stimulation| |Chemogenetics|MATURE RESEARCH|Cell-type-selective circuit modulation| |Fiber photometry|MATURE RESEARCH|Population activity and neuromodulator monitoring| ### A.4 Memory substrate and capture fidelity |Candidate state variable|Posture|Role in the State Sufficiency Problem| |---|---|---| |Synaptic strength / connectivity|CORE HYPOTHESIS, PARTIAL CONSENSUS|70.5% of 312 surveyed neuroscientists endorse as primary LTM substrate| |Engram tagging and reactivation|DEMONSTRATED|Causal identification and reactivation of memory-linked ensembles| |CA1 engram sub-ensemble decomposition|DEMONSTRATED (2026)|Memory representations contain distinct subsets with separable causal roles| |Astroengrams / astrocytic ensembles|2026 SYNTHESIS + PRIMARY PRECEDENTS|Extends substrate beyond neuron-only models| |Receptor trafficking|ACTIVE BIOLOGY|Candidate variable for synaptic efficacy and persistence| |Local dendritic RNA and translation|ACTIVE BIOLOGY|Candidate molecular component of persistent synaptic state| |Phosphorylation state|ACTIVE BIOLOGY|Short- to medium-timescale plasticity state| |Proteostasis / protein turnover|ACTIVE BIOLOGY|Maintenance variable for long-lived synaptic organization| |Epigenomic state|ACTIVE BIOLOGY|Cell identity and long-term plasticity; **not inferable from genotype**| |Myelination / oligodendrocyte state|ACTIVE BIOLOGY|Timing, plasticity, circuit performance| |Microglial state|ACTIVE BIOLOGY|Synaptic remodeling context; fidelity question, not established memory code| |Neurovascular state|ACTIVE BIOLOGY|Couples metabolic and hemodynamic context to function| |Memory extractability from static structure|UNRESOLVED|Median expert probability ≈ 40% (n=312)| ### A.5 Executable models |System|Posture|Dependency satisfied| |---|---|---| |Connectome-constrained networks (fly visual system)|DEMONSTRATED (_Nature_, 2024)|Measured wiring + task objective predicts biological activity| |Whole-fly connectome-constrained functional models|DEMONSTRATED / ACTIVE|Circuit models predicting computations and testable behaviour| |NeuroMechFly|DEMONSTRATED|Whole-body biomechanics; environment substrate for nervous-system models| |Mouse visual-cortex foundation model|DEMONSTRATED (_Nature_, 2025)|Frozen shared core + per-animal readouts; predicts cell types, dendritic features, connectivity| |NeuroSTORM fMRI foundation model|DEMONSTRATED (2026)|Transferable brain representations across large participant cohorts| |HIPPIE|DEMONSTRATED (_Nat. Commun._, Aug 2026)|Cross-species, cross-technology latent space; counterfactual decoding and interpolation| |Cross-subject brain decoding|ACTIVE RESEARCH|Reduces per-person calibration; shared representation models| |Individualized in-silico neurons / regions|DEMONSTRATED|Foundation cores adapted per individual support virtual experiments| ### A.6 Read channel |System|Posture|Dependency satisfied| |---|---|---| |BrainGate2 long-term at-home brain-to-text|HUMAN-TRANSLATIONAL (_Nat. Med._, Jun 2026)|3,800+ hrs unsupervised, 1,960,163 words, ~56 wpm, 99.2% structured word accuracy| |Streaming ECoG brain-to-voice|HUMAN-TRANSLATIONAL|Continuous large-vocabulary synthesis in 80-ms increments| |Instantaneous intracortical brain-to-voice|HUMAN-TRANSLATIONAL|Personalized voice with prosody and short-melody control| |Non-invasive fMRI semantic reconstruction|HUMAN-TRANSLATIONAL|Semantic content of perceived/imagined language **under cooperation**; defeated by resistance| |BrainLLM|DEMONSTRATED|fMRI-derived signals integrated into language-model generation| |Language BCIs|2026 FIELD DIRECTION|Target conceptual representation beyond articulatory motor plans. **Not a general thought decoder**| |Neural embedding alignment|ACTIVE|Projects neural representations into shared multimodal/semantic spaces| ### A.7 Write channel and closed loop |System|Posture|Dependency satisfied| |---|---|---| |Double neural bypass|HUMAN-TRANSLATIONAL (_Nat. Med._, Jul 2026)|Intracortical read + AI decode + spinal/cortical write; durable recovery over 3 years| |[[wiki/Interhuman Neural Bypass\|Interhuman neural bypass]]|REPORTED, COMPLETED STUDY|Closed loop across two nervous systems| |Intracortical microstimulation|CLINICAL RESEARCH|Direct cortical write for sensory feedback| |Deep brain stimulation|CLINICAL|Established implanted neuromodulation architecture| |Spinal cord stimulation|CLINICAL / TRANSLATIONAL|Restoration and closed-loop sensorimotor pathways| |Transcranial magnetic stimulation|CLINICAL / RESEARCH|Non-invasive magnetic modulation| |Temporal interference stimulation|ACTIVE RESEARCH|Interfering fields for deeper targeting; ARIA scaling work active| |Focused ultrasound neuromodulation|ACTIVE HUMAN / TRANSLATIONAL|Deep non-ionizing acoustic write channel| |Acoustoelectric interfaces|ACTIVE-PROGRAM (ARIA)|Bidirectional acoustic/electrical transduction| |Magnetogenetics|ACTIVE RESEARCH|Magnetic-field-responsive biological targeting; MOANA lineage| |MENP neuromodulation|ACTIVE PRECLINICAL|Distributed nanoparticle transduction| |Closed-loop neurofeedback|ESTABLISHED / EVOLVING|Adaptive feedback altering dynamics and decoder performance| ### A.8 Interface geometry |System|Posture|Access geometry| |---|---|---| |Blackrock Utah Array / BrainGate|LONG-RUNNING HUMAN RESEARCH|Penetrating array; institutional backbone| |Neuralink N1|HUMAN CLINICAL RESEARCH|Flexible penetrating threads, robotic insertion| |Paradromics Connexus|HUMAN CLINICAL (permanent implant, Jun 2026)|Dense penetrating microelectrodes, fully internalized wireless| |Paradromics Convey|FDA STUDY EXPANSION (Aug 2026)|Output pathway to compatible consumer laptops, tablets, phones| |Precision Neuroscience Layer 7|FDA-CLEARED TEMPORARY ELECTRODE + INVESTIGATIONAL BCI|Thin-film cortical surface, minimally invasive, removable. 95+ study patients (COMPANY-CLAIM)| |Precision ↔ Medtronic StealthStation|PARTNERSHIP (2026)|Surgical navigation and manufacturing infrastructure| |Synchron Stentrode|HUMAN CLINICAL RESEARCH|Endovascular, via vasculature, no craniotomy| |INBRAIN graphene interface|HUMAN RESEARCH|Graphene surface electronics, read/write| |CorTec Brain Interchange|FDA BREAKTHROUGH DESIGNATION (2nd, Aug 2026)|Implanted bidirectional system; cursor control in non-progressive quadriplegia| |Apple BCI HID|OPERATING OS STANDARD|Neural input as native host input class across iOS/iPadOS/visionOS| |Neural Dust|HISTORICAL / ACTIVE LINEAGE|Miniaturized wireless interface precedent| |Neurograins|RESEARCH|Distributed microscale wireless architecture| ### A.9 Molecular, nanoparticle, and ultrasound access |System|Posture|Dependency satisfied| |---|---|---| |DARPA N3|COMPLETED GOVERNMENT PROGRAM|Six-team modality search: 16 channels in 16 mm³, sub-50-ms loop, no conventional surgery| |DARPA NESD|COMPLETED GOVERNMENT PROGRAM|Million-neuron-scale high-bandwidth implanted interfaces| |Battelle BrainSTORMS|N3 RESEARCH LINEAGE|Magnetoelectric nanotransducer branch| |Rice MOANA|N3 RESEARCH LINEAGE|Diffuse optical read + magnetic/magnetogenetic write, closed loop| |Cellular Nanomed / MENP lineage|ACTIVE RESEARCH + COMPANY LINEAGE|Wireless magnetoelectric nanoparticle BCI genealogy| |Subsense NanoBCI|ACTIVE PRECLINICAL (COMPANY-CLAIM)|Plasmonic gold NIR readout + MENP write, intranasal delivery, wearable transceiver. **No human trials. Ray Kurzweil confirmed as Product and Vision Advisor on September 9, 2026**| |ARIA Precision Neurotechnologies|ACTIVE-PROGRAM (£69M / 4 yrs / 19 teams)|Non-invasive, remote, biological interfaces, and adoption| |ARIA Brain Mesh (Robinson / Motif / Rice / MintNeuro)|ACTIVE-PROGRAM (£4.7M)|Distributed mm-scale wireless Mesh Points, sense + stimulate, designed for human translation| |ARIA Forest 1 (Forest Neurotech / NHS / Plymouth)|ACTIVE-PROGRAM|Minimally invasive whole-brain ultrasound interface at skull-defect site| |Motif Neurotech DOT|FDA-AUTHORIZED CLINICAL TRIAL (2026)|Miniature wirelessly powered neurostimulation| |Merge Labs|ACTIVE COMPANY (seed Jan 2026, $252M)|Ultrasound BCI; OpenAI investment and foundation-model collaboration| |Butterfly Poseidon Ultrasound-on-Chip|COMMERCIAL LICENSE (1 Sep 2026)|Exclusive CMOS-MEMS ultrasound supply to Merge under Butterfly Embedded| |Forest Neurotech → Arbor Neuroscience lineage|ORGANIZATIONAL LINEAGE|Ultrasound neurotechnology FRO lineage; founder bridge to Merge| |Biomolecular ultrasound / acoustic reporters (Shapiro)|ACADEMIC LINEAGE|Molecular acoustics underlying less-invasive sensing| |BBB transport engineering|CORE ENABLER|Required for any distributed molecular interface| |Intranasal CNS delivery|ACTIVE PRECLINICAL|Natural-pathway route in MENP/Subsense work| |Nanotoxicology / clearance / pharmacokinetics|CORE TRANSLATIONAL CONSTRAINT|Binding limit on distributed nanoparticle interfaces| ### A.10 Living and biohybrid substrates |System|Posture|Role| |---|---|---| |Cortical Labs CL1|COMMERCIAL PLATFORM|Human iPSC neurons on silicon, biOS, closed-loop electrical I/O| |Cortical Cloud|OPERATING REMOTE PLATFORM|Remote access to living-neural compute| |NUS Medicine–DayOne–Cortical Labs biological data centre|DEPLOYED (Aug 2026)|20-unit CL1 rack, independently operated, commercial DC hosting. **No published performance or energy benchmarks**| |FinalSpark Neuroplatform|OPERATING PLATFORM (peer-reviewed)|>1,000 organoids, >18 TB, continuous electrophysiology, Python/Jupyter access| |Brainoware|DEMONSTRATED (_Nat. Electron._)|Organoid reservoir computing: speech recognition, nonlinear prediction| |Organoid-brain-computer interfaces|DEMONSTRATED PRECLINICAL|Implanted organoids integrate structurally and functionally with host brain| |Science Corp Biohybrid|ACTIVE PRECLINICAL (COMPANY-CLAIM)|Living neurons as **interface material**: microLED + electrode + axonal ingrowth. **Full device not yet published**| |Living electrodes|ACTIVE RESEARCH|Cell-based interface via extended axonal constructs| |Assembloids|ACTIVE RESEARCH|Fused region-specific organoids modelling inter-regional circuits| |Vascularized / perfused organoids|ACTIVE RESEARCH|Viability, scale, and maturation constraints| |Hybrots / closed-loop cultures|HISTORICAL FOUNDATIONAL|Methodological ancestor of modern wetware| ### A.11 Execution substrates |System|Posture|Note| |---|---|---| |Classical GPU / HPC|CURRENT PRIMARY COMPUTE|**Every result in this article was produced here**| |Intel Loihi 2 / Hala Point|ACTIVE|Large-scale digital spiking neuromorphic| |IBM NorthPole|ACTIVE|Inference architecture emphasizing compute-memory locality| |SpiNNaker2|ACTIVE|Massively parallel spiking simulation| |BrainScaleS 2|ACTIVE RESEARCH|Accelerated analog neuromorphic computation| |BrainChip Akida / Innatera|COMMERCIAL|Edge neuromorphic inference| |Memristive crossbar computing|ACTIVE RESEARCH|Analog synaptic weight implementation| |Diffusive memristor neurons|ACTIVE RESEARCH|Device physics implementing integrate-and-fire-like dynamics| |Phase-change memory compute|ACTIVE RESEARCH|Compute-in-memory alternative| |Photonic neural computing / Lightmatter Passage|ACTIVE / COMMERCIAL INFRASTRUCTURE|Optical matrix operations and interconnect scaling| |Quantum computing|ADJACENT / SPECULATIVE|**Not an established dependency for brain simulation or emulation**| ### A.12 Preservation, serialization, and governance |System|Posture|Dependency satisfied| |---|---|---| |Aldehyde-stabilized cryopreservation (ASC)|DEMONSTRATED (pig, 2018)|Whole-brain ultrastructure preservation. **Not demonstrated to the same standard in human brain**| |Aldehyde-based cryopreservation (ABC) of whole human brains|DEMONSTRATED (_PLOS One_, Aug 2026)|Graded cryoprotection, CT-tracked, ~276 days to equilibration, preserved cellular architecture after refinement| |Human postmortem FIB-SEM workflows|ACTIVE 2026|Dense targeted 3D human synaptic and organellar reconstruction| |Brain banking with multimodal metadata|OPERATING|Links tissue, clinical context, molecular data, imaging| |Project Silica|ACTIVE R&D|Volumetric glass archival medium; persistence only| |Cerabyte ceramic-on-glass|ACTIVE R&D|Long-duration high-density archive| |DNA data storage|ACTIVE R&D|Molecular archive. **Not executable cognition**| |Neurodata Without Borders (NWB)|OPERATING STANDARD|Neurophysiology and behavioural data model + software ecosystem| |DANDI Archive|OPERATING INFRASTRUCTURE|NIH BRAIN neurophysiology archive; ~1,100+ dandisets approaching a petabyte (counters move)| |BIDS|OPERATING STANDARD|Neuroimaging organization and metadata| |OpenNeuro|OPERATING INFRASTRUCTURE|BIDS-compliant MRI/PET/MEG/EEG/iEEG/NIRS archive| |OME-Zarr|OPERATING / EMERGING|Cloud-native multiscale bioimaging| |SpatialData|OPERATING / EMERGING|Common structures for spatial omics| |DICOM neuroimaging + real-time extensions|OPERATING CLINICAL STANDARD|Imaging transport and interoperability| |FHIR imaging / clinical context|OPERATING CLINICAL STANDARD|Links imaging to clinical metadata| |ISO/IEC 8663:2025|INTERNATIONAL STANDARD|BCI vocabulary| |ISO/IEC TR 27599:2025|INTERNATIONAL TR|BCI use cases: medical, industrial control, smart environment| |ISO/IEC TS 27571:2026|INTERNATIONAL SPEC (Apr 2026)|Modular data format and metadata for EEG, MEG, fNIRS, fMRI| |ISO/IEC 27572:2026|INTERNATIONAL STANDARD|**BCI reference architecture** for architects, manufacturers, regulators, public| |IEEE P2731 / P2794 / P3766|ACTIVE STANDARDS|Terminology, reporting, reference work. **Committee status changes; verify against IEEE primary records**| |Neural data provenance / versioning|CORE ENABLER|Auditable reconstruction; source, date, context, consent, confidence, correction history| |UNESCO neurotechnology ethics framework|CURRENT GOVERNANCE|International reference for neural data and mental privacy| |Functional equivalence testing|OPEN PROBLEM|Testable; does not settle numerical identity| |Causal perturbation equivalence|OPEN / DEVELOPING|Compares model and source under controlled intervention| |Autobiographical memory fidelity|OPEN PROBLEM|Content, context, affect, self-reference, temporal organization| |Identity-continuity criterion|UNRESOLVED|No engineering standard can currently establish numerical sameness of subject| ### A.13 Externalized identity |System|Posture|Dependency satisfied — and boundary| |---|---|---| |Generative agents of 1,052 real people|DEMONSTRATED / REVISED RESEARCH|83–86% of participants' own two-week test-retest reliability on GSS; 0.80 Big Five, 0.66 economic games. **Not consciousness, not neural fidelity**| |Behavioral digital twins|ACTIVE RESEARCH / COMMERCIAL|Phenotype layer only| |Lifelogging / multimodal personal archives|OPERATING TECHNOLOGY|Autobiographical and provenance layer external to neural state| |Voice / linguistic style models|OPERATING|Expressive phenotype; prosthetic personalization| |Embodied avatars and robotics|ACTIVE|Sensorimotor embodiment for externalized models| --- ## Appendix B — Disposition of the original ninety Published as a ledger rather than implemented silently. **KEEP** items survive with the stated correction; **PROMOTE** items were under-weighted in 2025; **RETIRE** items had no primary evidentiary basis or were mine as conceptual constructs and should not have carried numbered technical authority. |#|2025 item|Action|Disposition| |---|---|---|---| |1|Ultra-high-field MRI (7T+)|KEEP / CORRECT|Macro-mesoscale human imaging; explicitly not synapse-complete, does not resolve microtubules| |2|Diffusion Tensor Imaging|KEEP / CORRECT|Probabilistic macro-connectivity; does not trace every connection| |3|Functional MRI|KEEP / EXPAND|Add laminar fMRI, encoding/decoding models, NeuroSTORM, semantic decoding| |4|Molecular MRI|KEEP / NARROW|Requires named probes and human-vs-animal status| |5|Hyperpolarized MRI|KEEP / NARROW|Metabolic imaging; not a consciousness-state capture technology| |6|MRI-guided focused ultrasound|PROMOTE|Recast inside the full ultrasound stack (§IX)| |7|Resting-state fMRI|KEEP|Network inference; correlation ≠ causal mapping| |8|MRI with machine learning|PROMOTE / RENAME|→ neural foundation models, cross-subject transfer, in-silico individualized models| |9|BOLD imaging|KEEP / CONTEXTUALIZE|Hemodynamic proxy with stated limits| |10|Magnetic resonance spectroscopy|KEEP / EXPAND|Metabolic and molecular state layers| |11|Brain-computer interfaces (generic)|EXPLODE|→ competing access geometries (§IX, A.8)| |12|Cryonics and brain preservation|PROMOTE / CORRECT|→ ASC history + 2026 ABC whole-human protocol; "atomic precision" retired| |13|Whole-brain emulation (Blue Brain)|REBUILD|→ connectome-constrained models, MICrONS, foundation models, embodiment (§VI)| |14|Quantum computing for brain simulation|**DEMOTE**|Not an established dependency. Retained only as speculative substrate| |15|Optogenetics|PROMOTE|+ all-optical interrogation, holographic stimulation, closed-loop causal perturbation| |16|Nanotechnology for neural interfacing|PROMOTE HEAVILY|→ N3 nanotransducers, MENPs, plasmonics, BBB transport, Subsense, ARIA (§IX)| |17|AI modeling neural architectures|PROMOTE HEAVILY|→ foundation models, connectome-constrained nets, HIPPIE, semantic decoders| |18|Connectomics (HCP)|PROMOTE HEAVILY|→ FlyWire, BANC, MaleCNS, H01, MICrONS, BRAIN CONNECTS, LICONN (§IV)| |19|Synthetic biology for artificial neurons|REBUILD|→ biohybrid interfaces, organoids, assembloids, living electrodes| |20|Electrophysiology (EEG/ECoG)|EXPLODE|→ separate EEG, OPM-MEG, ECoG/iEEG, Neuropixels, HD-MEA, calcium/voltage imaging| |21|Neuromorphic computing|KEEP / UPDATE|Loihi 2/Hala Point, SpiNNaker2, BrainScaleS 2, NorthPole, Akida, Innatera| |22|Brain organoids|PROMOTE HEAVILY|→ Brainoware, FinalSpark, CL1, NUS 20-unit rack (§X)| |23|Photonic neural networks|KEEP / REFRAME|Execution and interconnect substrate; not evidence of transfer| |24|Swarm robotics|RELOCATE|Embodiment and multi-agent infrastructure only| |25|Epigenetic mapping and editing|PROMOTE AS FIDELITY LAYER|+ transcriptomics, chromatin, local RNA, phosphorylation, proteostasis (A.4)| |26|Holographic data storage|KEEP AS STORAGE|Archival substrate only| |27|DNA data storage|KEEP AS STORAGE|Persistence ≠ executable cognition| |28|Magnetic nanoparticle neural control|PROMOTE HEAVILY|→ documented MENP/MEnT lineage, MOANA, magnetogenetics, Subsense| |29|Closed-loop neurofeedback|PROMOTE HEAVILY|→ central; double neural bypass, adaptive BCI, read-write neuroprostheses (§VIII)| |30|Biohybrid neuro-AI interfaces|PROMOTE HEAVILY|→ Science Biohybrid, organoid-BCIs, living electrodes| |31|Quantum entanglement communication|**REMOVE**|Does not transmit usable information instantaneously without a classical channel| |32|Digital twin simulations|SPLIT|Neural/physiological twins vs behavioural generative agents — different observables| |33|Hive mind networks|RENAME / HISTORICIZE|→ documented brain-to-brain experiments, collaborative BCIs; interhuman neural bypass| |34|Neural dust expansion|KEEP AS PRECEDENT|Miniaturized wireless lineage; not deployment-ready| |35|Neuroprosthetic augmentation|PROMOTE HEAVILY|→ BrainGate2 home use, brain-to-voice, Paradromics, Synchron, Precision, CorTec| |36|Brain-on-a-chip platforms|REBUILD|→ organoid intelligence, HD-MEA cultures, CL1, FinalSpark, assembloids| |37|Exocortex development|KEEP AS CONCEPT|Ground in BCI-to-computer control, OS-level neural input, external memory| |38|Blockchain for consciousness data|**REMOVE FROM CORE**|Not a technical prerequisite; optional provenance/rights infrastructure| |39|Neuroplasticity induction|PROMOTE|Ground in closed-loop stimulation, rehabilitation, adaptive decoders, engram work| |40|Ethical AI governance frameworks|PROMOTE / RELOCATE|→ neurorights, neural data governance, UNESCO, ISO/IEC and IEEE BCI standards| |41|Cortical stacks via embedded nanobots|**RETIRE**|No source establishes DARPA-produced synaptic-resolution recording nanobots| |42|Memristive synaptic arrays|KEEP / UPDATE|+ diffusive memristors, phase-change memory, compute-in-memory| |43|Neural lace, graphene-hybrid meshes|REPLACE|→ Precision Layer 7, INBRAIN graphene, flexible ECoG| |44|Femtosecond laser optogenetics|RENAME / VERIFY|→ all-optical interrogation and holographic optogenetics with sourced specifications| |45|Cryo-electron tomography connectomics|KEEP / CORRECT|Local molecular and subcellular imaging; **not a whole-brain method**| |46|Self-modeling AI architectures|REBUILD|→ individualized neural models, generative agents, autobiographical architectures| |47|Mitochondrial bioengineering|DEMOTE|Living-substrate maintenance; not a central transfer technology| |48|Topological qubit brain simulations|**REMOVE AS DEPENDENCY**|Not established as necessary for anything in this stack| |49|Glial cell interface systems|PROMOTE HEAVILY|→ astroengrams, astrocytic ensembles, oligodendrocytes, microglia, vascular cells| |50|Biophotonic neural interfaces|KEEP / MAKE CONCRETE|→ DOT, fNIRS, NIR plasmonic sensing, calcium/voltage imaging| |51|Synthetic mRNA neuroplasticity enhancers (Japan/Switzerland)|**RETIRE**|Geography-tagged, unsourced as written| |52|CRISPR-activated neural substrates (South Korea)|REPLACE|→ NIH BRAIN precision cell-access tools, enhancer-based targeting, named vectors| |53|Quantum dot optogenetic probes (China)|OPTIONAL|Only with specific peer-reviewed systems; no country label| |54|Mycelium-based neural networks (Slovenia)|REMOVE FROM CORE|Unconventional computation; does not strengthen the case| |55|Holographic optogenetics (France)|KEEP / CONSOLIDATE|Folded into all-optical/holographic optogenetics with primary sources| |56|Neuroimmunomodulation interfaces (Israel)|OPTIONAL|State-fidelity layer, not the transfer proof chain| |57|DNA nanobots for synaptic mapping|**RETIRE**|No named primary demonstration| |58|Magnetoelectric nanoparticle gene delivery (Germany)|CORRECT / PROMOTE|→ actual MENP neural-interface lineage; separate stimulation, localization, delivery, recording| |59|AI-optimized neuropharmaceutical cocktails (Canada)|**RETIRE**|Replace only with named computational neuropharmacology platforms| |60|Electroceutical vagal interfaces (Austria)|REFRAME|Closed-loop bioelectronic medicine; peripheral to transfer| |61|Neural entanglement via quantum dots|**REMOVE**|No replicated neural-interface demonstration| |62|4D bioprinted neural networks|REFRAME|→ neural tissue engineering, vascularized organoids, assembloids| |63|Microbiome-gut-brain modulation|RELOCATE|Biological context; retained only under complete organismal state capture| |64|Holographic neural avatars (South Korea)|REPLACE|→ digital humans, embodied avatars, externalized phenotype models| |65|Cortical WiFi via terahertz waves (Israel)|**RETIRE**|No peer-reviewed neural-interface evidence| |66|Neuro-symbolic AI integration|KEEP AS AI SUPPORT|Modeling/representation; not a capture technology| |67|Plasmonic nano-imprinting (Australia)|CORRECT / PROMOTE|→ plasmonic neural sensing, gold nanoshell lineage, NIR scattering, Subsense readout| |68|Blood-brain barrier engineering|PROMOTE|Central to distributed molecular interfaces (A.9)| |69|Dark matter neural sensors|**RETIRE**|No demonstrated neuroscience use case| |70|Consciousness validation Turing protocols|REBUILD AS OPEN PROBLEM|No such accepted standard; → validation criteria in A.12| |71|Neutrino networking (N3-UbiqNet)|**RETIRE**|No verifiable primary evidence| |72|MOANA tri-modal non-invasive BMI|PROMOTE HEAVILY|Retained as named N3 lineage → Motif → ARIA Brain Mesh (§IX)| |73|Global SuperGrid Human-Node Architecture|**RETIRE AS NAMED TECHNOLOGY**|Conceptual systems model of mine; belongs in speculative essay| |74|Phase-Dynamic Harmonic Signal Lattice|**RETIRE AS NAMED TECHNOLOGY**|Conceptual construct of mine| |75|Photonic Computational Connectomes|**RETIRE / REPLACE**|Fused field name; → real photonic computing + connectome-constrained models| |76|BIOE-Driven Organoid Autonomy Modules|**RETIRE / REPLACE**|→ Brainoware, CL1, FinalSpark, organoid-BCIs| |77|Neural Terraforming Nanolithography|**RETIRE / REPLACE**|→ nanofabricated interfaces, CMOS-MEMS, thin films, graphene, microLEDs| |78|Biocomputational Cognitive Operating Systems|**RETIRE / REPLACE**|→ Cortical Labs biOS / Cortical Cloud, FinalSpark APIs| |79|Reflexive Field-Intelligence Sensor Mesh|**RETIRE / REPLACE**|→ documented multimodal sensor fusion, ambient sensing| |80|Neuro-Electromagnetic Field Entrainment Interfaces|**RETIRE / REPLACE**|→ TMS, temporal interference, focused ultrasound, MENP transduction| |81|Ambient BCI|KEEP / GROUND|HCI direction grounded in wearable EEG/EMG/OPM-MEG/fNIRS, not field coupling| |82|Cognitive-responsive environments|KEEP AS INTERFACE CONTEXT|External loop; not direct consciousness capture| |83|Atmospheric data field interfaces|**RETIRE FROM EVIDENTIARY CORE**|Move to speculative essay absent neural sensing/modulation evidence| |84|Phase-dynamic environmental computing|**RETIRE AS NAMED TECHNOLOGY**|No independent primary sources| |85|Electromagnetic field entrainment (NEFEI, dup.)|**RETIRE DUPLICATE**|Consolidate legitimate neuromodulation under named systems| |86|Embedded bio-sensor networks|REBUILD|→ verified wearable/implantable physiology, closed-loop bioelectronics| |87|Quantum computing for biomedical simulation|DEMOTE|Adjacent capability; not a transfer dependency| |88|Federal research ecosystem integration|RELOCATE|Not a technology → funding architecture: NIH BRAIN, BICAN, CONNECTS, DARPA, ARIA| |89|Bell Labs legacy technologies|MOVE TO HISTORY|Transistor, laser, information theory belong in lineage, not the current stack| |90|Atmospheric Wi-Fi field networks|**REMOVE FROM CORE**|Not a demonstrated consciousness-transfer technology| --- ## Appendix C — Relationship map, with typed predicates The predicates are not interchangeable, and the 2025 article's central inferential failure was allowing adjacency to read as coordination. **Lineage** means one program or organization descends from another. **Personnel** means individuals moved. **Partnership** and **funding** mean a documented commercial or financial relationship. **Advisor** means a named advisory role. **Technical convergence** means shared mechanism with no documented organizational connection. **Platform** means one party's system is a supported target of another's. Nothing in this table should be read across predicate boundaries. |Relationship|Predicate|Status|What it explains| |---|---|---|---| |OpenAI ↔ Merge Labs|FUNDING + STATED COLLABORATION|Verified|OpenAI invested in the Jan 2026 seed and states collaboration on scientific foundation models and frontier tools| |Merge Labs ↔ Butterfly Network|COMMERCIAL LICENSE|Verified, 1 Sep 2026|Exclusive CMOS-MEMS ultrasound-on-chip supply under Butterfly Embedded; upfront, milestones, hardware commitments, royalties| |Butterfly ↔ Forest Neurotech|PRIOR PARTNERSHIP|Verified, 2023|~$20M five-year collaboration; antecedent of the Merge agreement| |Forest Neurotech → Arbor Neuroscience → Merge|LINEAGE + PERSONNEL|Verified|Norman and Aflalo bridge the ultrasound FRO lineage into Merge. **Organizational succession ≠ IP transfer**| |Mikhail Shapiro ↔ Forest/Arbor ↔ Merge|PERSONNEL + SCIENTIFIC LINEAGE|Verified|Biomolecular ultrasound and acoustic reporter genes as the scientific bridge| |Rice / Robinson → DARPA N3 MOANA → Motif → ARIA Brain Mesh|PROGRAM + PERSONNEL LINEAGE|Verified|Cleanest government-to-academic-to-startup-to-new-program chain in the ecosystem| |DARPA NESD → Paradromics / Matt Angle|PROGRAM LINEAGE|Verified|Neural Input-Output Bus was a NESD performer architecture| |DARPA N3 → six modality branches|PROGRAM TREE|Verified|Battelle, CMU, JHU APL, PARC, Rice, Teledyne across EM, optical, acoustic, magnetic paths| |ARIA ↔ MintNeuro|FUNDING + INTEGRATION ROLE|Verified|MintNeuro leads low-power ASIC integration across three funded projects incl. Brain Mesh| |ARIA ↔ Forest Neurotech / NHS / Plymouth|FUNDING|Verified|Forest 1 whole-brain ultrasound interface at skull-defect site| |Subsense ↔ DARPA N3 / MOANA|**TECHNICAL CONVERGENCE ONLY**|No program link|Falls inside the N3/MOANA transduction search space. **No evidence of N3 funding or IP descent**| |Subsense ↔ UC Santa Cruz, ETH Zurich|COLLABORATION|Verified|Named research collaborations| |Subsense ↔ Ray Kurzweil|ADVISOR|**VERIFIED_CURRENT**|Subsense's current About page lists him as a product advisor; its September 9, 2026 news page announces him as **Product and Vision Advisor**. This supersedes the stale contested label| |Synchron ↔ Apple|PLATFORM|Verified|Demonstrated Apple-platform BCI interaction| |Apple BCI HID ↔ vendor-neutral neural input|STANDARD + ANALYTIC CONSEQUENCE|Verified|BCI represented as a host input class, not a one-off lab interface| |Precision ↔ Medtronic|PARTNERSHIP|Verified 2026|Imports surgical navigation, manufacturing, clinical infrastructure| |Paradromics ↔ Michigan / UC Davis / Mass General|CLINICAL NETWORK|Verified|Connect-One translational consortium around permanent high-data-rate BCI| |Paradromics ↔ consumer computing|REGULATORY MILESTONE|Verified Aug 2026|FDA pathway to compatible personal devices without per-model approval| |Science Corp ↔ Khosla / Lightspeed / YC / IQT / Quiet|FINANCING|Verified|$230M Series C, Mar 2026; ~$490M total reported by company| |Science Corp ↔ Yale / Murat Günel|CLINICAL LEADERSHIP|Verified 2026|Medical Director for BCIs as Biohybrid moves toward translation| |Cortical Labs ↔ DayOne ↔ NUS Medicine|DEPLOYMENT PARTNERSHIP|Verified Aug 2026|Wetware hardware + commercial DC operator + academic neurobiology in one rack-scale environment| |Allen ↔ Baylor ↔ Princeton ↔ MICrONS|CONSORTIUM|Verified|Mammalian structure-function connectomics| |Janelia ↔ Google Research ↔ Cambridge Connectomics|CONSORTIUM|Verified|Decade-long pipeline producing FlyWire, BANC, and the 2026 male CNS connectome| |ISTA ↔ Google Research (LICONN)|RESEARCH COLLABORATION|Verified|Danzl/Tavakoli with Januszewski and Jain| |NWB ↔ DANDI ↔ NIH BRAIN ↔ Allen/MICrONS|DATA INFRASTRUCTURE|Verified|Converts experiments into reusable, versioned, programmatically accessible data| |BICAN ↔ BRAIN CONNECTS ↔ Armamentarium|FEDERAL ARCHITECTURE|Verified|Parts list, connectivity across scales, and precise experimental access| |OpenAI Stargate ↔ large-scale AI compute|INFRASTRUCTURE CONTEXT|Verified|Execution infrastructure for large models. **Not evidence that it hosts reconstructed minds**| --- --- [[about/About Bryant McGill|Bryant McGill]] is a Wall Street Journal and USA Today bestselling author, systems architect, technologist, and strategic advisor, as well as a Congressionally Recognized Ambassador of Goodwill and United Nations–appointed Global Champion. His work spans naval intelligence systems, computational linguistics, artificial intelligence, digital transformation, and civilizational governance architecture. His forward analysis on U.S.–Israel Pax Silica frameworks has appeared in Jewish/Jerusalem News Syndicate (JNS). --- ## References ### Connectomics and reference maps - [Sexual dimorphism in the complete _Drosophila_ male central nervous system connectome](https://www.sci.news/biology/complete-fruit-fly-connectome-15053.html) — Berg, Beckett, Costa et al., _Cell_, September 3, 2026. - [A connectomics milestone: Mapping the complete male fruit fly brain](https://research.google/blog/a-connectomics-milestone-mapping-the-complete-male-fruit-fly-brain/) — Januszewski and Jain, Google Research, 2026. - [Distributed control circuits across a brain-and-cord connectome](https://www.nature.com/articles/s41586-026-10735-w) — _Nature_, 2026 (the BANC brain-and-nerve-cord connectome). - [Functional connectomics spanning multiple areas of mouse visual cortex](https://www.nature.com/articles/s41586-025-08790-w) — MICrONS Consortium, _Nature_ 640, 2025. - [Cubic Millimeter dataset overview](https://www.microns-explorer.org/cortical-mm3) — MICrONS Explorer. - [Scientists complete largest wiring diagram and functional map of the brain to date](https://www.sciencedaily.com/releases/2025/04/250409114838.htm) — Allen Institute, 2025. - [Light-microscopy-based connectomic reconstruction of mammalian brain tissue](https://www.nature.com/articles/s41586-025-08985-1) — Tavakoli, Lyudchik, Januszewski, Jain, Danzl et al., _Nature_ 642, 2025 (LICONN). - [Piecing together the brain puzzle](https://www.sciencedaily.com/releases/2025/05/250507125852.htm) — ISTA on LICONN. ### Neural foundation models and executable twins - [Foundation model of neural activity predicts response to new stimulus types](https://www.nature.com/articles/s41586-025-08829-y) — Wang, Fahey, Ding, Tolias et al., _Nature_ 640, 2025. - [Author Correction: Foundation model of neural activity predicts response to new stimulus types](https://www.nature.com/articles/s41586-026-10457-z) — _Nature_ 652, 2026. - [HIPPIE: a generative model for electrophysiological analysis across species, technologies, and modalities](https://www.nature.com/articles/s41467-026-76939-w) — Gonzalez-Ferrer, Lehrer, Alvarez-Esteban et al., _Nature Communications_, 2026. ### Human interfaces, read and write - [Long-term independent use of an intracortical brain–computer interface for speech and cursor control](https://www.nature.com/articles/s41591-026-04414-6) — Card, Brandman et al., _Nature Medicine_ 32, 2026. - [Brain-computer interface enables independent, accurate communication for man living with ALS](https://health.ucdavis.edu/news/headlines/brain-computer-interface-enables-independent-accurate-communication-for-man-living-with-als/2026/06) — UC Davis Health, 2026. - [A neuroprosthesis for restoring hand movement and sensation in a person with complete tetraplegia](https://www.nature.com/articles/s41591-026-04498-0) — Bouton et al., _Nature Medicine_ 32, 2026 (double neural bypass). - [Feinstein Institutes technology that restores movement and sensation after paralysis featured in Nature Medicine](https://www.biospace.com/press-releases/feinstein-institutes-technology-that-restores-movement-and-sensation-after-paralysis-featured-in-nature-medicine) — Feinstein Institutes, 2026. - [Paradromics receives FDA approval to expand BCI access to personal computing devices in the Connect-One clinical study](https://www.medicaleconomics.com/view/fda-expands-device-compatibility-for-paradromics-brain-computer-interface) — reported August 2026. - [Paradromics and University of Michigan complete first Connexus BCI implantation](https://paradromics.com/news/paradromics-completes-first-human-brain-computer-interface-bci-implantation/) — Paradromics, June 2026. ### Nonsurgical, molecular, and ultrasound access - [Six paths to the nonsurgical future of brain-machine interfaces](https://www.darpa.mil/news/2019/nonsurgical-brain-machine-interfaces) — DARPA N3. - [Precision Neurotechnologies funded projects](https://aria.org.uk/opportunity-spaces/scalable-neural-interfaces/precision-neurotechnologies/funded-projects) — ARIA. - [Exploring new frontiers in neurotech](https://www.aria.org.uk/insights/exploring-new-frontiers-in-neurotech/) — Jacques Carolan, ARIA, on Brain Mesh and biological interfaces. - [Funding world-leading research into new neurotechnologies](https://aria.org.uk/insights/funding-world-leading-research-into-new-neurotechnologies) — ARIA, including Forest 1. - [Butterfly Network licenses ultrasound-on-chip technology to Merge Labs for brain-computer interfaces](https://www.stocktitan.net/news/BFLY/merge-labs-and-butterfly-network-partner-to-advance-ultrasound-based-h1d0ix3tmo7p.html) — September 1, 2026. - [Butterfly Network, Merge Labs partner on ultrasound brain-computer interfaces](https://www.mobihealthnews.com/news/butterfly-network-merge-labs-partner-ultrasound-brain-computer-interfaces) — MobiHealthNews, 2026. - [Subsense is developing inhalable BCI nanoparticles for brain signal sensing and neurostimulation](https://www.medicaldesignandoutsourcing.com/subsense-brain-computer-interface-nanoparticle-inhalable/) — Medical Design & Outsourcing. ### Living substrates - [NUS Medicine, DayOne and Cortical Labs unveil biological data center prototype in Singapore](https://medicine.nus.edu.sg/news/nus-medicine-dayone-and-cortical-labs-unveil-biological-data-center-prototype-in-singapore/) — NUS Medicine, August 2026. - [DayOne partners with Cortical Labs and NUS Medicine for Singapore's first biological data center prototype](https://www.datacenterdynamics.com/en/news/dayone-partners-with-cortical-labs-nus-medicine-for-deployment-of-singapores-first-biological-data-center-prototype/) — Data Center Dynamics, 2026. - [Science Corporation Biohybrid](https://science.xyz/technologies/biohybrid) — Science Corporation. ### Preservation and memory substrate - [Cryopreservation of aldehyde-fixed whole brains](https://pubmed.ncbi.nlm.nih.gov/42636208/) — McKenzie, Garrood, Keberle, Slaughter et al., _PLOS One_, August 2026. - [Aldehyde-based cryopreservation of whole brains](https://www.biorxiv.org/content/10.64898/2026.03.02.708967v1) — preprint, March 2026. - [What are memories made of? A survey of neuroscientists on the structural basis of long-term memory](https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0326920) — Zeleznikow-Johnston, Kendziorra, McKenzie, _PLOS One_ 20(6), 2025. - [Structural brain preservation: a potential bridge to future medical technologies](https://www.frontiersin.org/journals/medical-technology/articles/10.3389/fmedt.2024.1400615/full) — _Frontiers in Medical Technology_, 2024. ### Reusable priors: genomics, cells, and adaptation - [Human genomic variation](https://www.genome.gov/about-genomics/educational-resources/fact-sheets/human-genomic-variation) — NHGRI, on the 99.6 percent figure and the pangenome. - [A draft human pangenome reference](https://www.nature.com/articles/s41586-023-05896-x) — Human Pangenome Reference Consortium, _Nature_, 2023. - [CRAM: the genomics compression standard](https://www.ga4gh.org/news_item/cram-compression-for-genomics/) — GA4GH. - [CRAM 3.1: advances in the CRAM file format](https://academic.oup.com/bioinformatics/article/38/6/1497/6499262) — Bonfield, _Bioinformatics_, 2022. - [Genome modelling and design across all domains of life with Evo 2](https://www.nature.com/articles/s41586-026-10176-5) — Arc Institute and NVIDIA, _Nature_, 2026. - [AlphaGenome Atlas: molecular predictions for 9 billion human DNA variants](https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome/) — Google DeepMind, September 8, 2026. - [AlphaGenome Atlas maps 9 billion possible DNA variants](https://spectrum.ieee.org/alphagenome-atlas) — _IEEE Spectrum_, 2026. - [scGPT: toward building a foundation model for single-cell multi-omics using generative AI](https://www.nature.com/articles/s41592-024-02201-0) — _Nature Methods_, 2024. - [Large-scale foundation model on single-cell transcriptomics](https://www.nature.com/articles/s41592-024-02305-7) — scFoundation, _Nature Methods_, 2024. - [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) — Hu, Shen et al., 2021. - [BRAIN Initiative Cell Atlas Network](https://www.nih.gov/brain/research/tools-technologies-brain-cells-circuits/brain-initiative-cell-atlas-network) — NIH. ### Externalized identity - [LLM agents grounded in self-reports enable general-purpose simulation of individuals](https://www.alphaxiv.org/abs/2411.10109) — Park, Zou, Liang, Willer, Bernstein et al. - [AI agents simulate 1,052 individuals' personalities with impressive accuracy](https://hai.stanford.edu/news/ai-agents-simulate-1052-individuals-personalities-impressive-accuracy) — Stanford HAI. ### Standards and archives - [ISO/IEC 27572:2026 — Information technology, brain-computer interfaces, reference architecture](https://www.iso.org/standard/71680.html) — ISO/IEC JTC 1/SC 43. - [ISO/IEC TS 27571:2026 — Data format for noninvasive brain information collection](https://standards.iteh.ai/catalog/standards/iso/94d7f04b-a7eb-4194-ba78-5116c4b63195/iso-iec-ts-27571-2026) — ISO/IEC. - [ISO/IEC TR 27599:2025 — Brain-computer interfaces, use cases](https://committee.iso.org/standard/80419.html) — ISO/IEC. - [Brain-computer interfaces subcommittee](https://jtc1info.org/technology/subcommittees/brain-computer-interfaces/) — ISO/IEC JTC 1/SC 43. - [DANDI: Distributed Archives for Neurophysiology Data Integration](https://www.braininitiative.org/toolmakers/resources/dandi/) — NIH BRAIN Initiative. - [Facilitating analysis of open neurophysiology data on the DANDI Archive](https://www.nature.com/articles/s41597-025-06285-x) — _Scientific Data_, 2025. - [Brain-computer interface HID reference for connecting to Apple platforms](https://developer.apple.com/documentation/accessibility/brain-computer-interface-hid-reference-for-connecting-to-apple-platforms) — Apple Developer.