Abstract
This page is a long-form research note aimed at turning mind uploading (WBE) into something that can be evaluated as real research. It is organized around four technical questions: what can be observed directly, what remains latent, what can be perturbed or calibrated, and what has to stay stable during implementation. It treats recent results, failures, and limits with the same weight so that the boundary between scaffold progress and state-complete claims remains explicit.
Start Here: Verification Commons
The core of Mind-Upload is to define progress up front and build a verification commons that lets different results be compared on the same basis.
Open Verification Commons ->This is a long research note. It is easier to navigate if you first read WBE 101 and EEG 101, and then return here with the Glossary if needed.
If you want only the technology and natural-science side first, start with Wiki: Observability and Claim Ceilings by Measurement Stack, Wiki: From Observation to Estimation, and Wiki: Homeostatic Plasticity and Maintenance State. If the limits of EEG are the issue, go to Wiki: EEG Preprocessing and QC. For the bridge from measurement terms to modeling terms, use Wiki: Measurement and Modeling Terms. For multimodal integration, go next to Wiki: Human Proxy Composition and Route Maturity and Wiki: State, Trait, and Drift. For implementation burdens, use Wiki: Closed Loops, Latency, Jitter, and Safety Stops and Wiki: Thermodynamic Grounding Basics.
Perspective is the long technical notebook that keeps structural scaffold gains, hidden maintenance-state burdens, human observability ceilings, and implementation objections on one axis. If you want to align claim levels first, start with WBE 101. If you want the dependency map, use the technology roadmap. If you want a one-page guide to the remaining theory-specific routes, see Wiki: Guide to reading theory pages.
This page is a long-form research note that lists observations and limitations; it is not primarily a task list or a proposal list. If you want a one-page guide to the differences between facts, hypotheses, proposals, and execution tasks, please see Wiki: Difference between fact, hypothesis, proposal, and execution task.
If you want to see why the reading path moves from WBE 101 to this long note, including common misunderstandings and the shift toward design principles, please see Wiki: 4 routes to deepen the theory from WBE introduction.
This page covers many limitations and counterarguments, so you need to distinguish clearly between partial solutions, exploration-stage results, and undeveloped areas. If you want a one-page guide to those progress labels, please read Wiki: How to read partial solution / exploration stage / undeveloped first.
Things to check before reading
- Role of this page: This page does not simply list favorable positions; it tracks evidence and limitations at the same time.
- What you can say on this page: You can sort out which measurements, latent states, and implementation ceilings are the current technical bottlenecks.
- Things that can't be said on this page alone: This page does not justify state-complete observation, route-free controller identity, or a sufficient-condition claim for WBE.
First read the design principles section on state-space closure, then move into the introduction, measurement, modeling, and implementation sections. If the terminology becomes the blocker, step out briefly to the Glossary or Wiki: Measurement and Modeling Terms and then return.
| Issues that interest me now | What to read first | What you can find out there |
|---|---|---|
| I want to know which technical bottleneck bites first. | Design Principles | You can see why structural progress, hidden maintenance state, and human observability are not the same claim. |
| I want to see the overall design from measurement to implementation. | Technical Framework | You can follow the assumptions that connect the three stages of measurement, decoding, and implementation. |
| I want to know what is missing between decode and emulate | Decoding to Emulation Gap | Correlation-based readouts reveal gaps to proceed to causal verification. |
| I want to know what kind of research plan I will fall into. | Research Program / EEG Consciousness Roadmap | You can check what artifact classes, validations, and stop rules the site expects you to accumulate. |
| I want to know the limitations of this page first. | Limitations | You can separate and read the observability limits, engineering limits, and operational limits. |
| What can be said relatively strongly on this page | What this page still treats as a hypothesis |
|---|---|
| Structural atlas progress is real, but it is not route-free current-state readout. | Which structural scaffold gains actually shrink latent-state uncertainty enough for stronger emulation benchmarks is still unresolved. |
| EEG alone has fundamental limitations in inverse problems and spatial resolution. | The extent to which this limitation can be overcome by integrating other modalities is still a research topic. |
| Living-human evidence remains proxy-based even when several routes are combined. | Which multimodal bundles can be composed without shared-driver or model-burden confounding is still unresolved. |
| Strong claims that exclude causal perturbations, retestability, and thermodynamic constraints are dangerous. | Which threshold is considered "sufficient" depends on future bench design. |
Design Principles: Start from state-space closure
This section is the technical front door of the page. The question it asks is not which label sounds deepest, but which variables can be observed directly, which remain latent, which can be causally perturbed, and which must stay stable over time for any stronger emulation claim to survive. Theory still appears later on this page, but only after those empirical gates are fixed.
Design principles adopted
- Structural scaffold is necessary but not sufficient: Better atlases and connectomes improve priors, not route-free current-state readout.
- Hidden maintenance variables are first-class causal objects: Transcriptional, glial, vascular, extracellular, and metabolic state cannot be treated as optional residuals.
- Human proxy bundles need explicit composition rules: Shared-driver audit, calibrator role, time window, and model burden must stay visible before claims are strengthened.
- Benchmarks must test perturbation, drift, and energetic burden: Decode accuracy alone is not an implementation criterion.
| What to fix first in the design | The position of this page | Things still unresolved |
|---|---|---|
| Structural scaffold | Local ultrastructure, transcriptomic atlases, and connectomics should be read as scaffold gains that tighten priors and registration, not as complete access to current whole-brain state. | How much those scaffold gains reduce latent-state uncertainty before perturbation becomes the dominant gate remains unsettled. |
| Maintenance controllers | Transcriptional programs, astrocyte state, vascular and clearance controllers, extracellular geometry, and metabolic support have to be treated as causal variables when memory stability or adaptive function is claimed. | Which variable families and time-resolution targets are minimally sufficient for falsifiable emulation benchmarks remains unsettled. |
| Human observability | Living-human PET, MRI, MRSI, EEG, and clearance routes stay split by direct observable, calibrator role, spatial unit, and physiology-side burden; they are not automatically one state meter. | Which multimodal bundles can be calibrated in the same subject without shared-driver or model-burden confounding is still unresolved. |
| Implementation gates | Closed-loop and emulation claims must carry perturbation response, retest stability, drift monitoring, and energetic-cost audits alongside decoding performance. | Which benchmark thresholds should count as a meaningful failure or success condition remains a bench-design problem. |
Recent primary papers sharpen three different frontiers rather than one unified ladder. Whole-brain cell-type atlas work (Yao et al. (2023)), local human nanoscale reconstruction (Shapson-Coe et al. (2024)), and morphology-linked predicted transcriptomic connectomics (Gamlin et al. (2025)) improve structural scaffolds. But causal work on thalamocortical transcriptional programs (Terceros et al. (2026)), multiday astrocytic stabilization (Dewa et al. (2025)), and astrocyte-supported amygdala representations (Bukalo et al. (2026)) show that stabilization and retrieval depend on dynamic non-neuronal and transcriptional state. On the living-human side, CSF mobility and biomarker-efflux work (Hirschler et al. (2025); Dagum et al. (2026)) tighten bounded observability, not route-free maintenance-controller identity. That is why this page now puts state-space closure ahead of theory labels.
A wiring diagram helps you locate parts of a machine, but it does not tell you the current buffer occupancy, controller setting, or heat load. WBE faces the same problem. Structure matters, but the stronger failure mode is hidden dynamic state.
Measurements will require BIDS, synchronization, QC, perturbation logs, and route-specific calibrators. Modeling will require explicit uncertainty management for inverse problems, latent confounders, and non-equivalent multimodal rows. Implementation will require closed-loop stability, drift checks, and thermodynamic-cost audits. The correct way to use this section is therefore not as a slogan, but as the gate that binds the requirements of the later sections.
Introduction: Theoretical Foundations Revisited
1.1. Science of consciousness: Reexamining the theoretical foundations and the impact of “Adversarial Collaboration”
The point of this section is not to declare a winning theory. Instead, when the major theories are placed side by side, the first question is which verification conditions are common across them.
What is the "consciousness" that mind upload (WBE) should reproduce? This project historically leaned toward a combined reading of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT). However, the pre-registered adversarial collaboration published in 2025 partially supported predictions from IIT, GNWT, and RPT while also falsifying some key predictions, so no single theory emerged as the winner.[54] Candidate markers for conscious content were also separated from markers of report and task relevance, making it clear that confound control must be fixed before theory choice is elevated. Therefore, this site does not commit to a single theory. Instead, it translates theories into pre-registered prediction sets that must compete under no-report conditions, perturbation benchmarks, and external generalization tests. FEP / predictive coding remains one candidate route, not the default one.
| Theory | Main target of explanation | Main weakness / failure point | How to use with Mind-Upload |
|---|---|---|---|
| IIT | Seeks to explain the integrated quality and causal structure of consciousness. | The problem is the Unfolding Argument, which is computationally heavy and can be judged as unconscious even though it is functionally the same. | It is treated as a candidate measuring framework for the quality and integration of consciousness, but in implementation it is reduced to approximate indices such as PCI-family measures. |
| GNWT | Seeks to explain why conscious access emerges when information is shared widely. | It is difficult to separate prefrontal cortex ignition from reporting behavior, and its correspondence with consciousness itself tends to be unstable. | It is used as a working hypothesis for asking under what conditions wide-area sharing and reportability emerge. |
| FEP/Predictive Coding | Seeks to explain how a system can maintain stable inference while interacting with the environment. | It cannot fully explain phenomenal consciousness by itself, and because it is highly abstract, it is easy to appear as if it has explained everything. | It is treated as a candidate model family for describing closed loops, prediction errors, and adaptation, and must compete with DCM, SCM, and state-space models. |
Central issue: How to design prediction competition rather than theoretical preferences
Nature 2025's adversarial test showed that signals related to conscious content are distributed in multiple occipital/temporal/frontal regions, and that some markers also track task relevance and report requirements.[54]. Therefore, rather than deciding which theory to name, it is more scientific to first decide which predictions to make and under which conditions.
Direction of solution: Return theory to the ``prediction source'' rather than the ``winner''
This site treats IIT, GNWT, RPT, and FEP as competing prediction generators. Acceptance depends on whether the resulting predictions pass no-report conditions, perturbation conditions, task-relevance control, and cross-dataset / cross-center generalization.
- Theoretical layer:Each theory registers in advance which spatiotemporal patterns, which condition differences, and which failure conditions it predicts.
- Indicator layer: PCI / PCI-ST, complexity, and criticality are not theories themselves; they are benchmark candidates for operationalizing theoretical predictions.
- Design layer: A theory is not advanced unless it satisfies no-report conditions, artifact management, delay / jitter auditing, and external validation.
In addition, the time requirements and counterfactual virtual equivalence of closed-loop systems are not determined using a single threshold or a single theory, but rather the latency budget and failure mode for each task are first disclosed, and then treated as a verification issue.[76].
1.2. From theory to implementation: technical and philosophical challenges
Challenges of porting IIT to digital infrastructure: IIT 4.0[17] Applying it to WBE leaves the unresolved question of how its axiomatic system could be satisfied on a digital platform. In particular, the axiom of "intrinsic reality"[44] is unlikely to be satisfied in principle by a discrete computing system such as a standard von Neumann computer, which is one form of the Unfolding Argument. In this project, we therefore avoid calling digital emulation an "approximation" of IIT and instead specify a shift toward physical mapping of causal structure using neuromorphic hardware. Alternatively, following Albantakis et al. (2023),[17] we also treat hybrid systems that combine biological neurons with digital elements as candidates for preserving causal power at the physical level rather than merely matching outputs computationally.
Psychological continuity and copying problems:Derek Parfit's Psychological Continuity Theory[4]The basis of identity is continuity of memory and personality. This requires WBE to maintain a "dynamic process" rather than just a static data copy. In response to the ``copying problem (alter ego paradox)'' posed by this theory, we aim to translate thought experiments such as ``stepwise neural replacement'' and ``hybrid brain systems'' into verifiable engineering protocols.
Turning to process philosophy:Whitehead's process philosophy is based on the perspective of viewing consciousness not as a static "thing" but as a "process" that is constantly updated through interaction with the environment.[32], Friston's free energy principle[14]/active reasoning[45]It also resonates with me. This project focuses on the technical requirements (e.g. Slow Continuous Mind Uploading)[59].
Operate the following as "minimum guardrails": Attach sources such as primary/review articles to major claims; Distinguish between hypotheses, facts, and value judgments, and include uncertainty; Define evaluation indicators and procedures first, and prioritize reproducibility.
Technical Framework
The technological roadmap for realizing Mind Upload is organized into three stages: "Measurement," "Decoding," and "Implementation." It combines the classic WBE roadmap[8], more recent whole-brain-architecture roadmaps[7], and large-scale simulation plans such as Blue Brain[16].
Earlier versions gave similar weight to elements with very different roles, such as Block-Champagne, Active Inference, PCI, TDA, and thermodynamic logs. When we compare direct EEG source-imaging validation studies, TMS-EEG recommendations, and theoretical papers, the common core is much narrower: transparent reporting, disclosure of forward models and electrode geometry, validation against external criteria, and OOD / perturbation / abstention design. Specific solvers, theories, and thermodynamic indicators are better treated as conditional or exploratory tracks.[78][79][90][100].
| issue | Points that were weak before correction | Reasonable arrangement as of 2026-03 |
|---|---|---|
| EEG source imaging | The name of the algorithm was too prominent. | The core requirements are external validation, uncertainty reporting, and conductivity sensitivity analysis; Block-Champagne and related methods are promising candidates, not the standard by themselves.[78][79][101]. |
| Perturbation / PCI | I was writing PCI close to the ground truth. | PCI / PCI-ST is a strong external benchmark candidate, but it is not a universal KPI because TMS-EEG requires strict stimulus control and artifact management.[90][100]. |
| Active Inference / Counterfactual Equivalence | It was written in a way that read like a verified central indicator. | At present, we use it as a theory-driven source of hypotheses and model families, then narrow it through OOD generalization, intervention, and model competition.[76][80]. |
| Criticality / TDA / Irreversibility | It was previously treated too much like a core indicator. | Complexity and criticality are promising, but for now they remain auxiliary analyses while primary judgments rely on indicator bundles that are easier to audit.[52][56][92]. |
The weakness of this central page was that atlas data, patch-seq, EM connectomics, same-brain structure-function bridges, and neuromodulator / glia imaging could all be collapsed into the single phrase "advanced by multimodal." The primary literature makes clear that each stack observes different variables directly, so each stack also has a different upper limit on what can be claimed. If we blur those differences, advancing a structural scaffold, strengthening a cell-type prior, improving local conditional prediction, and calibrating slow-state proxies all start to look like the same kind of progress when they are not.
| measurement stack | what is directly observed | a stronger argument | What remains latent | ceiling on this page |
|---|---|---|---|---|
| EEG / MEG / fMRI / fNIRS | These are macro current fields, hemodynamic proxies, and wide-area state transitions. | Global state tracking, timing constraints, relatively coarse network occupancy, and for hemodynamic stacks the need for vascular transfer / CVR calibration can be audited. | Cell type, current synaptic efficacy, vascular transfer state, neuromodulatory field, glial substrate-routing, and astrocyte-state are not directly determined. | Macro-state tracking, hemodynamic-limited inference, and weak L2. It is not listed as state-complete reconstruction. |
| whole-brain spatial atlas | Yao et al. (2023) substantially expanded cell-type taxonomy and spatial organization. | Molecular identity and spatial prior can be greatly enhanced. | The destinations to return to after threshold/gain, current state, sleep-history, and perturbation remain. | Up to cell-type and spatial priors. It is not presented as dynamic completeness. |
| Patch-seq / morpho-electric bridge | Gouwens et al. (2021) and Gamlin et al. (2025) provide bridges between transcriptomes, morphology, electrophysiology, and local wiring. | The mapping from cell-type labels to morpho-electric phenotypes and local motifs can be enhanced. | Whole-brain coverage, same-brain network context, and longitudinal maintenance-state remain. | Local parameter prior and bridge. It is not used as evidence of whole-brain state completeness. |
| volume EM / same-brain connectomics | Dorkenwald et al. (2024) strengthen the structural scaffold, while MICrONS Consortium et al. (2025) and Ding et al. (2025) strengthen sequential same-brain structure-function links and task-bounded conditional prediction. | Sequential local structure-function scaffolds, connectome-conditioned response prediction under named stimulus regimes, and cell-type dependent wiring rules can be further enhanced. | Same-time whole-brain state, direct transcriptomic identity unless separately assayed, current synaptic efficacy / release state, unique dynamical parameterization, all-state generalization, and long-horizon maintenance-state remain. | Structural scaffold plus sequential local structure-function scaffold and local conditional prediction. It cannot be placed on the same level as human whole-brain WBE, current synaptic-state readout, or a unique local twin. |
| Human diffusion-MRI tractography connectome Thomas et al. (2014); Gajwani et al. (2023); McMaster et al. (2025); Bramati et al. (2026); Zhu et al. (2025) |
Indirect streamline and parcel-graph estimates derived from diffusion orientations under named sampling, endpoint-assignment, filtering, and parcellation choices. | Macro pathway priors, targeted bundle hypotheses, protocol-harmonized macro comparisons, uncertainty-aware tractography, and microscopy-calibrated structural priors can be strengthened. | Synapse-resolved edge truth, directionality, cortical endpoint truth, pipeline-invariant hub topology, current synaptic efficacy, and maintenance-state remain latent. | Acquisition-, endpoint-, graph-construction-, and calibration-conditioned macro pathway prior. It is not presented as a finished living-human connectome or a WBE-ready structural completion claim. |
| human neuromodulatory proxy family | Carro-Domínguez et al. (2025) strengthen mixed arousal gating, Hansen et al. (2022) plus Nakuci & Bansal (2025) strengthen receptor / transporter priors and modeling scaffolds, and Wong et al. (2013) plus Miederer et al. (2025) strengthen selected occupancy versus challenge-linked release routes. | Calibration of arousal proxies, receptor-density priors, administered-drug target engagement, and bounded task- or drug-challenge release proxies can be strengthened. | The sufficiency of the current whole-brain transmitter field, unsampled receptor families, cell-specific downstream effect, and chronic cross-regime neuromodulatory state remains. | Up to family-split neuromodulatory proxy calibration. It is not presented as whole-brain transmitter ground truth. |
| astrocyte / glia state routes | Cahill et al. (2024), Williamson et al. (2025), and Dewa et al. (2025) strengthen local astrocyte-network encoding and multiday stabilization routes. | Local glial-state causality, omission audits for astrocyte-state, and separation of glial from transmitter-only explanations can be strengthened. | Living-human whole-brain astrocyte-state, glial substrate-routing state, and route-free cross-day maintenance-state remain. | Up to local glial-state causality and omission control. It is not presented as a living-human whole-brain glial readout. |
From this point onward, this page treats atlas, bridge, scaffold, sequential same-brain scaffold / local conditional predictor, and proxy calibration as separate categories. In other words, we do not allow readings such as "state-complete because it is multimodal," "whole-brain because it is same-brain," or "long-term maintenance is sufficient because there is local causal evidence." Any claim must be issued together with its measurement stack, direct observables, remaining latent state, and abstention conditions.
Another structural shortcut still remained on this central page: human connectome wording could still slide from diffusion-MRI tractography to a stable graph object. The primary literature does not support that shortcut. Thomas et al. (2014) showed that anatomical accuracy is inherently limited when only voxel-averaged local orientations are available. Gajwani et al. (2023) then compared 1,760 group connectomes and showed that hub location and strength move with tractography algorithm, parcellation, and group-reconstruction choices. McMaster et al. (2025) showed in 44 HCP-YA participants with scan / rescan data that graph measures shift across voxel resolutions and recommended resampling to 1 mm isotropic for robust comparisons. Bramati et al. (2026) then held scanner and preprocessing constant and still showed that common diffusion-sampling schemes move tractography outputs. Finally, Zhu et al. (2025) improved reconstruction by adding microscopy to MRI, which means the gain came from explicit calibration rather than from MRI-only tractography becoming route-free structural truth. Therefore, on this page, a living-human tractography result is read as an acquisition-, endpoint-, graph-construction-, and calibration-conditioned macro pathway prior, or at most a targeted bundle-hypothesis / calibrated bundle-comparison route, not as a finished human connectome. The operating stop line remains Wiki: tractography route card plus Verification: Observability Budget.
Another weakness remained here even after the measurement-stack page was tightened: this page still let same-brain functional connectomics and digital-twin wording sound too close to a solved local twin. The primary literature does not support that shortcut. Bosch et al. (2022) already showed that structure-function bridging is a multistage correlative workflow from in vivo recording to later tomography and EM. MICrONS Consortium et al. (2025) then released a sequential same-brain dataset co-registering in vivo calcium imaging with later EM in one mouse, and Ding et al. (2025) used a validated stimulus-conditioned population-response model to uncover a general wiring rule. But Gamlin et al. (2025) still relied on morphology-linked predicted MET-types rather than direct transcriptomic assay inside the EM volume. And the remaining latent state is not trivial: Holler et al. (2021) showed that inferring function from wiring diagrams is limited by unresolved synaptic-strength structure, Dürst et al. (2022) showed that vesicular release probability strongly sets synaptic strength, Mittermaier et al. (2024) showed that membrane-potential state gates human synaptic consolidation, and Beiran & Litwin-Kumar (2025) showed that connectome-constrained recurrent networks often remain dynamically degenerate until additional recordings narrow the solution space. Therefore, the safe ceiling here is sequential local structure-function scaffold plus task-bounded conditional prediction, not direct current synaptic-state readout, direct transcriptomic truth, or a unique local twin.
One weakness of the earlier version of this page was subtler than a missing citation list: it still let several human support-state routes collapse into one row. The current literature does not support that shortcut. Qian et al. (2012), Fleysher et al. (2013), and Qian et al. (2025) constrain different macro ionic quantity types; Rzechorzek et al. (2022), Rogala et al. (2024), and Tan et al. (2025) constrain passive or task-linked macro thermometry, while Tan et al. (2024) and Inoue et al. (2025) constrain perturbation-conditioned human thermal routes; Hagiwara et al. (2018), Baadsvik et al. (2024), Galbusera et al. (2025), and Colaes et al. (2026) constrain non-equivalent myelin-sensitive versus broader tissue-health-sensitive routes; Morgan et al. (2024), Padrela et al. (2025), and Chung et al. (2025) constrain BBB water-exchange versus tracer-specific transport; Zhao et al. (2020), Sun et al. (2024), Petitclerc et al. (2021), Anderson et al. (2022), Wu et al. (2026), and Petitclerc et al. (2026) constrain non-equivalent choroid-plexus perfusion, blood-to-CSF transport, water-cycling, and apparent-exchange routes; and Villemagne et al. (2022), Villemagne et al. (2022), Tyacke et al. (2018), Livingston et al. (2022), plus Jaisa-Aad et al. (2024) constrain target-defined astrocyte-related PET routes rather than one generic glial-state meter. Therefore, this page now names quantity type and calibrator role explicitly before any human proxy row is read as progress.
| Human route | What it actually advances | What it still does not fix | Safe reading on this page |
|---|---|---|---|
| Local human nanoscale ultrastructure Shapson-Coe et al. (2024) |
Fixed-tissue local organization of human neurons, glia, axons, and synapses at nanoscale resolution. | Living whole-brain dynamics, current synaptic efficacy, maintenance-state, and cross-brain generalization. | Local structural scaffold, not living human state completeness. |
| Regional synaptic-density PET atlas Finnema et al. (2016); Naganawa et al. (2021); Johansen et al. (2024) |
Regional in vivo proxy for synaptic density and its spatial distribution. | Release probability, tagged-spine history, receptor mobility, and branch-local stabilization routes. | Regional synaptic-density proxy, not direct ground truth of current synaptic state. |
| Human receptor / transporter atlas prior Hansen et al. (2022); Nakuci & Bansal (2025) |
Normative chemoarchitectural priors and follow-on modeling scaffolds for selected receptor / transporter systems across healthy human cortex. | Momentary transmitter release, current receptor occupancy during the task, and an individual's time-varying neuromodulatory state. | Regional chemoarchitectural prior / modeling scaffold, not current whole-brain neuromodulatory state. |
| Human occupancy PET Wong et al. (2013); Schlosser et al. (2025) |
Selected receptor-target engagement under an administered drug, named ligand, and bounded scan window, including informative null occupancy. | Endogenous transmitter release, unsampled receptor families, lamina- or cell-specific downstream effect, and continuous whole-brain transmitter state. | Ligand- and dose-limited target-engagement proxy, not endogenous release or direct whole-brain transmitter ground truth. |
| Human displacement / release-sensitive PET Koepp et al. (1998); Erritzoe et al. (2020); Miederer et al. (2025) |
Challenge-linked dopamine or serotonin release proxies under named task, pharmacological challenge, ligand, and time window. | Task-general transmitter state, unsampled receptor families, stable neuromodulatory identity outside the measured window, and cell-specific downstream effect. | Challenge-limited endogenous release proxy, not a free readout of the current whole-brain transmitter field. |
| Whole-brain MRSI metabolic similarity scaffold Lucchetti et al. (2025) |
Gray-matter parcel similarity graph built from five 1H-MRSI metabolite profiles. | Axonal edge-level connectivity, current glucose metabolic rates / ATP turnover, current transcriptional controller, and branch-local energetic reserve. | Macro biochemical similarity scaffold, not tractography, flux imaging, or a local maintenance-state snapshot. |
| High-resolution 1H-MRSI metabolite-distribution route Guo et al. (2025) |
High-resolution ultrahigh-field metabolite maps from 1H-MRSI under explicit reconstruction and artifact-control burden. | Parcel-similarity structure, deuterium absolute quantification, kinetic-rate maps, current transcriptional controller, and branch-local energetic reserve. | High-resolution metabolite-distribution proxy, not biochemical similarity, kinetic-rate imaging, or a local maintenance-state snapshot. |
| Human 31P metabolite / pH balance route Ren et al. (2015) |
Resting ATP-synthesis, phosphorus-metabolite, and pH balance estimates in living human brain. | Model-conditioned exchange flux, whole-brain NAD content, task-evoked NAD dynamics, branch-local mitochondrial reserve, and controller identity. | Resting macro metabolite / pH balance proxy, not whole-brain energetic completeness or branch-local mitochondrial ground truth. |
| Human 31P MT exchange-flux route Ren et al. (2017) |
Model-conditioned PCr→ATP and Pi→ATP exchange-flux estimates from a 7 T magnetization-transfer route. | Whole-brain NAD-content mapping, task-locked NAD dynamics, parcel-similarity structure, direct mitochondrial positioning, and route-independent energetic control. | Model-based macro exchange-flux proxy, not a route-free map of current energetic state. |
| Human 31P NAD-content mapping Guo et al. (2024) |
Whole-brain intracellular NAD-content mapping at 7 T with reproducibility on repeated scans. | Task-evoked local NAD dynamics, ATP exchange-flux, branch-local energetic fragility, and task-general controller identity. | Whole-brain NAD-content proxy, not a direct readout of dynamic local energy control. |
| Localized functional 31P NAD-dynamics route Kaiser et al. (2026) |
Task-evoked NAD+ dynamics in a visual-cortex voxel functionally localized by prior fMRI. | A whole-brain NAD map, route-independent energetic balance, task-general controller identity, and branch-local mitochondrial positioning. | Localized task-locked NAD-dynamics proxy, not a field-ready whole-brain energetic-state readout. |
| Human deuterium metabolite-mapping / absolute quantification Karkouri et al. (2026) |
Absolute HDO / Glc / Glx / Lac maps from a named 7 T deuterium quantification pipeline. | Equivalence to 31P energetic balance, exchange-flux, or NAD routes, a route-free kinetic interpretation, generic dose invariance, and direct mitochondrial-controller identity. | Absolute metabolite-distribution proxy, not a direct branch-local mitochondrial-state measurement. |
| Dynamic deuterium kinetic-rate imaging Li et al. (2025) |
Whole-brain glucose-transport and metabolic-rate maps from a specialized dynamic DMRSI route with blood-input and kinetic modeling. | Route-free high-resolution 1H-MRSI metabolite distributions, equivalence to 31P energetic balance, exchange-flux, or NAD routes, direct mitochondrial-controller identity, and same-subject hidden-state closure. | Macro energetic-rate proxy, not a direct branch-local mitochondrial-state measurement. |
| Quantity-defined human ionic routes Qian et al. (2012); Fleysher et al. (2013); Qian et al. (2025) |
Tissue-sodium mapping, SQ+TQF-derived intracellular sodium / volume fractions, and mono- versus bi-T2 sodium separation in living human brain. | Cell-specific chloride set point, KCC2 / NKCC1 balance, local EGABA, and extracellular K+ microdomains. | Macro ionic proxy family, not direct chloride-homeostasis or inhibitory-regime ground truth. |
| Human passive / task-linked macro thermometry Rzechorzek et al. (2022); Rogala et al. (2024); Tan et al. (2025) |
MRS-based brain-temperature mapping, daily rhythms, task-linked macro thermal shifts, and frontal-lobe spectroscopy temperatures. | Microtemperature gradients, synapse-specific heat load, and the controller that produced the measured thermal state. | Bounded macro thermal proxy, not local thermal-controller ground truth. |
| Human perturbation-conditioned thermal routes Tan et al. (2024); Inoue et al. (2025) |
Bounded human evidence that severe heat exposure or intraoperative focal cooling changes measured brain temperature together with motor, executive, perfusion, or neurovascular responses. | Route-general local thermal-controller identity, cell-specific microtemperature, routine same-subject whole-brain thermal readout, and transfer outside the perturbation setting. | Perturbation-conditioned human thermal proxy, not local thermal-controller ground truth. |
| Quantity-defined human myelin MRI / tissue-health-sensitive routes Hagiwara et al. (2018); Baadsvik et al. (2024); Galbusera et al. (2025); Colaes et al. (2026) |
Distinct myelin-related route families spanning MT / SyMRI contrast, bilayer-sensitive mapping, pathology-linked remyelination-sensitive qT1, and a T1w/FLAIR ratio that may remain broader than myelin-specific contrast. | Adaptive conduction microgeometry, cell-specific oligodendrocyte controller state, and local timing compensation. | Quantity-defined myelin proxy family plus broader tissue-health-sensitive ratios, not one generic myelin-state meter or direct conduction-timing ground truth. |
| BBB water-exchange MRI Morgan et al. (2024); Padrela et al. (2025) |
ASL-derived Tex / Kw estimates for blood-to-tissue water exchange across the BBB. | Endothelial / pericyte controller identity, molecule-specific transport, and local barrier-regulation mechanism. | Macro BBB water-exchange proxy, not one generic permeability meter and not neurovascular-controller ground truth. |
| Tracer-specific BBB PET transport Chung et al. (2025) |
Tracer-specific BBB permeability-surface-area estimates from high-temporal-resolution dynamic PET and kinetic modeling. | A single route-independent BBB leakiness scalar, local endothelial / pericyte controller identity, and cell-specific support-state control. | Tracer-specific BBB transport proxy, not direct molecular-controller readout. |
| Blood-CSF barrier / choroid-plexus perfusion / transport family Zhao et al. (2020); Sun et al. (2024); Petitclerc et al. (2021); Anderson et al. (2022); Wu et al. (2026); Petitclerc et al. (2026) |
Choroid-plexus perfusion, blood-to-CSF water transport, choroid-plexus water cycling, apparent BCSFB exchange, or joint BBB-versus-BCSFB ASL exchange. | A generic BBB scalar, a generic clearance truth, choroid-plexus epithelial transporter identity, and local barrier-controller attribution. | Bounded BCSFB / choroid-plexus proxy family, not one interchangeable barrier meter and not direct controller readout. |
| Target-defined astrocyte PET routes Villemagne et al. (2022); Villemagne et al. (2022); Tyacke et al. (2018); Livingston et al. (2022); Jaisa-Aad et al. (2024) |
Target-defined MAO-B or I2BS binding routes with separate target-validation and disease-regime-dependent interpretation. | Astrocyte ensemble identity, downstream causal role, and cross-target equivalence of "astrocyte state." | Target-defined astrocyte-related proxy, not one generic astrocyte-state meter or controller identity readout. |
| Human TSPO disease-context / pathology-validated PET Biechele et al. (2023); Wijesinghe et al. (2025) |
Disease-context TSPO interpretation bounded by species-specificity limits and, in PSP, post-mortem alignment of in vivo PET with microglial pathology. | A route-free human activation-state scalar, cell-exclusive microglia identity, and transfer of the PSP validation regime to other diseases or targets. | Validation-bounded TSPO proxy, not one reusable human microglia-state meter. |
| Human CSF1R first-in-human / route-setting PET Horti et al. (2022); Ogata et al. (2025) |
First-in-human CSF1R PET routes in healthy volunteers with explicit arterial-input modeling, tracer-kinetic selection, and declared quantification burden. | Disease-grounded immune-controller identity, route-free microglia-state interpretation, and deployment-ready same-subject whole-brain monitoring. | Route-setting CSF1R proxy, not a disease-validated immune-state readout. |
| Human COX-2 enzyme-defined PET Yan et al. (2025) |
Healthy-human COX-2 PET with celecoxib blockade, declared kinetic choices, and route-local test-retest characterization. | Route-free neuroimmune-state identity, disease-general inflammatory burden, and local controller attribution beyond the named enzyme-defined regime. | Enzyme-defined COX-2 proxy, not one generic neuroimmune-state meter. |
| Human macroscopic CSF-oscillation route Fultz et al. (2019) |
Sleep-linked coupling among slow electrophysiology, hemodynamics, and large-scale CSF oscillations. | Protein-specific efflux, crossed-boundary clearance capacity, and local immune-controller identity. | Macroscopic CSF-motion proxy, not direct glymphatic or local maintenance-controller ground truth. |
| Human parenchyma-CSF water-exchange route Kim, Huang, & Liu (2025) |
Noninvasive magnetization-transfer spin-labeling estimates of in vivo human parenchyma-CSF water exchange. | Protein-specific clearance, meningeal-lymphatic egress, and route-free local immune control. | Parenchyma-CSF water-exchange proxy, not a generic clearance scalar or direct controller readout. |
| Human respiration-conditioned net-flow route Lim et al. (2025) |
Plane-specific awake-state CSF displacement and net-flow changes linked to respiration and diaphragm motion. | Whole-brain bulk circulation, endogenous-solute clearance, and local maintenance-controller identity. | Respiration-conditioned CSF net-flow proxy, not a route-free whole-brain clearance measurement. |
| Human exercise-conditioned contrast-influx / meningeal-lymphatic route Yoo et al. (2025) |
Intravenous-contrast-derived putative glymphatic influx and parasagittal meningeal-lymphatic flow associated with long-term exercise. | Natural-sleep baseline clearance, endogenous-solute transport, and local immune-controller identity. | Intervention-conditioned contrast-influx / meningeal-lymphatic proxy, not an all-purpose clearance readout. |
| Human intrathecal-tracer / CSF-to-blood clearance route Eide et al. (2023) |
Intrathecal gadobutrol retention and pharmacokinetic CSF-to-blood clearance variables associated with plasma biomarkers in neurological disorders. | Healthy-baseline equivalence, route-free local clearance control, and synapse-specific maintenance identity. | Tracer-retention / CSF-to-blood-clearance-capacity proxy, not direct local maintenance control. |
| Human CSF-mobility route Hirschler et al. (2025) |
Region-specific CSF mobility in humans, including vasomotion-linked modulation and disease-related mobility alteration. | Protein-specific efflux, crossed-boundary clearance capacity, and route-free local controller identity. | CSF-mobility proxy, not direct readout of local immune or synaptic-maintenance control. |
| Model-based human biomarker-efflux route Dagum et al. (2026) |
Overnight brain-to-plasma Aβ / tau clearance inference under sleep-versus-deprivation using an investigational device and a multicompartment model. | Route-free ground truth, local immune-controller identity, and same-subject whole-brain maintenance closure. | Model-based biomarker-efflux proxy, not direct local clearance-controller measurement. |
| Hemodynamic transfer audit Murphy et al. (2011); Williams et al. (2023); Epp et al. (2025) |
Calibration of baseline vascular state, cerebrovascular reactivity, and interpretation limits for BOLD / HbO / HbR differences. | A clean neural difference unless the vascular transfer side has also been audited. | Hemodynamic-limited evidence unless vascular transfer / CVR is explicitly calibrated. |
This page had already separated transport-side clearance from target-defined immune support, but one shortcut still remained inside the human PET lane. The primary literature does not support a generic immune PET row. Biechele et al. (2023) showed that TSPO is not a species-invariant human activation-state meter across fibrillar tauopathies, Wijesinghe et al. (2025) then constrained a disease-context / pathology-validated TSPO route in PSP, Horti et al. (2022) plus Ogata et al. (2025) constrained first-in-human CSF1R route-setting PET under explicit arterial-input modeling, and Yan et al. (2025) constrained an enzyme-defined COX-2 route with celecoxib blockade and declared repeatability. Therefore, this page no longer lets target-defined neuroimmune PET stand in for one interchangeable human immune-support advance. A paper may strengthen disease-context validation, route-setting / quantification burden, or pharmacologic specificity plus repeatability without becoming a route-free measure of local immune-controller identity or synapse-support mechanism.
This page had already separated synaptic-density PET from presynaptic release-machinery ceilings, but one internal shortcut still remained: it could still let all SV2A papers sound like one general rise in human synaptic observability. The primary literature does not support that compression. Naganawa et al. (2021) established a tracer-specific quantification route for 18F-SynVesT-1 in eight healthy volunteers with a levetiracetam blocking study in four, Finnema et al. (2018) established a same-subject test-retest route for [11C]UCB-J, Johansen et al. (2024) built a 33-person healthy atlas calibrated to autoradiography, Snellman et al. (2024) provided a cross-sectional risk contrast in cognitively unimpaired APOE groups, Shatalina et al. (2024) provided a task / cognition association route in 25 healthy adults, Smart et al. (2021) showed that visual stimulation can raise K1 without measurably changing VT or BPND in seven healthy participants, and Holmes et al. (2022) found no measurable overall SV2A increase 24 h after ketamine in healthy controls or MDD/PTSD participants despite symptom improvement. Therefore, this page no longer allows SV2A PET advanced to stand in for one interchangeable bundle anchor. A paper may strengthen quantification, repeatability, atlas coverage, risk contrast, task association, activation-null stability, or intervention-response boundaries without becoming a direct readout of current synaptic efficacy, release probability, release-site number, docked-vesicle architecture, active-zone nanostructure / priming-site assembly, or rapid synaptic plasticity.
| SV2A / synaptic-density PET route role | What the cited paper directly constrains | What it still cannot be promoted to on this page |
|---|---|---|
| Tracer / quantification route Naganawa et al. (2021) |
A tracer-specific kinetic and blocking route for 18F-SynVesT-1, including the useful modeling window for quantitative analysis. | A route-free synaptic-state scalar interchangeable across tracers, models, scan windows, or reference choices. |
| Same-subject baseline / repeatability route Finnema et al. (2018) |
Route-local test-retest stability for [11C]UCB-J under a declared acquisition and analysis pipeline. | Longitudinal biological sensitivity, cross-tracer equivalence, or momentary synaptic-state tracking during task or intervention. |
| Healthy atlas / cohort-prior route Johansen et al. (2024) |
A calibrated 3D healthy-human SV2A atlas with cortical and subcortical spatial gradients. | Same-subject change evidence, current efficacy, or a bridge to current release competence in an individual scan. |
| Cross-sectional risk-contrast route Snellman et al. (2024) |
Lower hippocampal [11C]UCB-J signal in cognitively unimpaired APOE ε4/ε4 carriers relative to lower-risk genotype groups. | A task-general cognitive readout, same-subject progression witness, or direct measure of the mechanism generating risk. |
| Task / cognition association route Shatalina et al. (2024) |
An association between regional [11C]UCB-J signal, task-switching neural activity, and switch-cost behavior in healthy adults. | A proof that SV2A PET measures rapid task-evoked synaptic change, release probability, or one general cognitive-capacity axis. |
| Within-subject activation-null boundary Smart et al. (2021) |
A boundary showing that brief visual activation can increase tracer delivery without measurably changing SV2A binding measures. | A license to treat perfusion-linked uptake change as synaptic-density change, or to read SV2A PET as a moment-to-moment activation meter. |
| 24 h intervention-response boundary Holmes et al. (2022) |
A ketamine perturbation test showing no measurable overall SV2A-density increase at the 24 h human imaging window despite symptom response. | A claim that acute clinical improvement automatically implies a detectable whole-brain presynaptic-density increase at that time point. |
Fultz et al. (2019) constrained a macroscopic sleep-linked CSF-oscillation route, Kim, Huang, & Liu (2025) constrained a parenchyma-CSF water-exchange route, Lim et al. (2025) constrained a respiration-conditioned net-flow route, Yoo et al. (2025) constrained an exercise-conditioned contrast-influx / meningeal-lymphatic route, Eide et al. (2023) constrained an intrathecal-tracer / CSF-to-blood-clearance route, Hirschler et al. (2025) constrained a CSF-mobility route, and Dagum et al. (2026) constrained a model-based biomarker-efflux route. Those rows do not share the same direct observable, carrier / analyte class, crossed boundary, intervention regime, or model burden. Therefore, this page no longer allows clearance evidence exists to stand in for a single support-state rung. The minimum safe question is which route family was measured and which local controller still remained latent.
Ren et al. (2015) established a resting metabolite / pH balance route, Ren et al. (2017) established a band-inversion / MT exchange-flux route, de Graaf et al. (2017) showed that human NAD+ detection at 7 T already depends on explicit overlap handling when read through 31P-MRS, Hendriks et al. (2019) showed that functional 31P readouts can require a close-fitting coil, 7 T, and a large visual-cortex spectroscopy volume just to detect subtle stimulation-linked pH shifts, Guo et al. (2024) mapped whole-brain intracellular NAD content, and Kaiser et al. (2026) reported task-evoked NAD+ dynamics in a functionally localized occipital voxel. On the deuterium side, Karkouri et al. (2026) strengthened a named absolute-quantification route, Li et al. (2025) strengthened a blood-input kinetic-rate route, Ahmadian et al. (2025) showed that dose changes downstream metabolite visibility, and Bøgh et al. (2024) showed that repeatability depends on a stated 3 T protocol and time-point window. Therefore, this page no longer accepts a single macro energetic row: resting balance, model-based exchange flux, whole-brain NAD content, localized functional NAD dynamics, absolute deuterium metabolite distributions, and deuterium kinetic-rate imaging remain separate proxy classes, while dose and repeatability remain route-specific operating conditions rather than generic guarantees.
Johansen et al. (2024) built an SV2A atlas from 33 healthy participants calibrated against postmortem autoradiography, which is a cohort-level synaptic-density proxy. Carro-Domínguez et al. (2025) then provided a human mixed arousal proxy during sleep rather than a transmitter-specific row. Hansen et al. (2022) plus Nakuci & Bansal (2025) provide a normative receptor / transporter prior and modeling scaffold. Wong et al. (2013) and Schlosser et al. (2025) provide selected occupancy / target-engagement routes, including informative null occupancy under a named dose and tracer design, whereas Koepp et al. (1998), Erritzoe et al. (2020), and Miederer et al. (2025) provide challenge-linked dopamine or serotonin release proxies under bounded windows. Within spectroscopy, even sharing 7 T does not collapse the rows: Ren et al. (2015) provides a resting metabolite / pH balance route, Ren et al. (2017) a model-based exchange-flux route, Guo et al. (2024) a whole-brain NAD-content route, Kaiser et al. (2026) a functionally localized NAD-dynamics route, Karkouri et al. (2026) a specialized absolute-quantification route, and Li et al. (2025) a specialized deuterium kinetic-rate route in only five healthy participants using custom dual-frequency coils and blood-input functions. Ahmadian et al. (2025) then showed that dose changes metabolite visibility, while Bøgh et al. (2024) showed that repeatability depends on a stated 3 T protocol and time-point window. Baadsvik et al. (2024) demonstrated bilayer-sensitive myelin mapping in two healthy volunteers on high-performance hardware, Morgan et al. (2024) and Padrela et al. (2025) provide water-exchange BBB routes, Chung et al. (2025) provides a tracer-specific BBB transport route, Villemagne et al. (2022) plus Villemagne et al. (2022) provide first-in-human and AD-spectrum MAO-B route readings, Tyacke et al. (2018) provides an I2BS route, and the current clearance family already splits further into macroscopic CSF oscillation, parenchyma-CSF water exchange, respiration-conditioned net flow, exercise-conditioned contrast influx / meningeal-lymphatic flow, intrathecal tracer / CSF-to-blood clearance, CSF mobility, and model-based biomarker efflux. Therefore, these rows have to be read along three axes at once: what object the route measures, how specialized or deployment-limited the route still is, and which hidden-state family the route can safely calibrate. A row can be real, narrow, and high-burden all at once.
The point is not that these routes are weak. The point is that each route lowers a different latent-state error term. On this page, "human evidence advanced" now means "a specific layer advanced under a specific ceiling," not "the human whole-brain state became nearly observable." The cited papers do not yet show same-person, same-session, externally validated fusion of these rows into one state-complete readout, so proxy-rich human evidence remains weaker than same-subject state closure. That last sentence is an inference from the measurement properties summarized above.
| Composition failure that still remains | What the primary literature actually gives | Why the bundle still cannot be promoted directly | What this page now requires instead |
|---|---|---|---|
| Quantity-type collapse | Johansen et al. (2024) provide a cohort-level SV2A density atlas, Lucchetti et al. (2025) provide a five-metabolite similarity scaffold, Ren et al. (2015) provide resting 31P energetic balance, Ren et al. (2017) provide model-based 31P exchange flux, Guo et al. (2024) provide whole-brain NAD-content mapping, Kaiser et al. (2026) provide localized task-evoked NAD+ dynamics, Karkouri et al. (2026) provide absolute deuterium metabolite maps, Li et al. (2025) provide kinetic glucose-rate imaging, Fleysher et al. (2013) provide compartment-defined sodium quantities, Rzechorzek et al. (2022) provide macro thermometry, Morgan et al. (2024) provide BBB water exchange, Chung et al. (2025) provide tracer-specific BBB transport, Villemagne et al. (2022) plus Villemagne et al. (2022) provide first-in-human and AD-spectrum MAO-B PET, and the current clearance family already spans macroscopic CSF oscillation, parenchyma-CSF water exchange, respiration-conditioned net flow, exercise-conditioned contrast influx / meningeal-lymphatic flow, intrathecal tracer / CSF-to-blood clearance, CSF mobility, and model-based overnight biomarker-efflux inference. | Density, similarity, high-resolution metabolite distribution, resting balance, exchange flux, static NAD content, task-locked local dynamics, deuterium absolute metabolite mapping / quantification, kinetic-rate maps, ionic partition, temperature, water exchange, tracer-specific transport, target-defined binding, mobility, and model-based efflux are different inferential objects. Row count is therefore not yet a validated latent coordinate. | Name the claimed latent variable and the direct observable by row before any cross-row promotion. |
| Operational-maturity collapse | Karkouri et al. (2026) used a dedicated 7 T absolute-quantification pipeline, Li et al. (2025) used 7 T dynamic DMRSI in five healthy participants, Ahmadian et al. (2025) showed deuterium dose dependence, Bøgh et al. (2024) showed route-local repeatability at 3 T, Baadsvik et al. (2024) mapped myelin bilayer in two healthy volunteers on high-performance hardware, Lim et al. (2025) relied on real-time velocity-encoding MRI and a breathing-trained versus control comparison, Yoo et al. (2025) relied on intravenous contrast plus IR-ALADDIN and black-blood imaging, Hirschler et al. (2025) used a specialized 7 T CSF-mobility sequence, Eide et al. (2023) depended on neurological-patient intrathecal tracer pharmacokinetics, and Dagum et al. (2026) combined overnight physiology with an investigational device and a multicompartment model. | A route can be scientifically real while still being proof-of-principle, hardware-specialized, small-cohort, model-heavy, dose-dependent, or protocol-specific in its repeatability. Existence is not the same as deployment-ready composition. | Disclose cohort size, hardware class, model burden, and whether the row is routine, specialized, or proof-of-principle. |
| Common-driver / fusion collapse | Vafaii et al. (2024) showed both common and divergent cross-modal organization, Chen et al. (2025) showed coupled global progression plus two distinct network patterns in simultaneous EEG-PET-MRI, Bolt et al. (2025) showed that a major global fMRI mode is substantially coupled to autonomic physiology, and Epp et al. (2025) showed that significant BOLD changes can coexist with opposite oxygen-metabolism changes. | Same-session agreement still does not prove one solved biological state axis. Some rows can share nuisance or autonomic drivers, while others diverge even when acquired together. | Require a shared-driver audit, an explicit shared-versus-specific decomposition, and a report of what the bundle adds beyond the strongest single row. |
On this page, several living-human proxy rows are promoted together only after the bundle names the claimed latent variable, the direct observable by row, the calibrator role of each row, same-subject / same-session / same-perturbation status, model and hardware burden, whether the apparent agreement survives a common-driver audit, and what the bundle adds beyond the strongest single row under matched conditions. That operational rule is implemented on the site as the Verification: Human Proxy Composition Card. If the bridge is sequential rather than simultaneous, the site also adds the State-Continuity Bridge Card rather than letting same-subject wording stand in for same-state evidence. This promotion rule is an inference from the measurement properties summarized above.
1. Sensing: Precise reading of brain activity and quantification of uncertainty
Electroencephalography (EEG) offers high temporal resolution and is therefore a powerful input signal for WBE, but its low spatial resolution remains a fundamental limitation. EEG source imaging (ESI) is a computational response to that problem, yet it remains an ill-posed problem[5] and therefore does not yield a unique solution. For engineering goals that demand very high reliability, such as WBE, relying only on traditional minimum-norm methods or point estimates such as dSPM risks propagating estimation error throughout the system.
What matters here is not the solver name but the evidence chain. Empirical Bayesian methods,[78] high-density EEG, and individualized MRI are all promising ways to improve estimation conditions, but no method should be called a standard unless it is paired with EEG-BIDS-aligned reporting,[26] disclosure of electrode coordinates and forward models, sensitivity analysis for conductivity uncertainty,[79] and external validation.
- Positioning of high-density EEG and Bayesian estimation: The Block-Champagne framework,[78] FEM / BEM forward models, high-density EEG, and individualized MRI are important for improving estimation conditions. However, greater sensor density does not guarantee source uniqueness, because skull-induced spatial smoothing and inverse-problem non-uniqueness still remain.[101] In this project, they are treated not as a guarantee but as preconditions for narrowing error sources.
- Visualization and external validation of uncertainty:The estimated brain activity map includes not only amplitude but also"Credible Intervals"Alternatively, indicate the concentration of the posterior distribution to clearly indicate areas of high uncertainty. Furthermore, we cannot call something ``improved'' unless we report how much the error has been reduced under which conditions relative to external standards such as simulations, phantoms, simultaneous invasive recordings, and intracranial stimulation.
- Uncertainty and error propagation for forward problems:Errors in the conductivity and shape of head tissues (especially the skull) directly affect localization errors.[79], include at least a sensitivity analysis or range assessment in your submission. Although full probabilistic modeling is powerful, it is not universally required as of 2026-03, and we will prioritize allowing a third party to audit the error range.
2. Decoding: Introducing active inference and counterfactual virtual equivalence
In the "Decoding" section, traditional decoding techniques such as Mind Captioning are still mappings based on correlation. To make a claim that approaches WBE, it is not enough to imitate inputs and outputs; the system must also expose unlearning conditions, intervention conditions, failure conditions, and generative predictive performance. What matters here is not fixing a single theory as the correct answer, but arranging multiple generative models so that they can be compared.
Laukkonen et al.'s discussion of "counterfactual equivalence"[76] together with active-inference-oriented discussions[80] is useful for deciding what should be tested. However, as of 2026-03, it is still not a shared acceptance standard by itself. On this page it is treated as a design hypothesis that must be evaluated together with OOD generalization, perturbation response, calibration, abstention, and alternative-model reporting.
- Position of Active Inference: It is a powerful way to view the brain as a generative model that interacts with the environment rather than as a passive decoder.[80] In this project, however, it is not the only implementation principle; together with DCM, state-space models, and SCM, it is treated as a candidate in model competition.
- Extension of the Turing test (causal perturbation protocol): Because static counterfactuals cannot be tested by observation alone, we use a "Causal Perturbation Protocol" that extends the Turing Test. The biological brain's response to physical perturbations such as TMS is compared with the response distribution to virtual perturbations in emulation, and PCI is treated as one external benchmark, not as ground truth itself.[90][100].
2026-03-29 addendum: Candidate-model comparison is necessary, but it is still not enough to let a directed graph read as discovered causal wiring. Smith et al. (2011) showed that lag-based approaches perform poorly for fMRI and that functionally inaccurate ROIs are especially damaging to network estimation, Barnett & Seth (2017) showed that subsampling can create detectability black spots for Granger-causal interactions, Vink et al. (2020) showed that resting-state EEG functional connectivity explains less than 10% of TMS-evoked propagation variance, Novelli et al. (2025) showed that slow BOLD sampling can still induce spurious Granger-causal inference even when HRF variability alone need not do so, and Yan et al. (2026) showed that latent confounders remain an active challenge for biological network reconstruction. Therefore, this page now requires observed-subsystem closure / latent-confound audit, node-definition policy, and sampling / transformation sensitivity in addition to model comparison, perturbation validation, and abstention.[141][142][143][144][146]
3. Implementation: Consideration of computational limitations of IIT 4.0 and alternative indicators
This project has traditionally relied on integrated information theory (IIT 4.0), but its biggest implementation problems are the explosion of computational cost (NP-hard) and the ambiguity of "intrinsic reality" in digital infrastructures. Rather than treating IIT dogmatically, we use the Adversarial Collaboration results[54] to separate main judgments from auxiliary analyses in the implementation stack.
- Position of computable approximate indicators: Direct calculation of Phi (integrated information) scales exponentially with system size. Approximation methods using PCI-ST and low-dimensional embeddings[81] are therefore treated as engineering proxies, not as direct substitutes for consciousness or identity.
- Geometric comparison of neural activity manifolds (Neural Manifold Geometry): To capture structural differences that scalar values such as PCI cannot show, graph indicators, TDA, and persistent homology are introduced as auxiliary analyses. Interpretation consistency and noise robustness are still not standardized well enough for primary pass/fail decisions, so those decisions remain tied to simpler and more auditable metrics.
- Introducing thermodynamic constraints: We keep the idea of auditing computational cost and physical cost separately, but irreversibility and entropy production are currently only exploratory auxiliary logs. Quantities extracted from coarse-grained neural data should not be equated directly with microscopic dissipation or with necessary conditions for consciousness.[92].
From decoding to emulation: Explanation of logical gaps and verification design
Decoding that “reads” sentences from brain activity is powerful, but it is basically a translation of observed outputs, whereas WBE requires the generation of brain dynamics through an autonomous causal model.[8] To close that gap, the generative model must be explicit about inputs, internal state, and outputs, and it must be validated with predictions about interventions and perturbations.[13][45].
Contents of the gap (paraphrase that even high school students can understand)
For example, even if you read the "test answers",what the person usually thinks(How to respond to new problems) cannot necessarily be reproduced. Similarly, although brain-to-text can express ``brain activity at this moment,'' it does not guarantee the ``process of continuously updating the state while interacting with the environment'' required by WBE.
(Translation)
Output: Text/Media
(Static)
(Causal Model)
Output: Dynamics
(Process)
Points that are academically problematic (what should be additionally shown)
- Difference between decode and emulate:Mind Captioning[11]and continuous language recovery[30]showed that meaning and sentences can be reconstructed from brain signals. However, WBE requires the reproduction of the causal process itself: how internal states transition and how future outputs (actions, thoughts, self-models) are generated when the same input (sensation) is given.[8].
- Problems where the language-prior distribution (LLM) wins: Because LLMs are fluent, weak evidence can still produce plausible-looking sentences, i.e. hallucinated outputs.[28] Therefore, it is necessary to baseline how much brain-signal information actually contributes to the output by using counterfactual inputs such as shuffles, and to define abstention conditions up front.
- Many-to-one (different models explaining the same observation) problem:Just as EEG source estimation is an ill-posed problem,[5], there can be multiple explanations that fit the observed data. Definitional non-uniqueness is also discussed in IIT.[3]. For WBE, intervention prediction and partial-observation audit, not observational agreement alone, are what narrow the model family.
- Verification of preservation of consciousness and identity:The position that views identity in terms of psychological continuity includes the copying problem (multiplicity).[4][58]. Even if one adopts the Slow Continuous Mind Uploading hypothesis[59], it is still necessary to audit whether "conscious capacity" is preserved using behaviorally independent indicators such as the PCI family.[47].
Demonstration plan to fill the gap (minimum)
- Make the generative model explicit:Using DCM or related generative models, clarify what is a state and what is a parameter, publish the node-definition / parcellation policy, audit observed-subsystem closure and latent confounders, and present the result in a form that allows model comparison plus recovery checks.[13][141][145][146].
- Preregister intervention (perturbation) predictions:Incorporate the propagation and complexity of reactions to perturbations into the evaluation axis, like PCI/PCI-ST[47][51].
- Require a counterfactual baseline:Quantify the output produced by linguistic prior distribution alone by input shuffling, trial replacement, model temperature fixation, etc., and report it as an effect size.[28].
Verification boundaries for causal structure, state completeness, and physical constraints (updated in 2026-03 audit)
In translating the demonstration plan into implementation terms, the most important correction was to stop treating different limitations as a single kind of "difficulty." In the primary literature, at least observability, identifiability, maintenance-state, intervention, and thermodynamic readout are separate barriers, and progress on any one of them cannot substitute for all the others.
| wall | What the primary literature now supports | Don't support it yet | Correction policy on this page |
|---|---|---|---|
| Wall of Observability | Dorkenwald et al. (2024), MICrONS Consortium et al. (2025), and Ding et al. (2025) significantly advanced local connectomics, sequential same-brain structure-function bridging, and stimulus-conditioned response prediction. On the scalp EEG side, forward-model and head-conductivity sensitivity analysis can also be audited.[5][79]. | Still, we cannot say that we have been able to directly observe the human whole-brain in a state-complete manner. Improvements in sequential same-brain scaffolds, stimulus-conditioned cortical digital-twin models, and source imaging cannot be translated directly into observability of the whole brain and all states. | Instead of writing local connectomics, non-invasive ESI, and whole-brain WBE in the same breath, we first clarify the claim ceiling for each measurement stack. |
| Wall of identifiability | The EEG inverse problem can be improved considerably if the head model, conductivity assumptions, and model space comparisons are carefully handled.[5][79]. On the model side, Regression DCM and related large-scale estimators improve tractability, while systems-identification and network-reconstruction work make clear that unknown inputs, latent confounders, ROI policy, and sampling regime remain first-order conditions for inference.[96][141][142][145][146]. | However, improved predictability, denser graphs, or better localization do not guarantee uniqueness recovery. As long as other model families or equivalence classes still fit the observed data, or the observed subsystem is not closed, ``good accuracy'' cannot be written as ``the internal state or causal structure was uniquely known.'' | This page prioritizes evidence chain over solver name and now requires an explicit effective-connectivity route card: family comparison, observed-subsystem closure / latent-confound audit, node-definition policy, sampling / transformation sensitivity, perturbation validation, and abstention conditions. |
| Maintenance-state wall | Hengen et al. (2016), Torrado Pacheco et al. (2021), and Xu et al. (2024) support homeostatic recovery that depends on sleep / wake dynamics. Looser et al. (2024) highlights oligodendrocyte-axon metabolic coupling, while Cahill et al. (2024) and Lee et al. (2022) strengthen the case for glial and active maintenance. | From same-day decode performance and short-term activity matches to cross-day stability, overnight recovery, and timing-sensitive maintenance, they cannot be considered the same. | State-completeness gates include sleep history, myelin/delay, glial substrate-routing, astrocyte-state, and cross-day maintenance-state longitudinal logs. |
| Intervention wall | Hernandez-Pavon et al. (2023) for organized TMS-EEG, Flesher et al. (2021) for bidirectional BCI, and Oehrn et al. (2024) for adaptive DBS together provide strong causal evidence in local subsystems and disease settings. | Still, the success of locally closed loops does not directly support whole-brain branch-equivalence or identity preservation. If you remove the stimulus site, intensity, latency/jitter, and artifact windows, the comparison itself collapses. | Treat intervention evidence as staged evidence, and write separately for passive observation, held-out perturbation, online loop, and long-term adaptive operation. |
| Thermodynamic readout wall | Lynn et al. (2021) and de la Fuente et al. (2023) showed that time-irreversibility signatures can appear in coarse-grained neural dynamics, while Ishihara & Shimazaki (2025) estimated task-dependent entropy flow only under an explicit state-space kinetic Ising model. | However, these routes do not directly measure microscopic physical dissipation or a WBE-ready energy budget. Blom et al. (2024) showed that coarse observations can hide dissipative cycles and acquire memory, and Epp et al. (2025) showed that significant BOLD changes can oppose oxygen-metabolism changes across the cortex. | Thermodynamic claims therefore stay behind an irreversibility route card: signal route and state definition, coarse-graining / timescale, observed-state closure / memory order / reverse-transition support, estimator family and dynamical assumptions, physiology-side grounding, cost isolation, and abstention must be disclosed before any energetic language is raised. Without that card, the result remains an auxiliary log beside OOD, perturbation, external validation, and abstention conditions. |
One remaining weakness in the older wording was that irreversibility, EPR, and energy cost could still sound like one continuous evidential object. The current primary literature does not support that shortcut. Lynn et al. (2021) estimated entropy-production lower bounds only after coarse-graining BOLD dynamics into macrostates, de la Fuente et al. (2023) showed that reversibility detection in ECoG depends on principal-component choice, feature set, and model complexity, Ishihara & Shimazaki (2025) estimated task-dependent entropy flow only under an explicit state-space kinetic Ising model, Blom et al. (2024) showed that coarse observations can hide dissipative cycles and acquire memory, and Epp et al. (2025) showed that significant BOLD changes can oppose oxygen-metabolism changes across the cortex. Therefore, on this page, a thermodynamic claim is not promoted beyond an exploratory auxiliary analysis unless it names signal route and state definition, coarse-graining / timescale, observed-state closure / memory order / reverse-transition support, estimator family and dynamical assumptions, physiology-side grounding when energetic language is used, cost isolation, and abstention boundary.
One remaining weakness in the older wording was that a larger or faster directed graph could still sound too close to a discovered causal object. The primary literature does not support that shortcut. Villaverde et al. (2019) showed that unknown inputs, states, and parameters often have to be estimated jointly, not one by one. Smith et al. (2011) showed that functionally inaccurate ROIs are especially damaging to fMRI network estimation, Barnett & Seth (2017) showed detectability black spots under subsampling, Vink et al. (2020) showed that resting-state EEG connectivity is a weak predictor of perturbation propagation, Novelli et al. (2025) showed that slow BOLD sampling can still generate spurious Granger-causal inference, and Yan et al. (2026) showed that latent confounders remain an unresolved practical challenge in biological network reconstruction. Therefore, on this page, a DCM graph, regression DCM estimate, or other effective-connectivity result is not promoted beyond a model-conditioned causal hypothesis unless the route card names what part of the system was actually observed, how nodes were defined, how temporal and observation transforms were stress-tested, and what perturbation or external falsification route was passed.
Interpretations such as ``the human whole-brain state was achieved because of advances in local connectomics'', ``the internal state could be uniquely restored because the ESI was improved'', ``the identity of the person could be verified because the closed loop was activated'', and ``physical identity was guaranteed because of irreversibility'' are not supported by the current primary literature.
- DCM and causal indicators are used with route-card assumptions: DCM, Effective Information, Causal Density, and Symbolic Transfer Entropy (STE)[99] are useful as additional analyses, but they should not be used as stand-alone universal pass/fail criteria. Key judgments still require reporting equivalence classes, observed-subsystem closure, node-definition policy, sampling / transformation sensitivity, perturbation validation, and test-retest reliability.
- Fix the minimal evidence chain:For L2 and higher claims, the lowest line is (a) pre-registered hold-out/OOD conditions, (b) validation with perturbations or external criteria, (c) uncertainty and abstention conditions, (d) reporting of alternative models or equivalence classes, (e) maintenance-state logs including cross-days, and (f) separation of computational cost and hardware power.
Substance and Reproducibility of the project
As of 2026-03, this repository is still documentation-first. That is useful for setting claim ceilings, route cards, and audit rules, but it is not yet a third-party executable EEG or multimodal benchmark package. The current public artifacts are mainly the site text, the wiki/export toolchain, schema examples such as dataset_description.json, and the audit contracts that explain what evidence must be shown. What is still missing for L0 is a fixed analysis environment, a named public dataset snapshot, machine-executable preprocessing / metric scripts, and expected outputs that another group can rerun without hidden manual steps. Therefore, this page now separates what is already public from what remains missing before L0 reproducibility instead of speaking as if one continuous executable pipeline already existed.[25][26]
| Public lane | What is already public now | What is still missing before L0 |
|---|---|---|
| Documentation / audit rules | Claim ceilings, route cards, benchmark-reading rules, and validation ladders are already public on the site. | A third party still cannot rerun a named EEG benchmark from raw inputs to expected outputs using this repository alone. |
| Data organization | BIDS / EEG-BIDS are already adopted as the target organizational language, and schema examples are being used to define what must be logged.[25][26] | A concrete public dataset snapshot, sidecars, validator output, and a complete reproduction manifest still need to be published together. |
| Tooling | The repository already contains public-content and wiki/export automation, which helps keep the documentation layer auditable. | The executable analysis layer is still missing: preprocessing, metrics, baselines, and expected-result checks are not yet packaged as one public rerunnable bench. |
| Model / evaluation layer | The site now defines the negative controls, abstention rules, and validation classes that future decoding or source-imaging work must satisfy. | The code that instantiates those rules on a named public dataset, with fixed environment and regression tests, remains to be published. |
For now, the strongest safe public claim is that this repository publishes measurement / verification specifications and reading rules. It does not yet publish a completed EEG analysis package that a third party can execute end to end. On this site, the correct next step is therefore to turn the existing audit rules into a small public L0 benchmark bundle rather than to imply that the executable layer already exists.
Key Technical Challenges
Gap between connectome and dynamics
Research on the structural connectome has advanced substantially in 2024-2025. Dorkenwald et al. (2024) mapped the wiring diagram of an adult Drosophila whole brain, while MICrONS Consortium et al. (2025) released a sequential same-brain dataset that co-registers in vivo visual responses with later EM in one mouse, and Ding et al. (2025) used a validated stimulus-conditioned response model together with that dense reconstruction to test a general wiring rule. What follows directly from these results is structural scaffold plus sequential local structure-function bridge plus conditional response prediction, not that wiring alone determines the full state, not that current synaptic-state was directly read out, and not that one unique local twin was recovered.
Living-human diffusion-MRI tractography sharpens a different structural rung, but at a lower ceiling than human connectome language often suggests. Thomas et al. (2014) showed that anatomical accuracy is inherently limited when tractography is built from voxel-averaged local orientations, Gajwani et al. (2023) showed across 1,760 group connectomes that hub location and strength move materially with processing choices, McMaster et al. (2025) showed that graph measures shift across voxel resolutions in scan / rescan HCP data, Bramati et al. (2026) showed that common diffusion-sampling schemes still move tractography outputs even when scanner and preprocessing are held fixed, and Zhu et al. (2025) improved reconstruction by adding microscopy to MRI rather than by making MRI-only tractography route-free. The safe ceiling here is therefore acquisition-conditioned macro pathway prior, targeted bundle-hypothesis route, or calibrated bundle comparison, not a finished living-human connectome.
Likewise, Yao et al. (2023) strengthened whole-brain spatial atlases for cell identity and spatial priors, Gouwens et al. (2021) and Gamlin et al. (2025) strengthened the bridge from transcriptomes to morpho-electric phenotypes and local motifs, and Shapson-Coe et al. (2024), Johansen et al. (2024), Lucchetti et al. (2025), Guo et al. (2025), Ren et al. (2015), Ren et al. (2017), Guo et al. (2024), Kaiser et al. (2026), Karkouri et al. (2026), Li et al. (2025), Qian et al. (2012), Rzechorzek et al. (2022), Baadsvik et al. (2024), Morgan et al. (2024), Padrela et al. (2025), Chung et al. (2025), Zhao et al. (2020), Sun et al. (2024), Petitclerc et al. (2021), Anderson et al. (2022), Wu et al. (2026), Petitclerc et al. (2026), Villemagne et al. (2022), Tyacke et al. (2018), Fultz et al. (2019), Kim, Huang, & Liu (2025), Lim et al. (2025), Yoo et al. (2025), Eide et al. (2023), Hirschler et al. (2025), and Dagum et al. (2026) strengthened different human evidence classes. Importantly, these solve different kinds of deficits: cell identity and spatial priors, route-conditioned human macro pathway priors, local morpho-electric bridges, local human structural scaffold, regional synaptic-density proxy, biochemical similarity scaffold, high-resolution metabolite-distribution structure, 31P metabolite / pH balance, 31P MT exchange-flux, 31P NAD-content mapping, localized functional 31P NAD-dynamics, deuterium metabolite-mapping / absolute-quantification, deuterium kinetic-rate imaging, quantity-defined ionic / thermal / myelin routes, macro BBB water-exchange, tracer-specific BBB transport, a blood-CSF-barrier family that itself splits into choroid-plexus perfusion, blood-to-CSF transport, water cycling, and apparent exchange, target-defined astrocyte PET, and a clearance family that itself splits into macroscopic CSF oscillation, parenchyma-CSF water exchange, respiration-conditioned net flow, exercise-conditioned contrast influx / meningeal-lymphatic flow, intrathecal tracer / CSF-to-blood clearance, CSF mobility, and model-based biomarker efflux. Improvement in one stack therefore cannot be treated as if it meant "we now see everything."
Current public evidence still leaves at least nineteen maintenance-state families plus a separate shared electrical-state class outside connectome plus cell-type description. Gouwens et al. (2021), Grubb & Burrone (2010), and O'Leary et al. (2014) show that intrinsic excitability, AIS / ion-channel configuration, and post-perturbation recovery rules are not fixed by graph plus label alone, while Hengen et al. (2016) shows that firing-rate set points themselves remain a separate controller. Santoni et al. (2024), Wang et al. (2015), Dai et al. (2019), Shi et al. (2018), Peterson et al. (2025), Lee et al. (2003), Tomita et al. (2005), Vierra et al. (2023), and Rodriguez et al. (2025) show that transcription / chromatin state, post-transcriptional RNA control, and phospho-signaling / second-messenger routing remain distinct controller layers even when the same connectome or bulk abundance background is held fixed.
Frey & Morris (1997), Govindarajan et al. (2011), Pandey et al. (2021), and Thomas et al. (2025) show that late stabilization depends on local proteostasis and synaptic tagging, while Park et al. (2006), Swarnkar et al. (2021), and de Queiroz et al. (2025) show that compartment-specific cargo delivery remains another state layer. Torrado Pacheco et al. (2021) and Xu et al. (2024) show that sleep/wake renormalization changes the network's operating regime, while Ngo et al. (2013), Geva-Sagiv et al. (2023), Schreiner et al. (2024), and Deng et al. (2025) show that replay-coupling and specific NREM windows matter for consolidation. Gibson et al. (2014), McKenzie et al. (2014), and Looser et al. (2024) show that timing support remains partly myelin-dependent, Hardingham & Larkman (1998), Volgushev et al. (2000), and Long & Fee (2008) show that local thermal-state can alter kinetics without rewiring, Pizzorusso et al. (2002), Frischknecht et al. (2009), and Gogolla et al. (2009) show that ECM / PNN state changes plasticity gates, and Glykys et al. (2014), Heubl et al. (2017), and Ding et al. (2016) show that ionic milieu / chloride homeostasis changes inhibitory polarity and state transitions.
Galarreta & Hestrin (1999), Anastassiou et al. (2011), Burman et al. (2023), Yang et al. (2024), and Selfe et al. (2024) show that shared extracellular / electrical state can alter synchrony outside the chemical synapse graph, while Carro-Domínguez et al. (2025), Hansen et al. (2022), Nakuci & Bansal (2025), Wong et al. (2013), Schlosser et al. (2025), Miederer et al. (2025), and Neyhart et al. (2024) show that neuromodulatory specificity is not one mood scalar but a split among mixed arousal proxies, normative receptor / transporter priors and modeling scaffolds, administered-drug occupancy routes, challenge-linked release proxies, and local transmitter routes. Rangaraju et al. (2014), Rangaraju et al. (2019), Divakaruni et al. (2018), Bapat et al. (2024), and Hu et al. (2025) show that branch-local bioenergetic / mitochondrial support remains consequential. Separately, Bell et al. (2010), Pandey et al. (2023), Swissa et al. (2024), and Mai-Morente et al. (2025) show that neurovascular-unit / BBB / pericyte state is not only a measurement-side vascular confound but another maintenance-side controller, while Padrela et al. (2025) and Chung et al. (2025) still remain macro permeability routes rather than cell-specific controller readouts. Suzuki et al. (2011), Cahill et al. (2024), Williamson et al. (2025), Dewa et al. (2025), and Bukalo et al. (2026) show that astrocyte-state is part of active maintenance rather than generic background, and the current human clearance family already spans macroscopic CSF oscillation, parenchyma-CSF water exchange, respiration-conditioned net flow, exercise-conditioned contrast influx / meningeal-lymphatic flow, intrathecal tracer / CSF-to-blood clearance, CSF mobility, and model-based biomarker efflux. Those human rows are informative, but they still differ in direct observable, crossed boundary, intervention regime, and model burden, so this page does not treat them as one support-state meter or as direct identification of the local immune controller.
The remaining weakness on this central page was that even after separating shared extracellular / electrical state from the nineteen maintenance-state families, the electrical row could still be read as one hidden-state bucket. The primary literature does not support that compression. Galarreta & Hestrin (1999) is about gap-junction network topology. Anastassiou et al. (2011) is about endogenous-field / ephaptic spike-timing bias. Graydon et al. (2014), Kilb et al. (2006), Lauderdale et al. (2015), and Xie et al. (2013) constrain extracellular-space geometry / dilution / osmotic regime rather than electrical-synapse topology. Burman et al. (2023) and Selfe et al. (2024) constrain inhibitory driving-force regime, with the latter using a specialized direct optical assay rather than a field-ready whole-brain route. Yang et al. (2024) then shows activity-dependent electrical-synapse remodeling, which is again not the same object as a fixed gap-junction graph. On the human side, Voldsbekk et al. (2020) remains a diffusion-MRI clue about wakefulness-related extracellular-space change, while Feld et al. (2026) remains a sleep-perturbation clue that electrical synapses matter for declarative-memory retention. Therefore, this page no longer permits `electrical-state evidence exists` to be translated into `the shared extracellular / electrical regime was measured` unless the claim also says which family, which direct observable, which spatial regime, and which human ceiling apply.
| shared extracellular / electrical family | what the cited papers directly constrain | safe ceiling on this page |
|---|---|---|
| gap-junction network routes Galarreta & Hestrin (1999); Yang et al. (2024) |
Electrical coupling topology between named cells and activity-dependent engagement / disengagement of those junctional routes in local preparations. | Local electrical-coupling route, not proof that a chemical connectome or nominal inhibitory edge list already fixes the electrical regime. |
| endogenous-field / ephaptic routes Anastassiou et al. (2011) |
Field-driven spike-timing bias under local endogenous electric-field geometry. | Local ephaptic clue, not a generic certificate that field effects matched across preparations or scales. |
| extracellular-space geometry / osmotic routes Graydon et al. (2014); Kilb et al. (2006); Lauderdale et al. (2015); Xie et al. (2013) |
Extracellular-volume geometry, neurotransmitter dilution, osmotic contraction / edema-linked excitability, and sleep-linked interstitial-space expansion. | Extracellular-space route family, not direct whole-brain proof that the same graph implies the same spillover, dilution, or state-switch regime. |
| inhibitory driving-force routes Burman et al. (2023); Selfe et al. (2024) |
State-dependent shunting versus driving inhibition in active cortex and direct optical reporting of inhibitory receptor driving force in specialized preparations. | Local inhibitory-regime route, not a routine whole-brain readout of local EGABA or electrotonic state. |
| human clue layer Voldsbekk et al. (2020); Feld et al. (2026) |
Wakefulness-related diffusion-MRI evidence consistent with extracellular-space change in white matter and a pharmacological sleep-perturbation clue that electrical synapses contribute to declarative-memory retention. | Bounded human clue class, not direct ground truth of coupling topology, extracellular-space geometry, or local inhibitory-driving-force regime in the living whole brain. |
On this page, the maintenance-side taxonomy follows the site's current rule of 19 maintenance-state families in Wiki: Homeostatic Plasticity and Maintenance State, while shared extracellular / electrical state remains a separate class linked to the Wiki: electrical-state route card. This keeps maintenance mechanisms and electrical-regime claims from being silently collapsed into one generic hidden-state bucket.
| Hidden-state class | Why connectome + cell type still leaves it open | Safe reading on this page |
|---|---|---|
| Intrinsic excitability / AIS / ion-channel state | Threshold, gain, rebound, burstiness, AIS geometry, and ion-channel configuration can still differ within the same transcriptomic label. | Do not read cell-type labels or short activity matches as proof that the underlying input-output law also matched. |
| Firing-rate set point / recovery controller | The post-perturbation return point, recovery timescale, and compensatory path can still differ even when baseline firing rates look similar. | Do not read matched average firing or short-term stability as proof that the maintenance controller also matched. |
| Activity-dependent transcription / chromatin state | Allocation eligibility, late stabilization programs, and locus-specific plasticity control can still differ over hours to weeks on the same graph. | Do not read atlas completeness as if the current memory-allocation controller were already fixed. |
| Post-transcriptional RNA-state | Isoform choice, m6A-dependent translation / degradation, and RNA-editing ratios can still change plasticity on the same graph even when gene-level abundance looks similar. | Do not read transcript abundance as if it already fixed the active RNA controller. |
| Phospho-signaling / second-messenger state | Phosphosite occupancy, kinase/phosphatase balance, and compartment-specific cAMP / Ca2+ / PKA signaling can still differ on the same transcript and protein background. | Do not read transcriptomics, proteomics, or nominal weights as if they already fixed the active phospho-controller. |
| Local proteostasis / synaptic-tagging state | Tagged spines and branches can differ in PRP capture, translation, degradation, and autophagy balance even when current weights look similar. | Do not treat a weight snapshot as the late-stabilization route. |
| Cargo-transport / cytoskeletal trafficking state | Branch-, spine-, and bouton-specific delivery and retention of receptors, endosomes, RNA cargoes, mitochondria, and presynaptic components still remain variable. | Do not treat a graph plus local protein availability as proof that the correct cargo reached the correct compartment. |
| Sleep/wake-dependent renormalization | A same-day fit is a different issue from overnight maintenance of the operating regime. | Do not treat daytime success as proof of sleep-consistent network recovery. |
| Sleep architecture / replay-coupling state | Slow-oscillation / spindle / ripple coordination and consolidation-permissive NREM windows are not fixed by the daytime graph. | Do not treat sleep duration or same-day recovery as proof of replay-equivalent maintenance. |
| Delay / myelin state | Conduction timing and axonal support remain plastic and can shift synchrony even without rewiring. | Timing-sensitive claims need more than a structural graph plus one scalar delay constant. |
| Thermal-state | Local temperature still shifts membrane kinetics, spike generation, field-potential amplitude, and sequence timing on the same graph. | Macro thermometry is not proof that local timing-state matched. |
| Perisynaptic ECM / PNN state | Plasticity gate, receptor mobility, and memory-update resistance still vary without changing the synapse list. | Do not treat synapse counts or static weights as sufficient for adult plasticity claims. |
| Ionic milieu / chloride homeostasis | Local chloride set point and extracellular ion composition can change inhibitory sign and state-transition logic. | Do not treat connectome plus nominal weights as sufficient for inhibitory polarity or rhythm claims. |
| Shared extracellular / electrical state | Gap-junction coupling, endogenous field effects, extracellular-space geometry / diffusion barriers / osmotic regime, and local inhibitory driving force can still shift fast synchrony and spike timing outside the chemical synapse graph. | Do not treat a chemical connectome, a nominal inhibitory edge list, or a bounded human perturbation clue as proof that the local electrotonic regime also matched. |
| Neuromodulatory specificity / transmitter context | Mixed arousal proxies, local transmitter signals, receptor / transporter priors, occupancy routes, and displacement / release proxies constrain different quantities. | Do not read one neuromodulatory covariate as if it fixed the current transmitter-specific state of the circuit. |
| Bioenergetic / mitochondrial state | Branch-local ATP reserve, mitochondrial positioning, and fission/fusion still change repeated-burst reliability and dendritic plasticity. | Macro energetic proxies are not ground truth for branch-local energetic state. |
| Glial substrate-routing and astrocyte-state | Glial fuel-routing routes and astrocyte network / ensemble responses remain active variables rather than one background support scalar. | Do not reinterpret persistence as static storage; active maintenance remains visible in the evidence. |
| Clearance / immune support | Transport-side clearance routes and target-defined neuroimmune PET now split into different human proxy families: meningeal lymphatics / CSF-interstitial exchange on one side, and TSPO / CSF1R / COX-2 bounded PET routes on the other. | Do not read macro clearance evidence or one PET target as if the local immune controller were already identified or route-free across families. |
Brain emulation therefore requires more than reproducing neuronal connectivity. Current connectomics, atlas work, and human proxy layers are all valuable, but they push different parts of the problem. Same-day activity matching, cross-day stability, and maintenance-consistent dynamics therefore remain separate claims. On this page the correction rule is simple: each claim must state which hidden-state families were directly measured, which were only constrained by priors or proxies, and which remained latent and forced abstention. Bridging the gap between static structure and dynamic state still requires transcriptomic connectomics[33] plus stepwise verification through state-completeness gates.
302 neurons
~139,000 neurons
Dorkenwald et al., Nature 2024
~71M neurons
~86B neurons
Figure 1
Progress and scale of connectome research. Nematodes (C. elegans), the connectome of the entire nervous system is being developed.[20]In Drosophila, the adult whole brain connectome (approximately 139,000 neurons) was completed in 2024 by the FlyWire project (Dorkenwald et al., 2024).[21]. However, identification of functional dynamics (state-dependent changes in connectivity) and dynamics involving synaptic strength, neuronal modification, and glia remains a challenge. Although saturated reconstructions have been achieved in small cortical volumes in mice,[22], whole-brain-scale reconstruction is currently in progress. The human brain has approximately 86 billion neurons[57].
Quality assurance/synchronization
retained as a distribution
Verification of counterfactual hypothesis
minimum branch set
Figure 2
The chain of "measurement → reconstruction → cause and effect → verification" required for WBE verification. Uncertainty at each stage is carried over to the next stage, and counterfactual equivalence is evaluated through intervention.
Research Program
The weak point of the older version of this section was that it compressed several different evidence classes into one linear phase table: sharable measurement packages, source-imaging validation, multimodal fusion, local perturbation evidence, speech neuroprostheses, and WBE-relevant verification. The primary literature does not support that compression. Tang et al. required participant-specific cooperation for non-invasive semantic reconstruction.[30] Willett et al., Littlejohn et al., and Wairagkar et al. demonstrated strong communication subsystems, not whole-brain equivalence.[108][109][110] Hernandez-Pavon et al., Gogulski et al., and Biabani et al. showed that TMS-EEG interpretation still depends on stimulus conditions, target-specific reliability, and sensory contamination control.[100][124][125] Rohaut et al. and Manasova et al. showed that multimodal gains are bundle-conditioned and modality-specific rather than one interchangeable meter.[128][105] Beiran and Litwin-Kumar further showed that connectome-constrained models remain dynamically degenerate until additional recordings collapse the compatible state space.[140] Therefore, this section now separates repository-executable work from external-dependency experiments and keeps each lane at its own claim ceiling.
Repository-executable work
| Lane | Concrete output that this repository can publish now | Safe claim ceiling |
|---|---|---|
| L0 benchmark packaging | Publish a small public benchmark bundle that fixes dataset identity, split policy, QC log, metric bundle, and expected outputs for one EEG use case. | Third-party rerunnable analysis package, not yet causal or WBE-level evidence. |
| Language-decoding evaluation pack | For brain-to-text or speech decoding claims, publish fixed negative controls such as no-brain, LM-only, time-shuffle, held-out slices, drift slices, and latency / abstention logs. |
Task-limited decoding audit, not evidence of subject-complete readout or WBE. |
| Source / fusion audit package | Publish validation-class registries, uncertainty logs, and fusion cards that force source-imaging and multimodal claims to disclose geometry, conductivity, synchronization, missing-modality policy, and abstention. | Conditional source / fusion hypothesis audit, not automatic state identification. |
| Maintenance-state omission ledger | When reporting connectome-based or macro-measurement models, publish which fast-state and maintenance-state families remain omitted and what claim ceiling follows from those omissions. | Scoped model comparison, not state-complete reconstruction. |
External-dependency tasks
| Task | Why it is external dependency | Safe wording on this site until completed |
|---|---|---|
| Simultaneous HD-EEG/fMRI or TMS-EEG acquisition | Requires IRB approval, specialized hardware, synchronization control, participants, and site operations. | Planned validation route, not current repository evidence. |
| Intracranial stimulation or simultaneous SEEG/ECoG validation | Requires clinical partnership, patient access, and a named external calibration protocol. | External calibration class, not a public benchmark already executed here. |
| Chronic speech neuroprosthesis or real-time voice synthesis | Requires implanted hardware, longitudinal participant operations, and real-time clinical/engineering support. | Local communication-subsystem evidence, not whole-brain verification. |
| WBE-relevant perturbation equivalence across long horizons | Requires stacked observability, intervention, longitudinal maintenance, and body/environment audits far beyond the current repository scope. | Long-range research target, not a scheduled near-term deliverable. |
The site can already publish benchmark rules, audit cards, and claim ceilings. It cannot honestly present simultaneous acquisition, intracranial validation, implanted speech BCIs, or WBE-grade perturbation equivalence as if they were in-repo deliverables. The correct scientific move is to publish the measurement and verification contract first, then attach external experiments only when they are actually executed and logged.
Brain-to-Text update: language priors, streaming, evidence gate
As of 2026-03, the primary literature does not show that a "general-purpose LLM operating theory" is the core of brain decoding. What matters is separating which modality, which task, which generalization condition, and whether a neural contribution beyond the language-prior distribution was actually confirmed. Tang et al.'s non-invasive semantic reconstruction required participant cooperation during both training and application.[30] Horikawa et al.'s Mind Captioning advanced the generation of descriptions for visual content,[11] but the question remains how much meaning can be restored through the communication subsystem alone. Non-invasive speech-perception decoding by Defossez et al.[106] and word decoding by d'Ascoli et al.[107] strengthen different non-invasive routes. On the invasive side, the current literature is no longer one row either: Willett et al., Littlejohn et al., and Wairagkar et al. strengthen throughput / expressivity, Singh et al. strengthens cross-subject transfer initialization, and Karpowicz et al. plus Wilson et al. strengthen alignment-based rescue / unsupervised recalibration.[108][109][110][158][159][160] They do not, however, amount to WBE or self-model reproduction.
In this section, general discussions of RLHF, RAG, and agentic workflows are not the main evidence. The main evidence comes from the primary literature on brain-to-text, speech neuroprostheses, and neural encoding; general-purpose LLM papers are kept only as implementation notes.
Dividing the evidence hierarchy into four parts
| track | What the primary literature now supports | What this still does not justify | Treatment on this site |
|---|---|---|---|
| Non-invasive semantic/caption decoding | Tang et al. showed semantic reconstruction of continuous language from fMRI, and Horikawa et al. showed description generation for visual content from brain activity.[30][11]. | It does not show subject-independent decoding, streaming of everyday interaction, or reproduction of causal internal states. Tang et al. themselves report that both training and application require participant cooperation.[30]. | L1-equivalent meaning restoration |
| Non-invasive word/speech decoding | Defossez et al. demonstrated discrimination of 3-second speech intervals from non-invasive recordings, and d'Ascoli et al. showed that MEG, task design, and dataset size strongly affect large-scale word-decoding performance.[106][107]. | It does not demonstrate open-vocabulary's stable communication, reliable decoding in a single attempt, or long-term drift resistance. | L1 enhancement candidates |
| Invasive language-BCI route family | Willett et al. strengthened high-throughput speech-to-text from intracortical microelectrodes, Littlejohn et al. strengthened streaming brain-to-voice synthesis in 80-ms increments, Wairagkar et al. strengthened instantaneous voice synthesis with paralinguistic control and silence fallback, Singh et al. strengthened cross-subject transfer initialization from distributed intracranial recordings, and Karpowicz et al. plus Wilson et al. strengthened alignment-based rescue and unsupervised recalibration under accumulating neural change.[108][109][110][158][159][160]. | These papers do not yet add up to one solved route for generic transfer, long fixed-decoder durability, low-burden chronic deployment, or whole-brain WBE. They answer different operational questions. | Split L2~L3 route family |
| Neural encoding with LLM embedding | Zada et al. and Goldstein et al. showed that contextual embeddings and representations of semantic relationships can predict brain activity during natural-language tasks.[111][112]. | This is neither thought reading nor decoder. Even if the LLM embedding explains brain responses well, it does not mean that it has "read the inside of the brain." | measurement model / encoding benchmark |
The remaining weakness in this section was that it still let speech BCI sound too much like one monotonic ladder. The current primary literature does not support that shortcut. Willett et al. (2023), Littlejohn et al. (2025), and Wairagkar et al. (2025) support communication throughput / expressivity. Singh et al. (2025) supports cross-subject transfer initialization from distributed intracranial recordings. Karpowicz et al. (2025) and Wilson et al. (2025) support alignment-based rescue / unsupervised recalibration under accumulating neural change. Therefore, this page no longer reads those papers as one interchangeable route to chronic deployability or fixed-decoder stability.
| Operational slice | What the cited papers support | What this page still refuses to merge automatically |
|---|---|---|
| Throughput / expressivity | Large-vocabulary speech decoding, near-real-time voice synthesis, silence fallback, and paralinguistic control in bounded closed-loop communication settings. | These gains are not merged automatically into generic transfer, low-burden chronic deployment, or WBE-like state recovery. |
| Transfer initialization | Shared latent manifolds or grouped decoders can improve initialization when cortical coverage is sparse, heterogeneous, or clinically constrained. | Transfer initialization is not treated as proof that the resulting decoder is stable for long horizons without later rescue or recalibration. |
| Adaptive rescue / recalibration | Alignment or hidden-target inference can restore performance over weeks to months under accumulating neural change. | Adaptive rescue is not treated as fixed-decoder durability, zero-maintenance operation, or stable-neuron truth by default. |
Minimum required evaluation pack
- Isolation of neural contribution:
no-brain,time-shuffle,trial-shuffle,LM-only, andno-LM. When searching from a candidate set, specify the candidate set size. - Boundary of generalization: Report held-out sentences, held-out stories, held-out vocabulary, cross-day, cross-task, and cross-subject separately, and state openly if participant cooperation or individual adaptation is required.[30][107].
- Real-time metrics: If you claim streaming performance, do not report words per minute alone. Also report P50 / P95 / P99 latency, silence / abstention rate, dropout, recalibration burden, and recovery time.[109][110].
- Longitudinal operation split: State whether the paper is claiming same-session throughput, transfer initialization, a bounded fixed-decoder interval, or adaptive rescue. Report
time since last supervised calibration, the stabilization / alignment strategy, and if chronic microelectrode decoding is central, a unit-identity or channel-stability audit.[158][159][160]. - Reproduction log:The brain encoder, language model, vocoder, context window, beam width, external corpus, prompt, and calibration procedure are fixed, and if the model update crosses evaluation, it will be treated as a separate run.
Operation rules on Mind-Upload side
- Conditions for claiming L1:Show a neural contribution that exceeds the LM-only/shuffle baseline and do not hide candidate sets or evaluation conditions.
- Conditions for claiming L2:Include held-out conditions, cross-day or cross-task generalizations, and abstinence when confidence is low.
- Route typing for invasive language BCIs:State explicitly whether the evidence is throughput / expressivity, transfer initialization, a bounded fixed-decoder slice, or adaptive rescue. This page does not let rescue results silently stand in for fixed-decoder stability.[158][159][160].
- Conditions for claiming L3:Submit streaming log, tail latency, silence/freeze, fixed decoder interval, time since last supervised calibration, recalibration burden, and disclose closed-loop failure mode.
- Prohibitions for higher claims:Success in decoding, resemblance in embedding, and naturalness of conversation cannot be interpreted as emulate/WBE/preservation of identity.
Measuring consciousness with EEG: reading perturbation indicators as the main axis and resting indicators as an aid
EEG is effective when handling consciousness-related information in the "measurement" stage of WBE, but it is not a device that can independently determine whether someone is "conscious or not." Current primary literature strongly supports that (a) perturbation response complexity is a candidate state-level benchmark, (b) no-report paradigms and criterion placement control are design conditions to reduce report/post-perceptual confound, (c) resting-state complexity/criticality is promising but remains an auxiliary readout, and (d) multimodal/multisite validation is required for clinical operation.[47][51][52][100][102][103][104][105][127][128][113]. Therefore, in this section we treat EEG not as a “single consciousness meter” but as a bundle of indicators with different strengths of evidence.
On this site, no-report / criterion placement is the construct-validity gate, PCI / PCI-ST is the main benchmark candidate when sensory control and reliability logs are present, resting-state complexity / criticality remains an auxiliary proxy that must be calibrated within the same cohort, and clinical claims will not be promoted unless they come from a multimodal panel that exceeds the behavior-only baseline.
Visual and auditory no-report studies are design conditions, not direct proof that EEG alone can serve as a bedside consciousness meter. On this site, cross-modal no-report evidence is used only as evidence for confound control and is kept separate from EEG benchmark evidence.[102][103][113].
Earlier versions of this page had the right direction, but the latest primary literature shows that even for the same "awareness index," the claim ceiling changes unless construct validity, perturbational validity, same-cohort calibration, and incremental validity over behavior are audited as separate gates. Remove criterion-placement control and the interpretation of no-report breaks down. Remove sensory contamination control and target-specific reliability, and the PCI / TMS-EEG readout becomes unstable. Remove same-cohort calibration and resting-state indicators remain proxies. Remove the behavior-only baseline and multimodal panels cannot claim deployability. Promotion therefore depends on whether the 4 Gates of Verification are passed.[113][124][125][126][127][128].
5 conditions to fix first in this section
- Remove report confounds first: No-report paradigms are not consciousness readouts by themselves; they are design conditions for separating perception from post-report processing.[48][102][103].
- Audit criterion placement in a separate log: Even with a no-report setup, leaving the response criterion unspecified risks picking up judgment strategy rather than conscious content.[113].
- Keep perturbation benchmarks on the main axis: PCI / PCI-ST is a powerful benchmark across altered states and disorders of consciousness, but TMS-EEG stimulus conditions and artifact management are prerequisites.[47][51][100].
- Use resting indicators only with external calibration: LZ complexity and criticality are promising, but they should not be used as primary judgments unless calibrated against perturbation metrics and clinical outcomes.[50][52][104].
- Require multimodal external validation for clinical claims: We prioritize incremental validity from bundles of behavior, imaging, and electrophysiology rather than the apparent strength of a single indicator.[105][127][128].
4 tracks with different strengths of evidence
| Track | What the primary literature supports now | What it still does not justify | Position on this site |
|---|---|---|---|
| PCI / PCI-ST | Complex responses to perturbation are candidates for state-level benchmarks across anesthesia, sleep, and disorders of consciousness.[47][51][55]. | They do not by themselves determine conscious content, personal identity, or theory validity. In prefrontal TMS-EEG, reliability and sensory contamination are target- and window-dependent, so omitting stimulus conditions, controls, or artifact windows makes benchmark comparisons invalid.[124][125]. | Main benchmark candidate |
| No-report / criterion placement | The literature increasingly supports these designs as ways to separate post-perceptual processing from response criterion in both vision and audition.[48][102][103][113]. | They do not by themselves become bedside meters, and without logs for criterion placement and report strategy, neural markers may still reflect judgment strategy rather than conscious content. | Confound control |
| Resting complexity/criticality | Changes associated with anesthesia, links to PCI, and the possibility of classifying consciousness without perturbation have been reported.[50][52][56][126]. | These are not replacements for perturbation-based indicators. Because spontaneous and evoked markers can dissociate in MCS,[104] they will not be promoted to primary judgment unless calibrated against PCI, behavioral outcomes, and clinical outcomes in the same cohort. | Auxiliary / exploration track |
| Multimodal clinical panel | A multimodal panel that combines behavioral evaluation, HD-EEG, MRI, PET, and clinical variables may outperform behavior-only baselines for diagnosis and prognosis in coma / DoC.[105][127][128]. | It still cannot be described as ready to deploy unless it shows incremental validity, out-of-site generalization, and robustness to missing modalities beyond the baseline. | Deployability gate |
4 gates that determine promotion
| gate | Minimum requirement | Claim to stop if it is not passed |
|---|---|---|
| Construct validity | Separate no-report from report, log criterion placement separately, and predefine failure conditions. | Do not claim that a neural marker directly reads conscious content itself. |
| Perturbational validity | TMS-EEG or intracranial perturbation conditions, sensory control, artifact windows, and target-specific reliability. | Do not call PCI-like values alone a state-level benchmark. |
| Same-cohort calibration | Calibrate PCI, behavior, clinical outcomes, and pipeline sensitivity analyses within the same cohort. | Do not call a resting-state metric a standalone bedside meter. |
| Incremental validity | Behavior-only baseline comparisons, external-site generalization, and calibration error under missing-modality conditions. | Do not describe a multimodal panel as ready to deploy. |
The remaining operational shortcut was that a familiar label such as IIT, PCI, criticality, or multimodal could still sound like one continuous evidential ladder. To stop that compression, the public operating artifact is now Verification: Consciousness Readout Card, which forces theory label, claimed gate, perturbation log, same-cohort denominator, and deployability denominator onto one line before the claim ceiling is raised.
Research priorities (A→C)
| Track | Aim | Why prioritize now? | Conditions to proceed |
|---|---|---|---|
| A | Establish a perturbation benchmark | It currently offers the strongest verification basis and is comparatively easy to compare across state differences.[47][51]. | Publish TMS-EEG stimulus-system logs, artifact windows, and test-retest reliability.[100]. |
| B | Calibrate the resting proxy to the benchmark | It may reduce equipment burden, but it has not yet been shown to replace perturbation-based indices.[52][104]. | Proceed only with same-cohort calibration against PCI, behavioral outcomes, and clinical outcomes. |
| C | External validation with multimodal/multicentre | Clinical value is determined by incremental predictive power and cross-site reproducibility, not by single-lab success.[105]. | Exceed the behavior-only baseline, be robust to missing measurements, and disclose calibration errors. |
Roadmap (assuming graduate school)
Phase 0: Infrastructure development (~1 year) - solidify “measurable” and “reproducible”
Fix the EEG analysis pipeline so that the resting state index and report/criterion confound control can be reproduced on the same data.
- Reproducible analysis including preprocessing, artifact removal, and logging (BIDS, etc.)
- In published data, associated with anesthesia/sedationComplexity/Spectrum/ConnectivityReproduce the changes in[23][50][53]
- Design the task so that both conditions with and without report can be run concurrently, and align with both visual/auditory no-report literature and criterion placement audits.[102][103][113]
- First check the test-retest reliability of LZ complexity, criticality, and spectral index using the same data.
Phase 1: Perturbation benchmark (equivalent to years 1-2)—make the “PCI/PCI-ST line” auditable
The response complexity to perturbations (TMS, sensory stimulation, etc.) is used as the main benchmark to enable state-level comparison.
Phase 2: Calibration of resting proxy (equivalent to 2nd to 3rd year)—“no perturbation” is included as an adjunct rather than a substitute
Externally calibrate resting EEG complexity and criticality to perturbation metrics, behavioral, and clinical outcomes. The purpose is not to replace PCI, but to quantify how far it can be used as a proxy.
Maschke et al. showed a relationship between criticality index and PCI under anesthesia induction.[52], Casarotto et al. reported that spontaneous markers and evoked markers can dissociate in MCS.[104]. Therefore, we do not say "sufficient without perturbation", but instead audit condition-specific proxy performance.
Phase 3: External validation (equivalent to years 3-4)—showing multimodal gains rather than “single indicators”
In the final stage, we bundle EEG indicators with behavior/imaging/clinical variables and evaluate incremental validity from the perspectives of diagnosis, prognosis, and tolerance for missing data.
- Rather than reporting the AUC of a single metric, it reports how much it beats the baseline and maintains calibration.[105]
- Evaluate deployability, including facility differences, measurement burden, and missing measurements
ToDos from the past 1 to 3 months (in order of least effective)
- Fix the analysis pipeline and log schema and test-retest reliability of resting indicators first
- Correlate report presence/absence conditions with visual/auditory no-report documents and audit criterion placement in a separate log.[102][103][113]
- Reproduce changes associated with anesthesia/sedation once using public data[50][52]
- Pre-register stimulus logs and artifact windows for PCI/PCI-ST lines depending on whether or not to introduce perturbations.[51][100]
- Decide the public metrics for multimodal validation first to avoid misreading the resting proxy as a bedside meter.[105]
Technical position and goals
This roadmap does not aim to prove the correctness of a particular theory of consciousness, but rather to make it possible to compare perturbation indicators, no-report control, resting state proxies, and multimodal clinical evaluations using the same audit schema. What Ferrante et al.'s adversarial test showed was not ``convergence to a single theory,'' but the point that theory and task design should be audited separately.[54].
Therefore, the goal of this section is not to declare a "world-standard standalone consciousness meter." As the multisite study by Manasova et al. shows, diagnosis and prognosis can improve when multimodal integration is added to behavioral assessment, but the indicators do not substitute for one another.[105] The goal of this site is to accumulate a public benchmark that includes pre-registration, external validation, and abstention.
Technical Proposals
This is a concrete technology proposal that supports the research roadmap. Here, we have integrated the proposals into the main text and organized them to provide an overview of the evidence and implementation focus.
Integration of measurement QA, synchronization and BIDS compliance
Measurement-quality visualization (impedance / noise floor / CMRR) and synchronization standardization should be linked directly to the EEG-BIDS metadata structure. EEG-BIDS provides a framework centered on explicit metadata such as dataset_description.json, eeg.json, channels.tsv, and electrodes.tsv, which improves reproducibility and portability.
- Link the structure and essential metadata of BIDS-EEG with the QA log to leave a reproducible "measurement trail"[83]
- Multimodal synchronization standardizes LSL sample timestamp and jitter correction[84]
→ To organize the proposal status and external dependencies, please refer to the contribution guide.
Enhanced preprocessing reproducibility and connectivity ceilings
The emphasis is on ensuring reproducible artifact and line-noise control first, while keeping a separate audit ceiling for connectivity claims. Automatic ASR removal and ZapLine can improve reproducibility of cleanup, but they do not by themselves solve volume conduction, source leakage, or directional identifiability for wPLI, source-space connectivity, or STE.[85][86][129][130]
- ASR is being evaluated as an automatic artifact removal method, and guidelines for parameter ranges have been provided.[85]
- ZapLine is a proven method for line noise removal and can be applied to EEG/MEG[86]
- wPLI reduces sensitivity to some zero-lag mixing and noise, but it is not a leak-proof inter-areal coupling meter; simulated EEG and source-space analyses still show source leakage and ghost-interaction ceilings.[87][129][130]
- STE is useful as a directed-dependence estimator, but observational EEG alone does not settle causality. Ye et al. evaluated STE under TMS perturbation precisely because causality is difficult to identify from observations alone.[88][131]
- Recent benchmarking still finds that rereferencing, epoch design, and metric choice materially change sensor-space connectivity estimates, so this site treats connectivity pipelines as auditable configurations rather than stable readouts.[132]
→ To organize the proposal status and external dependencies, please refer to the contribution guide.
Motion-tolerant OPM-MEG / hyperscanning with shielding and calibration gates
Recent OPM-MEG work moves standing, ambulatory movement, and even two-person interaction from "not measurable at all" to proof-of-concept under disclosed engineering conditions. The safe reading is still narrower than ``wearable MEG solves naturalistic measurement.'' Wearability helps because the sensors move with the head, but zero-field operation, active field control, sensor calibration, crosstalk management, and anatomy choice remain first-class conditions for source modeling and comparison.[89][133][134][135]
- Boto et al. (2018) and Seymour et al. (2021) showed that OP-MEG can recover neuromagnetic signals during seated and standing/mobile paradigms, but still in magnetically shielded environments with motion tracking and interference suppression.[89][133]
- Holmes et al. (2023) extended this to ambulatory motion with matrix-coil active shielding, and Holmes et al. (2023) also demonstrated proof-of-concept OPM hyperscanning in two-person interactive tasks. The advance is therefore movement-tolerant macro electrophysiology under active magnetic control, not shield-free open-world sensing.[134][135]
- Holmes et al. (2025) showed that lighter and cheaper shielded rooms can become usable when tSSS and active compensation are added, which means ``more portable'' still does not mean ``no specialized magnetic environment.''[136]
- Iivanainen et al. (2022) and Rhodes et al. (2025) show that source modeling still depends on sensor localization / gain calibration and on the anatomy route; pseudo-MRI is useful when MRI is difficult, but individual MRI remains the gold standard.[137][138]
- Wu et al. (2025) showed that array crosstalk remains a sensor-level engineering limit in denser OPM layouts, so higher channel density is not read here as automatic measurement maturity.[139]
→ To organize the proposal status and external dependencies, please refer to the contribution guide.
Identifiability and Causal Intervention (PCI/do-calculus)
Counterfactual hypotheses cannot be tested through observation alone; they require intervention data. PCI, as TMS-EEG-based perturbation complexity, can therefore serve as one external benchmark, but it is not ground truth by itself. It should be used for comparing intervention-response distributions only when TMS-EEG recommendations on stimulation site, intensity, auditory masking, and myoelectric / stimulation artifact windows are satisfied.[90][100].
- The causal hierarchy consists of three layers: observation, intervention, and counterfactual hypotheticals.[91]
- PCI has been proposed as a consciousness index using TMS-EEG response complexity, but should be read in conjunction with OOD conditions, calibration, and abstention conditions.[90]
→ To organize the proposal status and external dependencies, please refer to the contribution guide.
Keep irreversibility logs separate from cost and theory-selection claims
Logical complexity, wall-plug power, and irreversibility-derived quantities are different objects and must be logged separately. On this site, irreversibility remains an exploratory auxiliary lane unless its route card exposes the state definition, closure assumptions, reverse-transition support, and physiology-side grounding. Main promotion still depends on perturbation, OOD, external validation, and abstention conditions.
- Lynn et al. (2021) and Ishihara & Shimazaki (2025) support route-dependent nonequilibrium summaries, not direct microscopic dissipation or WBE-ready energy cost.
- IIT-related MIP search remains an engineering tractability problem rather than evidence that a thermodynamic KPI was measured.[93]
→ To organize the proposal status and external dependencies, please refer to the contribution guide.
Causal preservation and thermodynamic readouts answer different questions
Causal claims ask whether interventions and counterfactual structure are preserved; thermodynamic logs ask whether a chosen observable shows nonequilibrium signatures under stated assumptions. Because these objects fail differently, this page does not let an irreversibility result stand in for causal preservation, and it does not let a causal benchmark stand in for energetic grounding.
→ To organize the proposal status and external dependencies, please refer to the contribution guide.
Raise claims only after causal and thermodynamic route cards are explicit
When PCI[90], SCM / counterfactual arguments[91], and irreversibility logs appear in the same proposal, each keeps its own ceiling. PCI remains one perturbation benchmark under explicit TMS-EEG conditions, model-based causal claims remain conditional on their observation and validation assumptions, and irreversibility remains auxiliary unless signal route, state definition, closure, physiology-side grounding, and cost isolation are disclosed. None of the three is treated here as a standalone pass/fail indicator.
→ To organize the proposal status and external dependencies, please refer to the contribution guide.
Limitations and Epistemic Humility
As a research note, we recognize and specify the following structural limitations. These are not "weaknesses" but requirements of scientific integrity.
theoretical limits
- Avoiding hard problems:This page uses functional equivalence as an operational definition, but the identity of phenomenal consciousness is not subject to verification. Whether functionally complete emulation is a sufficient condition for ``having consciousness'' cannot be determined within the framework of this project (Chalmers, 1995).
- Limits of theoretical neutrality:Although it declares that it is ``independent of theory,'' the selection of indicators such as PCI itself may include implicit assumptions that favor IIT. This potential bias cannot be completely eliminated.
- FEP falsifiability:The free energy principle has been criticized as being too comprehensive and unfalsifiable. This project adopts FEP as the "implementation principle," but its validity needs to be separately verified using empirical results.
- Implications of Unfolding Argument:Doerig et al. (2019)[40]If the above argument is correct, consciousness in the sense of IIT (Φ>0) does not arise in principle in digital emulation. A shift to neuromorphic infrastructure is essential, but its engineering feasibility remains untested.
engineering limits
- Observability upper limit: Non-invasive EEG / MEG / fMRI are macro proxies, and advances in local connectomics such as Dorkenwald et al. (2024) or MICrONS Consortium et al. (2025) do not directly imply state-complete observation of the human whole brain.
- Identifiability upper limit: The EEG inverse problem can be improved, but dependence on head model, conductivity, and candidate model family remains.[5][79][96] Therefore, high fit quality and localization accuracy cannot be treated as unique reconstruction of internal state.
- Maintenance-state upper limit: Hengen et al. (2016), Torrado Pacheco et al. (2021), Xu et al. (2024), Looser et al. (2024), and Cahill et al. (2024) together show that sleep / wake state, myelin, and glial / metabolic support remain separate variables. Reproducing same-day behavior and maintaining cross-day state are different problems.
- Upper limit of intervention scope: TMS-EEG recommendations, bidirectional BCI, and adaptive DBS support local causal gain, but not whole-brain branch-equivalence. Comparability cannot be guaranteed without disclosing latency, jitter, and artifacts.
- Current state of reproducibility:The L0 (third-party reproducibility) set forth by this project has not been achieved at this time. The repository mainly contains website content, and executable code, data, and environment information are not made public.
- The severity of the connectome-dynamics gap:Although FlyWire and MICrONS have strengthened the structural scaffold, mapping structure to function, maintenance, and intervention response remains a core open question.
Limits of project management
- Personal project:At this point, this is a personal research note and not a peer-reviewed study. External verification and expansion of joint research are necessary.
- Discrepancy between design and implementation:Many of the design policies described on this page remain at the document level and have not been made public as implementation code, test data, or evaluation results.
About
Yasufumi Nakata
Belongs to Keio University Faculty of Environment and Information Studies / Atsushi Aoyama Laboratory.
This site is a public research note regarding mind upload research.
References
- Fleming, S. M., et al. (2023). Open letter regarding "The integrated information theory of consciousness". Neuroscience of Consciousness, 2023(1), niad001. doi:10.31234/osf.io/zsr78
- Tononi, G., et al. (2016). Integrated information theory: from consciousness to its physical substrate. Nat. Rev. Neurosci., 17, 450–461. doi:10.1038/nrn.2016.44
- Hanson, J. R. (2023). On the non-uniqueness problem in integrated information theory. Neuroscience of Consciousness, 2023(1), niad014. doi:10.1093/nc/niad014
- Parfit, D. (1984). Reasons and Persons. Oxford University Press.
- Michel, C. M., & Brunet, D. (2019). EEG source imaging: a practical review of the methodology. Frontiers in Neurology, 10, 325. doi:10.3389/fneur.2019.00325
- Koch, C., Massimini, M., Boly, M., & Tononi, G. (2016). Neural correlates of consciousness: Progress and problems. Nature Reviews Neuroscience, 17(5), 307–321. doi:10.1038/nrn.2016.22
- Yamakawa, H., et al. (2024). Technology roadmap toward the completion of whole-brain architecture with BRA-driven development. Cognitive Systems Research, 88, 101300. doi:10.1016/j.cogsys.2024.101300
- Sandberg, A., & Bostrom, N. (2008). Whole Brain Emulation: A Roadmap. Future of Humanity Institute, Oxford University. Link
- Logothetis, N. K. (2008). What we can do and what we cannot do with fMRI. Nature, 453(7197), 869-878. doi:10.1038/nature06976
- Yuste, R., et al. (2017). Four ethical priorities for neurotechnologies and AI. Nature, 551(7679), 159-163. doi:10.1038/551159a
- Horikawa, T., et al. (2025). Mind captioning: Evolving descriptive text of mental content from human brain activity. Science Advances, 11(45), eadw1464. doi:10.1126/sciadv.adw1464
- Kozlov, M. (2025). 'Mind-captioning' AI decodes brain activity to turn thoughts into text. Nature, 647(8089), 297. doi:10.1038/d41586-025-03624-1(Note: News article. For primary research, see [11])
- Friston, K. J., Harrison, L., & Penny, W. (2003). Dynamic causal modelling. NeuroImage, 19(4), 1177-1202. doi:10.1016/S1053-8119(03)00202-7
- Friston, K. (2010). The free-energy principle: a rough guide to the brain. Nature Reviews Neuroscience, 11(2), 127–138. doi:10.1038/nrn2787
- Vorwerk, J., et al. (2014). A guideline for head volume conductor modeling in EEG and MEG. NeuroImage, 100, 590–607. doi:10.1016/j.neuroimage.2014.06.040
- Markram, H. (2006). The Blue Brain Project. Nature Reviews Neuroscience, 7(2), 153-160. doi:10.1038/nrn1860
- Albantakis, L., et al. (2023). Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms. PLOS Computational Biology, 19(10), e1011465. doi:10.1371/journal.pcbi.1011465
- Wipf, D., & Nagarajan, S. (2009). A unified Bayesian framework for MEG/EEG source imaging. NeuroImage, 44(3), 947–966. doi:10.1016/j.neuroimage.2008.02.059
- Özçete, Ö. D., et al. (2024). Mechanisms of neuromodulatory volume transmission. Molecular Psychiatry. doi:10.1038/s41380-024-02608-3
- Cook, S. J., et al. (2019). Whole-animal connectomes of both Caenorhabditis elegans sexes. Nature, 571(7763), 63–71. doi:10.1038/s41586-019-1352-7
- Scheffer, L. K., et al. (2020). A connectome and analysis of the adult Drosophila central brain. eLife, 9. doi:10.7554/eLife.57443
- Kasthuri, N., et al. (2015). Saturated reconstruction of a volume of neocortex. Cell, 162(3), 648–661. doi:10.1016/j.cell.2015.06.054
- Purdon, P. L., et al. (2013). Electroencephalogram signatures of loss and recovery of consciousness from propofol. PNAS, 110(12), E1142–E1151. doi:10.1073/pnas.1221180110
- Schumer, C., et al. (2025). Management of Individuals' Neural Data (MIND) Act of 2025. U.S. Senate Bill. (Note: Proposed bill as of Jan 12, 2026).
- Gorgolewski, K. J., et al. (2016). The brain imaging data structure, a format for organizing and describing outputs of neuroimaging experiments. Scientific Data, 3. doi:10.1038/sdata.2016.44
- Pernet, C. R., et al. (2019). EEG-BIDS, an extension to the brain imaging data structure for electroencephalography. Scientific Data, 6(1). doi:10.1038/s41597-019-0104-8
- Ienca, M., & Andorno, R. (2017). Towards new human rights in neuroscience. Life Sciences, Society and Policy, 13(1), 5. doi:10.1186/s40504-017-0050-1
- Ji, Z., et al. (2023). Survey of Hallucination in Natural Language Generation. ACM Computing Surveys, 55(12). doi:10.1145/3571730
- Huth, A. G., et al. (2016). Natural speech reveals the semantic maps that tile human cerebral cortex. Nature, 532(7600), 453–458. doi:10.1038/nature17637
- Tang, J., et al. (2023). Semantic reconstruction of continuous language from non-invasive brain recordings. Nature Neuroscience, 26(5), 858–866. doi:10.1038/s41593-023-01304-9
- Boly, M., et al. (2017). Are the neural correlates of consciousness in the front or in the back of the cerebral cortex? Clinical and neuroimaging evidence. Journal of Neuroscience, 37(40), 9603–9613. doi:10.1523/JNEUROSCI.3218-16.2017
- Whitehead, A. N. (1929). Process and Reality: An Essay in CosmologyMacmillan. (Note: A classic of process philosophy. The year of publication and publisher may vary depending on the edition.)
- Gamlin, C. R., et al. (2025). Connectomics of predicted Sst transcriptomic types in mouse visual cortex. Nature, 640(8058), 497–505. doi:10.1038/s41586-025-08805-6
- Jun, S., Altmann, A., Sadaghiani, S., et al. (2025). Modulatory Neurotransmitter Genotypes Shape Dynamic Functional Connectome Reconfigurations. Journal of Neuroscience, 45(10). doi:10.1523/JNEUROSCI.1939-24.2025 (Open Access: PMC11884390)
- Doerig, A., et al. (2019). The unfolding argument: Why IIT and other causal structure theories cannot explain consciousness. Consciousness and Cognition, 72, 49–59. doi:10.1016/j.concog.2019.04.002
- Santello, M., et al. (2019). Astrocyte-neuron interactions: from synapses to networks and behavior. Neuron, 103(6), 985-1000. doi:10.1016/j.neuron.2019.08.024
- Melloni, L., et al. (2023). An adversarial collaboration protocol for testing contrasting predictions of global neuronal workspace and integrated information theory. PLoS ONE, 18(3), e0282855. doi:10.1371/journal.pone.0282855
- Dehaene, S., Kerszberg, M., & Changeux, J. P. (1998). A neuronal model of a global workspace in effortful cognitive tasks. Proceedings of the National Academy of Sciences, 95(24), 14529-14534. doi:10.1073/pnas.95.24.14529
- Tononi, G. (2015). Integrated information theory. Scholarpedia, 10(1), 4164. doi:10.4249/scholarpedia.4164(Note: For a discussion of IIT's Intrinsic Existence axiom, see this review and [17])
- Friston, K. (2017). Active inference: a process theory. Neural Computation, 29(1), 1–49. doi:10.1162/NECO_a_00912
- Parr, T., & Friston, K. J. (2019). Generalised free energy and active inference. Biological cybernetics, 113(4), 495-513. doi:10.1007/s00422-019-00803-8
- Casali, A. G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., Casali, K. R., ... & Massimini, M. (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine, 5(198), 198ra105. doi:10.1126/scitranslmed.3006294
- Tsuchiya, N., et al. (2015). No-Report Paradigms: Extracting the True Neural Correlates of Consciousness. Trends in Cognitive Sciences, 19(12), 757–770. doi:10.1016/j.tics.2015.10.002
- Massimini, M., et al. (2005). Breakdown of cortical effective connectivity during sleep. Science, 309(5744), 2228–2232. doi:10.1126/science.1117256 (Open Access: air.unimi.it)
- Schartner, M., et al. (2015). Complexity of multi-dimensional spontaneous EEG decreases during propofol induced general anaesthesia. PLOS ONE, 10(8). doi:10.1371/journal.pone.0133532 (Open Access: PMCID: PMC4529106)
- Comolatti, R., et al. (2019). A fast and general method to empirically estimate the complexity of brain responses to transcranial and intracranial stimulations. Brain Stimulation, 12(5), 1280–1289. doi:10.1016/j.brs.2019.05.013
- Maschke, C., et al. (2024). Critical dynamics in spontaneous EEG predict anesthetic-induced loss of consciousness and perturbational complexity. Communications Biology, 7(1). doi:10.1038/s42003-024-06613-8
- Anusha, A. S., et al. (2021). Brain Functional Connectivity as Biomarker for Propofol-Induced Alterations of Consciousness. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 1928–1931. doi:10.1109/EMBC46164.2021.9629617
- Ferrante, O., et al. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature, 642(8066), 133–142. doi:10.1038/s41586-025-08888-1
- Sinitsyn, D. O., et al. (2020). Detecting the potential for consciousness in unresponsive patients using the perturbational complexity index. Brain Sciences, 10(12), 917. doi:10.3390/brainsci10120917
- Shew, W. L., & Plenz, D. (2013). The functional benefits of criticality in the cortex. The Neuroscientist, 19(1), 88–100. doi:10.1177/1073858412445487
- Azevedo, F. A., Carvalho, L. R., Grinberg, L. T., Farfel, J. M., Ferretti, R. E., Leite, R. E., ... & Herculano-Houzel, S. (2009). Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain. Journal of Comparative Neurology, 513(5), 532-541. doi:10.1002/cne.21974
- Weber, C. (2025). The multiplicity objection against uploading optimism. Synthese, 205(6). doi:10.1007/s11229-025-05057-9
- Clowes, R. W. (2021). Slow Continuous Mind Uploading. Studies in Brain and Mind, 18, 161–183. doi:10.1007/978-3-030-72644-7_8
- Vaswani, A., et al. (2017). Attention Is All You Need. Advances in Neural Information Processing Systems (NeurIPS). arXiv:1706.03762
- Ouyang, L., et al. (2022). Training language models to follow instructions with human feedback. arXiv:2203.02155. arXiv
- Rafailov, R., et al. (2023). Direct Preference Optimization: Your Language Model is Secretly a Reward Model. arXiv:2305.18290. arXiv
- Dettmers, T., et al. (2023). QLoRA: Efficient Finetuning of Quantized LLMs. arXiv:2305.14314. arXiv
- Lewis, P., et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. arXiv:2005.11401. arXiv
- Yao, S., et al. (2022). ReAct: Synergizing Reasoning and Acting in Language Models. arXiv:2210.03629. arXiv
- Schick, T., et al. (2023). Toolformer: Language Models Can Teach Themselves to Use Tools. arXiv:2302.04761. arXiv
- Manakul, P., et al. (2023). SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models. arXiv:2303.08896. arXiv
- Friston, K. J., et al. (2017). Active inference, mathematical constitution, and pure consciousness. Frontiers in Psychology, 8, 1322.
- Zanichelli, N., et al. (2025). State of Brain Emulation Report 2025. arXiv:2510.15745.
- Koulouri, A. (2025). Bayesian model parameter learning in linear inverse problems. Machine Learning: Science and Technology.
- Nzakuna, P. S., et al. (2025). Monte Carlo-based Strategy for Assessing the Impact of EEG Data Uncertainty. IEEE Transactions on Instrumentation and Measurement.
- Laukkonen, R. E., Friston, K., & Chandaria, S. (2025). A beautiful loop: An active inference theory of consciousness. Neuroscience & Biobehavioral Reviews, 183, 106296. doi:10.1016/j.neubiorev.2025.106296
- Clark, A. (2013). The Mark of the Mental: In Search of the Markov Blanket. In: The Mark of the Mental. Oxford University Press.
- Feng, Z., et al. (2025). Block-Champagne: Imaging Extended E/MEG Source Activation with Empirical Bayesian Uncertainty Quantification. IEEE Transactions on Medical Imaging. doi:10.1109/TMI.2025.3642620
- Aydin, U., Vorwerk, J., Küpper, P., et al. (2019). Influence of Head Tissue Conductivity Uncertainties on EEG Dipole Reconstruction. Frontiers in Neuroscience, 13, 531. doi:10.3389/fnins.2019.00531
- Whyte, C. J., Hohwy, J., Baltieri, M., et al. (2025). On the minimal theory of consciousness implicit in active inference. Physics of Life Reviews. doi:10.1016/j.plrev.2025.11.002
- Li, J., et al. (2025). Computational complexity reduction in Integrated Information Theory via low-dimensional embedding. Chaos, Solitons & Fractals.
- Cai, C., et al. (2021). Robust estimation of noise for electromagnetic brain imaging with the Champagne algorithm. NeuroImage.
- Pernet, C. R., Appelhoff, S., Gorgolewski, K. J., et al. (2019). EEG-BIDS, an extension to the brain imaging data structure for electroencephalography. Scientific Data, 6, 103. doi:10.1038/s41597-019-0104-8
- Kothe, C., Shirazi, S. Y., Stenner, T., et al. (2025). The lab streaming layer for synchronized multimodal recording. Imaging Neuroscience, 3, IMAG.a.136. doi:10.1162/IMAG.a.136
- Chang, C.-Y., Hsu, S.-H., Pion-Tonachini, L., & Jung, T.-P. (2018). Evaluation of Artifact Subspace Reconstruction for automatic EEG artifact removal. Proc. IEEE EMBC. doi:10.1109/EMBC.2018.8512547
- de Cheveigne, A. (2020). ZapLine: A simple and effective method to remove power line artifacts. NeuroImage, 207, 116356. doi:10.1016/j.neuroimage.2019.116356
- Vinck, M., Oostenveld, R., van Wingerden, M., Battaglia, F., & Pennartz, C. M. A. (2011). An improved index of phase-synchronization for electrophysiological data in the presence of volume-conduction, noise and sample-size bias. NeuroImage, 55(4), 1548–1565. doi:10.1016/j.neuroimage.2011.01.055
- Staniek, M., & Lehnertz, K. (2008). Symbolic Transfer Entropy. Physical Review Letters, 100, 158101. doi:10.1103/PhysRevLett.100.158101
- Boto, E., Holmes, N., Leggett, J., et al. (2018). Moving magnetoencephalography towards real-world applications with a wearable system. Nature, 555, 657–661. doi:10.1038/nature26147
- Casali, A. G., Gosseries, O., Rosanova, M., et al. (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine, 5(198), 198ra105. doi:10.1126/scitranslmed.3006294
- Correa, J. D., Lee, S., & Bareinboim, E. (2021). Nested Counterfactual Identification from Arbitrary Surrogate Experiments. arXiv:2107.03190. arXiv
- Seifert, U. (2012). Stochastic thermodynamics, fluctuation theorems and molecular machines. Reports on Progress in Physics, 75(12), 126001. doi:10.1088/0034-4885/75/12/126001
- Kitazono, J., Kanai, R., & Oizumi, M. (2018). Efficient Algorithms for Searching the Minimum Information Partition in Integrated Information Theory. Entropy, 20(3), 173. doi:10.3390/e20030173
- Dorkenwald, S., et al. (2024). Neuronal wiring diagram of an adult brain. Nature, 634, 124–138. doi:10.1038/s41586-024-07558-y
- Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200-219.
- Frässle, S., et al. (2021). Regression DCMs for fMRI. NeuroImage, 227, 117566. doi:10.1016/j.neuroimage.2020.117566
- Hoel, E., Albantakis, L., & Tononi, G. (2016). Can the macro beat the micro? Integrated information across spatiotemporal scales. Neuroscience, 311, 393–401. doi:10.1016/j.neuroscience.2016.09.049
- Seifert, U. (2012). Stochastic thermodynamics, fluctuation theorems and molecular machines. Reports on Progress in Physics, 75(12), 126001. doi:10.1088/0034-4885/75/12/126001
- Staniek, M. & Lehnertz, K. (2008). Symbolic Transfer Entropy. Physical Review Letters, 100(15), 158101. doi:10.1103/PhysRevLett.100.158101
- Hernandez-Pavon, J. C., Metsomaa, J., Mutanen, T. P., et al. (2023). TMS combined with EEG: Recommendations and open issues. Brain Stimulation, 16(2), 350–366. doi:10.1016/j.brs.2023.02.009
- Srinivasan, R., Nunez, P. L., Tucker, D. M., Silberstein, R. B., & Cadusch, P. J. (1996). Spatial sampling and filtering of EEG with spline Laplacians to estimate cortical potentials. Brain Topography, 8(4), 355–366. doi:10.1007/BF01186911
- Cohen, M. A., et al. (2024). Neural signatures of visual awareness independent of postperceptual processing. Cerebral Cortex, 34(11), bhae415. doi:10.1093/cercor/bhae415
- Dellert, T., Balster, M., Schlossmacher, I., Bruchmann, M., Moeck, R., Straube, T., & Jacobsen, T. (2025). Neural correlates of consciousness in an auditory no-report fMRI study. Current Biology. doi:10.1016/j.cub.2025.10.026
- Casarotto, S., Hassan, M., Rosanova, M., Sarasso, S., Derchi, C. C., Trimarchi, P. D., Viganò, A., Russo, S., Moraes, T., Naro, A., Pigorini, A., Leonardi, G., Gosseries, O., Laureys, S., Boly, M., Bodart, O., Casali, A. G., Colombo, M. A., Casali, K. R., & Comolatti, R. (2024). Dissociations between spontaneous electroencephalographic features and the perturbational complexity index in the minimally conscious state. European Journal of Neuroscience. doi:10.1111/ejn.16299
- Manasova, D., Belloli, T., Rosenfelder, M., Willacker, L., Fló Rama, M., Valota, A., Sinitsyn, D. O., Bourdillon, P., Comanducci, A., Heine, L., Sanz, L. R. D., Aabid, A., El-Daher, E., Liégeois-Chauvel, C., Biagioli, G., Woodruff, M., van Erp, W. S., Gomez, F., Gisquet, H., et al. (2026). Multimodal multicentre investigation of diagnostic and prognostic markers in disorders of consciousness. Brain. doi:10.1093/brain/awaf412
- Défossez, A., Caucheteux, C., Rapin, J., Kabeli, O., & King, J.-R. (2023). Decoding speech perception from non-invasive brain recordings. Nature Machine Intelligence, 5, 1097–1107. doi:10.1038/s42256-023-00714-5
- d'Ascoli, S., Ferrante, O., et al. (2025). Towards decoding individual words from non-invasive brain recordings. Nature Communications, 16, 10521. doi:10.1038/s41467-025-65499-0
- Willett, F. R., Kunz, E. M., Fan, C., et al. (2023). A high-performance speech neuroprosthesis. Nature, 620, 1031–1036. doi:10.1038/s41586-023-06377-x
- Littlejohn, K. T., Cho, C. J., Liu, J. R., et al. (2025). A streaming brain-to-voice neuroprosthesis to restore naturalistic communication. Nature Neuroscience, 28, 902–912. doi:10.1038/s41593-025-01905-6
- Wairagkar, M., Card, N. S., Singer-Clark, T., et al. (2025). An instantaneous voice-synthesis neuroprosthesis. Nature, 644, 145–152. doi:10.1038/s41586-025-09127-3
- Zada, Z., Goldstein, A., et al. (2024). Contextual embeddings from deep language models predict fMRI responses to naturalistic language. Neuron, 112(22), 3725–3740.e9. doi:10.1016/j.neuron.2024.09.018
- Goldstein, A., Bar, A., et al. (2025). Connecting concepts in the brain by mapping cortical representations of semantic relations. Nature Neuroscience, 28, 1841–1848. doi:10.1038/s41593-025-01903-8
- Fahrenfort, J. J., Johnson, D., Kloosterman, N. A., Stein, T., van Gaal, S., Pitts, M., & de Graaf, T. A. (2025). Criterion placement threatens the construct validity of neural measures of consciousness. eLife, 13, RP102335. doi:10.7554/eLife.102335
- Gouwens, N. W., et al. (2021). Phenotypic variation of transcriptomic cell types in mouse motor cortex. Nature, 598, 144–150. doi:10.1038/s41586-020-2907-3
- Yao, Z., et al. (2023). A high-resolution transcriptomic and spatial atlas of cell types in the whole mouse brain. Nature, 624, 317–332. doi:10.1038/s41586-023-06812-z
- MICrONS Consortium, et al. (2025). Functional connectomics spanning multiple areas of mouse visual cortex. Nature, 640, 435–447. doi:10.1038/s41586-025-08790-w
- Hengen, K. B., Torrado Pacheco, A., McGregor, J. N., Van Hooser, S. D., & Turrigiano, G. G. (2016). Neuronal firing rate homeostasis is inhibited by sleep and promoted by wake. Cell, 165(1), 180–191. doi:10.1016/j.cell.2016.01.046
- Torrado Pacheco, A., et al. (2021). Sleep Promotes Downward Firing Rate Homeostasis. Neuron, 109(3), 530–544.e6. doi:10.1016/j.neuron.2021.04.004
- Xu, W., et al. (2024). Sleep restores an optimal computational regime in cortical networks. Nature Communications, 15, 3820. doi:10.1038/s41467-024-47838-5
- Neyhart, E., Zhou, N., Munn, B. R., et al. (2024). Cortical acetylcholine dynamics are predicted by cholinergic axon activity and behavioral state. Cell Reports, 43(10), 114808. doi:10.1016/j.celrep.2024.114808
- Looser, Z. J., et al. (2024). Oligodendrocyte-axon metabolic coupling is mediated by extracellular K+ and maintains axonal health. Nature Neuroscience, 27, 1598–1609. doi:10.1038/s41593-023-01558-3
- Cahill, M. K., et al. (2024). Network-level encoding of local neurotransmitters in cortical astrocytes. Nature, 629, 146–153. doi:10.1038/s41586-024-07311-5
- Lee, J.-C., Wang, C.-Y., Lin, C.-L., & Lu, H.-C. (2022). Synaptic memory survives molecular turnover. PNAS, 119(42), e2211572119. doi:10.1073/pnas.2211572119
- Gogulski, J., Cline, C. C., Ross, J. M., Parmigiani, S., & Keller, C. J. (2024). Reliability of the TMS-evoked potential in dorsolateral prefrontal cortex. Cerebral Cortex, 34(4), bhae130. doi:10.1093/cercor/bhae130
- Biabani, M., Perera, T., Zalesky, A., et al. (2024). Characterising the contribution of auditory and somatosensory inputs to TMS-evoked potentials following stimulation of prefrontal, premotor, and parietal cortex. Imaging Neuroscience, 2. doi:10.1162/imag_a_00349
- Breyton, M., Lado, M. J., Laureys, S., et al. (2025). Spatiotemporal brain complexity quantifies consciousness outside of perturbation paradigms. eLife, 13, RP98920. doi:10.7554/eLife.98920
- Amiri, A. M., Fisher, D., Raimondo, F., et al. (2023). Multimodal prediction of residual consciousness in the intensive care unit: the CONNECT-ME study. Brain, 146, 50–64. doi:10.1093/brain/awac335
- Rohaut, B., Calligaris, M., Hermann, B., Perez, P., et al. (2024). Multimodal assessment improves neuroprognosis performance in clinically unresponsive critical-care patients with brain injury. Nature Medicine, 30, 2349–2355. doi:10.1038/s41591-024-03019-1
- Haufe, S., Nikulin, V. V., Müller, K.-R., & Nolte, G. (2013). A critical assessment of connectivity measures for EEG data: A simulation study. NeuroImage, 64, 120–133. doi:10.1016/j.neuroimage.2012.09.036
- Palva, J. M., Wang, S. H., Palva, S., Zhigalov, A., Monto, S., Brookes, M. J., Schoffelen, J.-M., & Jerbi, K. (2018). Ghost interactions in MEG/EEG source space: A note of caution on inter-areal coupling measures. NeuroImage, 173, 632–643. doi:10.1016/j.neuroimage.2018.02.032
- Ye, S., Kitajo, K., & Kitano, K. (2020). Information-theoretic approach to detect directional information flow in EEG signals induced by TMS. Neuroscience Research, 154, 87–96. doi:10.1016/j.neures.2019.09.003
- Miljevic, A., Murphy, O. W., Fitzgerald, P. B., & Bailey, N. W. (2025). Estimating sensor-space EEG connectivity PART 1: Identifying best performing methods for functional connectivity in simulated data. Clinical Neurophysiology, 174, 73–83. doi:10.1016/j.clinph.2025.03.043
- Seymour, R. A., Alexander, N., Mellor, S., O'Neill, G. C., Tierney, T. M., Barnes, G. R., & Maguire, E. A. (2021). Using OPMs to measure neural activity in standing, mobile participants. NeuroImage, 244, 118604. doi:10.1016/j.neuroimage.2021.118604
- Holmes, N., Rea, M., Hill, R. M., Leggett, J., Edwards, L. J., Hobson, P. J., Boto, E., Tierney, T. M., Rier, L., Reina Rivero, G., Shah, V., Osborne, J., Fromhold, T. M., Glover, P., Brookes, M. J., & Bowtell, R. (2023). Enabling ambulatory movement in wearable magnetoencephalography with matrix coil active magnetic shielding. NeuroImage, 274, 120157. doi:10.1016/j.neuroimage.2023.120157
- Holmes, N., Rea, M., Hill, R. M., Boto, E., Leggett, J., Edwards, L. J., Rhodes, N., Shah, V., Osborne, J., Fromhold, T. M., Glover, P., Montague, P. R., Brookes, M. J., & Bowtell, R. (2023). Naturalistic hyperscanning with wearable magnetoencephalography. Sensors, 23(12), 5454. doi:10.3390/s23125454
- Holmes, N., Leggett, J., Hill, R. M., Rier, L., Boto, E., Schofield, H., Hayward, T., Dawson, E., Woolger, D., Shah, V., Taulu, S., Brookes, M. J., & Bowtell, R. (2025). Wearable magnetoencephalography in a lightly shielded environment. IEEE Transactions on Biomedical Engineering, 72(2), 609–618. doi:10.1109/TBME.2024.3465654
- Iivanainen, J., Borna, A., Zetter, R., Carter, T. R., Stephen, J. M., McKay, J., Parkkonen, L., Taulu, S., & Schwindt, P. D. D. (2022). Calibration and Localization of Optically Pumped Magnetometers Using Electromagnetic Coils. Sensors, 22(8), 3059. doi:10.3390/s22083059
- Rhodes, N., Rier, L., Boto, E., Hill, R. M., & Brookes, M. J. (2025). Source reconstruction without an MRI using optically pumped magnetometer-based magnetoencephalography. Imaging Neuroscience, 3, IMAG.a.8. doi:10.1162/IMAG.a.8
- Wu, T., Xiao, W., Peng, X., Wu, T., & Guo, H. (2025). Crosstalk reduction in optically pumped magnetometers arrays for biomagnetic measurement. Review of Scientific Instruments, 96(8), 085004. doi:10.1063/5.0273491
- Beiran, M., & Litwin-Kumar, A. (2025). Prediction of neural activity in connectome-constrained recurrent networks. Nature Neuroscience, 28, 2561–2574. doi:10.1038/s41593-025-02080-4
- Smith, S. M., Miller, K. L., Salimi-Khorshidi, G., Webster, M., Beckmann, C. F., Nichols, T. E., Ramsey, J. D., & Woolrich, M. W. (2011). Network modelling methods for FMRI. NeuroImage, 54(2), 875–891. doi:10.1016/j.neuroimage.2010.08.063
- Barnett, L., & Seth, A. K. (2017). Detectability of Granger causality for subsampled continuous-time neurophysiological processes. Journal of Neuroscience Methods, 275, 93–121. doi:10.1016/j.jneumeth.2016.10.016
- Vink, J. J. T., Klooster, D. C. W., Ozdemir, R. A., Westover, M. B., Pascual-Leone, A., & Shafi, M. M. (2020). EEG Functional Connectivity is a Weak Predictor of Causal Brain Interactions. Brain Topography, 33(2), 221–237. doi:10.1007/s10548-020-00757-6
- Novelli, L., Barnett, L., Seth, A. K., & Razi, A. (2025). Minimum-Phase Property of the Hemodynamic Response Function, and Implications for Granger Causality in fMRI. Human Brain Mapping, 46(10), e70285. doi:10.1002/hbm.70285
- Villaverde, A. F., Tsiantis, N., & Banga, J. R. (2019). Full observability and estimation of unknown inputs, states and parameters of nonlinear biological models. Journal of The Royal Society Interface, 16(156), 20190043. doi:10.1098/rsif.2019.0043
- Yan, J., Zhang, S.-W., Zhang, C., Huang, W., Shi, J., & Chen, L. (2026). Dynamical Causality under Latent Confounders for Biological Network Reconstruction. IEEE Transactions on Pattern Analysis and Machine Intelligence. doi:10.1109/TPAMI.2026.3658839
- Galarreta, M., & Hestrin, S. (1999). A network of fast-spiking cells in the neocortex connected by electrical synapses. Nature, 402, 72–75. doi:10.1038/47029
- Anastassiou, C. A., Perin, R., Markram, H., & Koch, C. (2011). Ephaptic coupling of cortical neurons. Nature Neuroscience, 14, 217–223. doi:10.1038/nn.2727
- Graydon, C. W., Cho, S., Diamond, J. S., Kachar, B., von Gersdorff, H., & Grimes, W. N. (2014). Specialized postsynaptic morphology enhances neurotransmitter dilution and high-frequency signaling at an auditory synapse. Journal of Neuroscience, 34(23), 7871–7885. doi:10.1523/JNEUROSCI.4493-13.2014
- Kilb, W., Dierkes, P. W., Syková, E., Vargová, L., & Luhmann, H. J. (2006). Hypoosmolar conditions reduce extracellular volume fraction and enhance epileptiform activity in the CA3 region of the immature rat hippocampus. Journal of Neuroscience Research, 84(2), 361–371. doi:10.1002/jnr.20871
- Lauderdale, K., Murphy, T., Tung, T., Davila, D., Binder, D. K., & Fiacco, T. A. (2015). Osmotic edema rapidly increases neuronal excitability through activation of NMDA receptor-dependent slow inward currents in juvenile and adult hippocampus. ASN Neuro, 7(5). doi:10.1177/1759091415605115
- Xie, L., Kang, H., Xu, Q., Chen, M. J., Liao, Y., Thiyagarajan, M., O'Donnell, J., Christensen, D. J., Nicholson, C., Iliff, J. J., Takano, T., Deane, R., & Nedergaard, M. (2013). Sleep drives metabolite clearance from the adult brain. Science, 342(6156), 373–377. doi:10.1126/science.1241224
- Burman, R. J., Brodersen, P. J. N., Raimondo, J. V., Sen, A., & Akerman, C. J. (2023). Active cortical networks promote shunting fast synaptic inhibition in vivo. Neuron, 111(22), 3545–3560.e6. doi:10.1016/j.neuron.2023.08.005
- Yang, Y.-C., Wang, G.-H., Chou, P., Hsueh, S.-W., Lai, Y.-C., & Kuo, C.-C. (2024). Dynamic electrical synapses rewire brain networks for persistent oscillations and epileptogenesis. Proceedings of the National Academy of Sciences, 121(8), e2313042121. doi:10.1073/pnas.2313042121
- Selfe, J. S., Steyn, T. J. S., Shorer, E. F., Burman, R. J., Düsterwald, K. M., Kraitzick, A. Z., Ju, C., Goldman, P. S., Ibrahim, B. A., Syed, P., Li, L., Akerman, C. J., Kullmann, D. M., Raimondo, J. V., & Baker, C. A. (2024). All-optical reporting of inhibitory receptor driving force in the nervous system. Nature Communications, 15, 9752. doi:10.1038/s41467-024-53074-y
- Voldsbekk, I., Maximov, I. I., Zak, N., Roelfs, D., Geier, O., Due-Tønnessen, P., Elvsåshagen, T., Strømstad, M., Bjørnerud, A., & Groote, I. (2020). Evidence for wakefulness-related changes to extracellular space in human brain white matter from diffusion-weighted MRI. NeuroImage, 212, 116682. doi:10.1016/j.neuroimage.2020.116682
- Feld, G. B., Niethard, N., Liu, J., Gebhardt, S., Kleist, L., Brugger, K., & Born, J. (2026). Electrical synapses contribute to sleep-dependent declarative memory retention. European Journal of Neuroscience. doi:10.1111/ejn.70401
- Singh, A., Thomas, T., Li, J., et al. (2025). Transfer learning via distributed brain recordings enables reliable speech decoding. Nature Communications, 16, 8749. doi:10.1038/s41467-025-63825-0
- Karpowicz, B. M., Ali, Y. H., Wimalasena, L. N., et al. (2025). Stabilizing brain-computer interfaces through alignment of latent dynamics. Nature Communications, 16, 4662. doi:10.1038/s41467-025-59652-y
- Wilson, G. H., Stein, E. A., Kamdar, F., et al. (2025). Long-term unsupervised recalibration of cursor-based intracortical brain-computer interfaces using a hidden Markov model. Nature Biomedical Engineering. doi:10.1038/s41551-025-01536-z
- Fultz, N. E., Bonmassar, G., Setsompop, K., Stickgold, R. A., Rosen, B. R., Polimeni, J. R., & Lewis, L. D. (2019). Coupled electrophysiological, hemodynamic, and cerebrospinal fluid oscillations in human sleep. Science, 366(6465), 628-631. doi:10.1126/science.aax5440
- Eide, P. K., Lashkarivand, A., Pripp, A. H., et al. (2023). Plasma neurodegeneration biomarker concentrations associate with glymphatic and meningeal lymphatic measures in neurological disorders. Nature Communications, 14, 2084. doi:10.1038/s41467-023-37685-5
- Kim, D., Huang, Y., & Liu, J. (2025). Non-invasive MRI measurements of age-dependent in vivo human glymphatic exchange using magnetization transfer spin labeling. NeuroImage, 310, 121142. doi:10.1016/j.neuroimage.2025.121142
- Hirschler, L., Runderkamp, B. A. R., Decker, A., et al. (2025). Region-specific drivers of CSF mobility measured with MRI in humans. Nature Neuroscience, 28, 2392-2401. doi:10.1038/s41593-025-02073-3
- Lim, C., Chen, C., Zhang, C., et al. (2025). Human cerebrospinal fluid net flow enhanced by respiration during the awake state. Nature Communications, 16, 595. doi:10.1038/s41467-025-66548-4
- Yoo, R.-E., Kim, J.-H., Moon, H. Y., et al. (2025). Long-term physical exercise facilitates putative glymphatic and meningeal lymphatic vessel flow in humans. Nature Communications, 16, 3360. doi:10.1038/s41467-025-58726-1
- Dagum, P., Elbert, D. L., Giovangrandi, L., et al. (2026). The glymphatic system clears amyloid beta and tau from brain to plasma in humans. Nature Communications, 17, 715. doi:10.1038/s41467-026-68374-8
- Colaes, R., Radwan, A., Billiet, T., Cappelle, S., Peeters, R., Christiaens, D., et al. (2026). Evaluating the T1w/FLAIR ratio as a proxy for myelin: Associations with myelin water imaging, diffusion metrics, and cognition. Neuroradiology. doi:10.1007/s00234-025-03875-9
- Carro-Domínguez, M., Huwiler, S., Oberlin, S., et al. (2025). Pupil size reveals arousal level fluctuations in human sleep. Nature Communications, 16, 2070. doi:10.1038/s41467-025-57289-5
- Nakuci, J., & Bansal, K. (2025). The neuroreceptors and transporters underlying spontaneous brain activity. Communications Biology, 8, 527. doi:10.1038/s42003-025-08492-z
- Schlosser, G., Murgaš, M., Godbersen, G. M., et al. (2025). Human in vivo assessment of ketamine binding of the serotonin transporter-follow up at a higher dose. Frontiers in Neuroscience, 19, 1651016. doi:10.3389/fnins.2025.1651016
- Miederer, I., Buchholz, H.-G., Rademacher, L., et al. (2025). Dopaminergic Mechanisms of Cognitive Flexibility: An [18F]Fallypride PET Study. Journal of Nuclear Medicine, 66(6), 908-915. doi:10.2967/jnumed.124.268317
- Thomas, C., Ye, F. Q., Irfanoglu, M. O., Modi, P., Saleem, K. S., Leopold, D. A., & Pierpaoli, C. (2014). Anatomical accuracy of brain connections derived from diffusion MRI tractography is inherently limited. Proceedings of the National Academy of Sciences, 111(46), 16574-16579. doi:10.1073/pnas.1405672111
- Gajwani, M., Oldham, S., Pang, J. C., Arnatkevičiūtė, A., Tiego, J., Bellgrove, M. A., & Fornito, A. (2023). Can hubs of the human connectome be identified consistently with diffusion MRI? Network Neuroscience, 7(4), 1440-1471. doi:10.1162/netn_a_00324
- McMaster, E. M., Newlin, N. R., Rudravaram, G., Saunders, A. M., Krishnan, A. R., Remedios, L. W., Kim, M. E., Xu, H., Yoon, J., Archer, D. B., Schilling, K. G., Rheault, F., Cutting, L. E., & Landman, B. A. (2025). Harmonized connectome resampling for variance in voxel sizes. Magnetic Resonance Imaging, 121, 110424. doi:10.1016/j.mri.2025.110424
- Bramati, I. B., Szczupak, D., Monteiro, M. C., Meireles, F., Guimarães, D. M., Dean, R. J., Paul, L. K., & Tovar-Moll, F. (2026). Diffusion MRI sampling schemes bias diffusion metrics and tractography. Frontiers in Neuroimaging, 5, 1670604. doi:10.3389/fnimg.2026.1670604
- Zhu, S., Huszar, I. N., Cottaar, M., Daubney, G., Eichert, N., Hanayik, T., Khrapitchev, A. A., Mars, R. B., Mollink, J., Sallet, J., Scott, C., Smart, A., Jbabdi, S., Miller, K. L., & Howard, A. F. D. (2025). Imaging the structural connectome with hybrid MRI-microscopy tractography. Medical Image Analysis, 102, 103498. doi:10.1016/j.media.2025.103498