From Empirical Problem
to Control Hypothesis
What problem is CCH trying to solve?
Which measurable brain dynamics distinguish a system that retains the capacity for conscious experience, and are those dynamics merely observed — or actively maintained, restored, and protected against perturbation by identifiable control mechanisms?
This page does not begin by assuming CCH, a Consciousness Field, or a privileged brain region. It begins with an empirical difficulty: behavioral responsiveness, arousal, reportability, the presence of experience, and the contents of experience can dissociate. Dreaming and ketamine show that unresponsiveness is not identical to absence of experience; disorders of consciousness show that behavior can underestimate residual capacity.
The target is therefore deliberately narrower than the metaphysical “hard problem.” CCH asks about the local causal organization of conscious-capable brain regimes: their measurable state, transitions, maintenance, recovery, and vulnerability to disturbance.
Four distinctions that prevent a false target
| Term | Operational meaning here | Not identical to |
|---|---|---|
| Arousal | Wakefulness and activation supported by cortical and subcortical systems. | Rich experience or reportability. |
| Responsiveness | Observable behavior following a command or stimulus. | Presence of experience. |
| Capacity | Current ability of the brain to sustain differentiated, integrated causal responses compatible with experience. | A report about a particular content. |
| Content | What is experienced at a given moment. | The overall level or capacity for consciousness. |
A classifier is not yet an explanation. CCH must identify candidate state variables, causal organization, lawful interventions, competing mechanisms, and failure conditions. If it only predicts labels, its central control claim fails even if its classifier is accurate.
Check the current role, status, allowed support, and forbidden inference for AC, CF, R, RC, CCH, DCC, MDL×DCC, AI8, X(t), ΩCCH, S*, and retired CCC.
Evidence before hypothesis
Perturbational complexity
supported marker
TMS–EEG perturbational complexity separates multiple benchmark conditions with and without reported experience and has been used to stratify behaviorally unresponsive patients.
It does not identify a controller or explain phenomenal presence.
Spontaneous signal diversity
state-sensitive
Several EEG diversity measures decrease under propofol and vary across sleep stages.
Within-stage dream-report results and preprocessing dependence block a simple “more complexity = more consciousness” law.
Causal subcortical modulation
causal evidence
Central lateral thalamic stimulation can restore wake-like cortical dynamics and behavior in anaesthetized macaques.
Any claustrum-centered account must beat this and other distributed alternatives.
Claustrum findings
mixed
Human stimulation results conflict; newer human and animal work supports roles in sleep, engagement, prefrontal dynamics, connectivity, and variability.
This supports “candidate modulator,” not “seat of consciousness.”
What these findings jointly support
The strongest common ground is not that one scalar, one region, or complexity itself is consciousness. It is that conscious-capable and unconscious brain regimes often differ in the structure of causal interactions, differentiation, temporal dynamics, and recoverability; these differences are measurable and perturbable.
What they do not establish
They do not establish necessity and sufficiency of any current metric, prove the claustrum is the controller, explain why experience exists, or show that an artificial controller would be conscious. Recent dissociations between spontaneous EEG features and PCI further warn that different measurements can capture different aspects of brain state.
The detailed source-by-source grounding, competing mechanisms, and claustrum evidence are kept in CCH Scientific Foundations. That companion begins with the literature rather than the CCH model.
From state markers to causal maintenance
Current markers can classify or stratify brain states. They do not, by themselves, specify how a brain enters, maintains, and recovers a conscious-capable regime under anaesthesia, sleep transitions, injury, seizures, or changing task demands.
CCH targets this narrower gap: is the relevant regime actively regulated in a multidimensional state space, and can that claim make causal predictions that simpler passive, arousal-only, fixed-controller, or alternative-network models do not?
Neuroscience already contains homeostatic, thalamocortical, recurrent, arousal, predictive, and network-control ideas. CCH earns value only if its explicit state-space and feedback formulation produces better discriminating predictions or cleaner experiments. Renaming known regulation as a controller is not a contribution.
Why the gap is explanatory, not merely predictive
A useful explanation should identify parts, activities, organization, and interventions. For CCH this means: what is sensed, what state is estimated, what counts as leaving a viable region, which biological pathways can alter the trajectory, and what neural and behavioral consequences should follow.
How CCH follows from the problem
CCH is introduced only after the following evidence-constrained premises:
Sleep, anaesthesia, seizures, injury, stimulation, and task changes alter trajectories rather than merely static labels.
Responsiveness, arousal, spontaneous diversity, perturbational complexity, and reported experience can dissociate.
Perturbational and stimulation studies show that how the network responds and propagates activity contains information not reducible to raw activity magnitude.
Thalamic, brainstem, cortical, neuromodulatory, and claustral systems are plausible contributors. The evidence favors competition among architectures, not a preselected winner.
If active regulation is required, preregistered controller models should outperform matched passive, PK/PD, arousal-only, and fixed alternatives. External closed-loop success would establish controllability; endogenous controller status requires a separate biological signature.
Four separable CCH hypotheses
| ID | Hypothesis | Can fail independently? |
|---|---|---|
| H1 | A conscious-capable brain regime occupies a reproducible multidimensional region ΩCCH, not a universal maximum of complexity or synchrony. | Yes |
| H2 | Robust maintenance and recovery of that region requires adaptive feedback beyond passive dynamics and simpler fixed rules. | Yes |
| H3 | The claustrum contributes distinctively as one node within a distributed controller. | Yes |
| H4 | DCC / MDL×DCC regulation can improve artificial-system stability, search, compression, and research performance under matched budgets. | Yes; independent engineering evidence lane only |
A minimal state-space and control blueprint
Controller, plant, sensor, and actuator are functional and intervention-relative roles defined by a specific intervention and prediction problem, not necessarily anatomical compartments. In recurrent brains, the roles may overlap anatomically and may be distributed. A candidate controller component is not the same as a complete controller architecture.
| Element | Scientific role | Current status |
|---|---|---|
| Plant | Distributed biological network evolving under internal and external perturbation. | Real brain; reduced models required. |
| Measurements y(t) | EEG/MEG, intracranial signals, imaging, perturbational response, behavior, and physiology. | Dataset- and modality-dependent. |
| State X(t) | Candidate vector of integration, differentiation, temporal structure, context, and uncertainty. | PRIMARY CANDIDATE MULTIDIMENSIONAL STATE REPRESENTATION NOT YET FROZEN FOR A CLAIM-BEARING TEST |
| Region ΩCCH | Candidate preregistered region associated with retained capacity, learned without held-out leakage. | CANDIDATE PREREGISTERED TARGET REGION NOT YET FROZEN |
| Controller π | Potentially distributed mapping from state/error and context to lawful modulation. | Open hypothesis; model class must be frozen. |
| Actuators u(t) | Gain, excitation/inhibition, routing, neuromodulation, arousal, timing, and network coupling. | Must be causally mapped with spread controls. |
| Outcomes z(t) | Preregistered neural trajectory endpoint plus separate secondary behavioral/perturbational outcomes. | Hierarchy defined below; phenomenality not inferred from state alone. |
Mandatory future measurement contract for X(t)
No final feature set, threshold, equation, channel list, or preprocessing value is asserted here. Before a claim-bearing test, the protocol must freeze:
Exploratory feature discovery and confirmatory evaluation must use separate datasets or properly nested folds. X(t) remains a candidate family until that contract is frozen.
Anti-circularity contract for ΩCCH
ΩCCH is a preregistered region learned from one evidence hierarchy and evaluated against independent outcomes in held-out conditions. It is not defined as whatever region happens to contain samples labelled conscious after inspection of the result.
- Immediate report where available.
- Covert command following measured independently of the features used to train X(t).
- Validated perturbational benchmark as an independent capacity anchor.
- Behavioral responsiveness as a separate outcome, not universal ground truth.
- Drug concentration and peripheral arousal as covariates, not consciousness labels.
Train and freeze the representation and ΩCCH boundary in one condition; test them against independent outcomes in another held-out condition. Discordant cases must be reported, not relabelled post hoc.
Primary state, optional compression
The primary scientific object is X(t). The optional candidate compression S*=k·Cn·(ERn·SI) is retained only as a preregistered adversarial target. It may fail while a multidimensional model survives. Here Cn is normalized coordination, ERn effective-rank diversity, and SI a temporal-structure calibration proxy.
The component estimators (Cn, ERn, SI), the scaling constant k, and the S* failure threshold are subject to the same measurement contract as X(t) and are not asserted here. The form above is illustrative of the candidate family, not a frozen definition.
High synchrony can be pathological; high randomness can be unstructured; different states can share the same product. CCH predicts a candidate viable region with context-dependent tradeoffs, not “more S* means more consciousness.”
Predeclared outcome hierarchy
No region owns consciousness
| Candidate architecture | Why it belongs in the competition | CCH burden |
|---|---|---|
| Thalamocortical / mesocircuit | Causal evidence links central lateral thalamus and deep cortical layers to consciousness-level transitions; recent cross-condition fMRI also shows nucleus-specific thalamic organization across anaesthesia, sleep, and disorders of consciousness. | Show whether state/error-control adds held-out prediction beyond thalamic and arousal models. |
| Brainstem & neuromodulatory | Core arousal and state transitions depend on subcortical systems. | Separate general arousal from differentiated capacity and residual PK/PD effects. |
| Frontoparietal / report-related | Networks contribute to access, task engagement, report, and control. | Distinguish experience-related dynamics from report and task demands. |
| Passive recurrent / whole-brain dynamics | Relevant state organization may self-stabilize without a distinct adaptive controller. | H2 fails if frozen dynamical models match prediction and recovery at lower complexity. |
| Predictive / active-inference control | Existing estimate–error–policy accounts may already implement the relevant control logic. | CCH must add a discriminating prediction, not only new vocabulary. |
| Claustrum subnetworks | Connectivity and circuit-specific modulation findings justify inclusion as a candidate node. | Demonstrate a distinctive, replicable contribution beyond spread, neighboring structures, and competing nodes. |
The claustrum evidence, stated without inflation
Human stimulation evidence is conflicting and anatomically limited. In 2014, stimulation at one contact near the left claustrum–anterior-dorsal-insula boundary in one epilepsy patient with a prior ipsilateral temporal lobectomy produced reversible unresponsiveness and amnesia under a stimulation protocol that differed from the later series. In 2019, direct—including bilateral—stimulation in five patients did not reproduce loss of consciousness. Human lesion and resection evidence further constrains necessity and uniqueness claims. Surgical cases are compatible with substantial functional compensation after unilateral claustrum involvement, while a penetrating-lesion series associated claustrum damage with the duration but not the frequency of loss of consciousness and found long-term outcome more strongly related to total lesion burden. Review-level synthesis does not support a crucial and unique role in maintaining wakefulness, while leaving open a modulatory or wider-network-component role. Complementing these limiting findings, Snider et al. (2020) found no single cortical lesion focus associated with loss of consciousness across 16 prolonged-LOC, 91 transient-LOC, and 64 no-LOC cases; lesion-network mapping instead identified a distributed circuit defined by anticorrelation with the dorsal brainstem, with peaks in the bilateral claustrum. This is positive network-level evidence involving the claustrum, but it supports a distributed circuit rather than ownership by one region.
In two intracranial subjects, 49 claustrum units tracked NREM slow waves. Mouse studies show projection-specific modulation of engagement, sleep, prefrontal activity, connectivity, and response variability, with effects dependent on circuit and task. These findings justify a candidate-modulator status only.
A claustrum pathway may carry control-relevant information or causal influence while remaining only a candidate controller component. It is not thereby the complete estimator–error–policy–actuator architecture, a seat of consciousness, a necessary/sufficient switch, or a receiver.
What would distinguish CCH from a loose idea?
Preregistered model classes
The comparison is not verbal. Every class must be frozen with the same target data, declared information access, and matched intervention budget.
| Class | Available inputs | Latent state | Adjustable parameters | Online update | Policy form | Intervention budget | Complexity / MDL penalty | Held-out target | Failure threshold |
|---|---|---|---|---|---|---|---|---|---|
| PASSIVE_DYNAMICAL | Past observed neural/physiological state and disturbance input; no privileged error channel. | Fixed stochastic dynamical state. | Dynamics/noise parameters frozen after training. | Forbidden. | No separate state estimate or error-dependent policy. | No intervention; observational prediction only. | Parameter count plus encoded model/residual cost. | State and recovery trajectory. | Fails if calibrated held-out prediction is inferior after penalty. |
| AROUSAL_ONLY | One preregistered arousal axis, drug/effect-site estimates, pupil/EMG/cardiorespiratory covariates. | Single arousal latent plus PK/PD state. | Frozen mapping and PK/PD parameters. | Forbidden. | Scalar state prediction; optional fixed arousal intervention. | Matched to compared controller. | Same likelihood/calibration and MDL accounting. | Neural trajectory plus separate behavior. | Fails if multidimensional model adds robust held-out value. |
| FIXED_CONTROLLER | Frozen X(t), target distance, context, allowed covariates. | Frozen state estimate. | Policy/gain frozen before test. | Forbidden. | State-dependent, non-adaptive policy. | Exactly matched energy, duration, and dose. | Policy + estimator + residual description length. | Restoration/maintenance of primary neural endpoint. | Fails if adaptive benefit survives matched budget and penalty. |
| ADAPTIVE_CONTROLLER | Preregistered X(t), signed error history, context, allowed covariates. | State and online policy/gain state. | Only preregistered gains/policy parameters may update. | Allowed under frozen update law. | Error-dependent online policy/gain adaptation. | Exactly matched energy, duration, and dose. | Full estimator, policy, update-law, and residual cost. | Primary neural trajectory; secondary outcomes separate. | Fails without penalized held-out superiority and controller signature. |
| DISTRIBUTED_CONTROL | Preregistered signals from multiple nodes/pathways and matched covariates. | Redundant distributed state/control representation. | Frozen or preregistered adaptive distributed parameters. | As preregistered; reported separately. | No privileged node; redundant control signals and actuator effects. | Matched total intervention energy across nodes. | Network/policy description length plus calibration penalty. | Trajectory, robustness to node loss, and rescue. | Fails if privileged-node model is simpler and equally predictive/causal. |
Every comparison reports predictive likelihood and calibration, held-out error, parameter count or description-length penalty, and intervention energy—not accuracy alone.
Endogenous biological controller signature
CCH H2 requires the conjunction of all six signals:
- Candidate activity carries signed target-distance or error information beyond PK/PD and arousal.
- The signal has temporal precedence over the predicted corrective change.
- Identified output pathways provide pathway-specific causal mediation.
- Selective disruption produces selective degradation of stability or recovery at matched disturbance.
- Activation through the predicted actuator provides rescue.
- The full model beats passive, PK/PD, arousal-only, and fixed-control baselines after complexity penalty.
Superiority of imposed closed-loop stimulation would establish that the measured regime is controllable. Evidence for an endogenous biological controller additionally requires preregistered error coding, pathway-specific causal mediation, selective impairment under controller disruption, and rescue through the predicted actuator.
External adaptive closed-loop superiority establishes controllability. It does not by itself establish an endogenous biological controller.
Experiment families
Freeze preprocessing, estimators, nulls, splits, uncertainty, movement/EMG/pupil/cardiorespiratory covariates, anaesthetic concentration, and stimulation-spread models.
Train and freeze X(t) and ΩCCH in one condition; evaluate held-out subjects and a distinct condition against independent outcomes. Report discordance.
Predict path and timing after explicit PK/PD/effect-site baselines rather than treating induction/recovery asymmetry as controller evidence.
Preregister candidate-node, thalamic, arousal, and distributed-node perturbations, control sites, dose/intensity, spread, pathway mediation, disruption, and rescue.
At matched intervention energy and information, test adaptive, fixed, yoked/open-loop, no-control, PK/PD, arousal-only, and distributed alternatives on the primary neural trajectory.
Preliminary staged experiment roadmap
This is a preliminary research roadmap, not a funded, ethically approved, or execution-ready animal study. Specialist neuroscience review, ethics approval, suitable institutional collaboration, preregistration, and independent replication are required.
AI engineering is a separate evidence and experiment family
DCC, MDL×DCC, AIM³/MAL, AI8, and ArenaLoop can test whether regulation improves search, compression, persistent research loops, recovery, and innovation under matched model, tool, compute, token, intervention, and human-attention budgets. A positive result supports the tested engineering claim and cross-domain transfer only—not biological CCH, AC/CFH, artificial phenomenality, or consciousness. See Projects Are Evidence and the MDL×DCC Domain Map.
CCH must lose cleanly when a simpler explanation wins
| Alternative | What it says | Result that favors it |
|---|---|---|
| A0 Passive dynamics | The relevant regime self-organizes; no adaptive controller is needed. | Passive model matches state transitions and recovery at lower complexity. |
| A1 Arousal axis | A single subcortical activation variable explains the useful variance. | Multidimensional Ω adds no held-out value. |
| A2 Marker-only | Complexity measures classify capacity but need not identify a controller. | Markers predict while controller interventions add no selective effect. |
| A3 Established distributed control | Thalamocortical, brainstem, and cortical networks already explain regulation. | CCH adds terminology but no new prediction. |
| A4 Claustrum-specific | A claustrum pathway has a distinctive control role. | Replicable, anatomically specific perturbations beat competing nodes. |
Preregisterable decision rule and falsification table
CCH H2 is supported only if the adaptive/endogenous model improves the primary held-out neural state/trajectory endpoint, remains calibrated, survives the full complexity/MDL penalty and equal intervention-energy budget, and meets the endogenous-controller conjunction. Improvement only in a secondary behavioral or arousal outcome does not satisfy H2.
| Decision object | Preregistered success condition | Falsification / rival-favoring result |
|---|---|---|
| X(t) / ΩCCH | Frozen representation and target region transport across held-out subjects and a distinct condition while retaining calibrated value for the primary neural endpoint. | Failure to generalize, preprocessing fragility, outcome leakage, or loss to a simpler compact baseline after penalty. |
| Adaptive / endogenous control (H2) | Penalized held-out superiority under equal intervention energy plus the full six-part endogenous-controller signature. | No advantage over passive, PK/PD, arousal-only, fixed, or distributed alternatives; external closed-loop benefit without endogenous signature supports controllability only. |
| Candidate controller component (H3) | Anatomically and pathway-specific error coding, mediation, selective disruption, and rescue that survive spread and competing-node controls. | No distinctive effect, broader-node explanation, or equal performance from a distributed architecture. |
| Optional compression S* | Preregistered robust compression adds held-out value beyond X(t) baselines without becoming the label definition. | Instability, non-transport, or no incremental value. S* may fail independently while X(t) survives. |
| AC / CFH interpretation | A preregistered prediction differs from and outperforms an otherwise identical no-AC/no-CF model. | Local CCH success alone, or equal predictions from no-AC/no-CF models, leaves ontology underdetermined. |
Hard failure conditions
Metric failure: X(t)/ΩCCH does not generalize, is preprocessing-fragile, or loses to simpler baselines.
Mechanism failure: closed-loop adaptive control offers no advantage over passive or fixed systems.
Claustrum failure: no distinctive causal contribution survives anatomy, spread, and competing-node controls.
Explanatory failure: the model predicts labels but cannot specify causal parts, activities, or intervention paths.
Hard moving-target falsifier: the research programme fails if definitions, target regions, or thresholds move after outcomes are known.
The scalar S* can fail while X(t) survives. H3 can fail while H2 survives. The entire biological program can fail while DCC remains useful engineering. No lower-level success is allowed to prove a higher-level claim.
A control question, not a universal replacement theory
| Framework | Main question | CCH relation |
|---|---|---|
| GNW | How does information become globally available for access and report? | CCH asks how a regime capable of such broadcasting is maintained and recovered. |
| IIT | What intrinsic causal structure corresponds to experience? | CCH does not identify a metric with experience; it tests control of observable dynamics. |
| Recurrent processing | Which recurrent interactions support conscious perception? | Recurrent dynamics may be part of the plant or may defeat the need for a separate controller. |
| Predictive / active inference | How do hierarchical generative systems regulate prediction and action? | Potential source of controller mechanisms and strong alternative models. |
| Mesocircuit / thalamocortical | How do subcortical–cortical circuits support loss and recovery? | Direct causal competitor and possible implementation of H2. |
| Adversarial theory testing · Cogitate 2025 | Which preregistered predictions of GNWT and IIT survive direct multimodal comparison? | The reported lack of sustained posterior synchronization challenges an IIT prediction; the general lack of offset ignition and limited prefrontal representation of some conscious dimensions challenge GNWT predictions. This does not support CCH. It strengthens the case for quantitative, theory-neutral comparison and leaves maintenance and recovery as separate questions. |
The Cogitate adversarial collaboration shows the value of preregistered, theory-neutral comparison and reports serious challenges to predictions from both IIT and GNWT. It does not make CCH the default winner; CCH must enter the same kind of quantitative contest and lose cleanly when its predictions fail.
Ontology-neutral science: ontology is not scientific grounding
AC/RC and CFH may remain separate ontological interpretations. They are not premises or evidence for CCH. If a no-CF model and an AC/CF model make identical empirical predictions, science should treat the ontology as underdetermined.
Research programme, not victory claim
The current deliverable is a grounded experimental architecture. No large-scale held-out result establishes ΩCCH; no closed-loop biological controller has been demonstrated; and claustrum necessity or sufficiency is not established.
A successful local CCH result would not prove AC or CFH; it would support only the preregistered local biological model that survived its competitors.
Next decisive work
Stage 0: existing-data measurement and identifiability preflight under predeclared estimability, calibration, leakage, and nested-fold stability criteria. Stage 1: freeze X(t), ΩCCH, model classes, endpoints, and decision table in a timestamped preregistration and pass held-out anti-circularity checks. Stage 2: only after those gates pass, pursue a specialist-reviewed and preregistered causal perturb-and-release study through an appropriate institution.
Every future CCH claim must name: the observed problem, established evidence, exact gap, competing explanation, proposed mechanism, discriminating prediction, preregistered test, and failure condition.
Provenance
BD originated the consciousness–claustrum–DCC research direction and retains release responsibility. AI collaborators contributed formalization, code scaffolding, literature structuring, adversarial examples, and editorial review. This provenance is not evidence for the hypothesis.
Sources that motivate or constrain CCH
These papers establish observations and constraints. None proves the full CCH package.
Numbered references. DOI and PMID destinations were checked on 18 August 2026. The list order is part of this release.
- Casali et al. (2013) — Introduced the Perturbational Complexity Index (PCI) using TMS–EEG and showed that perturbation-evoked spatiotemporal complexity can distinguish several reported conscious and unconscious conditions.
- Casarotto et al. (2016) — Validated PCI in a benchmark population with immediate or delayed subjective reports, then applied the threshold to patients with disorders of consciousness.
- Schartner et al. (2015) — Found reduced spontaneous multidimensional EEG complexity under propofol anaesthesia using several diversity measures.
- Aamodt et al. (2021) — Found that EEG signal diversity varies with sleep stage, while within-stage dream-report results constrain a simple “more complexity = more experience” interpretation.
- Casarotto et al. (2024) — Reported dissociations between spontaneous EEG features and perturbational complexity in minimally conscious patients, arguing against treating one spontaneous marker as the whole phenomenon.
- Redinbaugh et al. (2020) — Provided causal evidence that central lateral thalamic stimulation can modulate consciousness-related cortical dynamics in anaesthetized macaques; a major competing control candidate.
- Koubeissi et al. (2014) — A single-patient stimulation near a claustral electrode produced reversible unresponsiveness and amnesia; important but anatomically and statistically limited.
- Bickel & Parvizi (2019) — Direct, including bilateral, claustrum stimulation in five epilepsy patients did not produce loss of consciousness.
- Duffau et al. (2007) — Low-grade-glioma surgery involving the claustrum provides evidence of substantial functional compensation after unilateral involvement; it constrains simple indispensability claims without proving universal dispensability.
- Chau et al. (2015) — In penetrating brain injury, claustrum damage was associated with duration but not frequency of loss of consciousness; long-term outcome was more strongly related to total lesion burden.
- Snider et al. (2020) · PMID 31904898 — Standard voxel lesion–symptom mapping found no single cortical focus associated with loss of consciousness across 16 prolonged-LOC, 91 transient-LOC, and 64 no-LOC cortical-lesion cases; lesion-network mapping instead identified a distributed circuit anticorrelated with the dorsal brainstem, with peaks in the bilateral claustrum. This supports network involvement, not regional ownership.
- Liaw & Augustine (2023) — Review-level synthesis does not support a crucial and unique role for the claustrum in maintaining wakefulness, while retaining a possible contribution within wider networks.
- Pavel et al. (2019) — Claustrum-area electrical stimulation in rats transiently deepened slow-wave isoflurane anaesthesia.
- Luo et al. (2023) — Chemogenetic claustrum activation attenuated propofol sensitivity, shortened anaesthesia, and shifted EEG activity toward wakefulness.
- Lamsam et al. (2024) — Human claustrum single units tracked NREM slow waves in a small intracranial sample, supporting a regulatory role in sleep dynamics rather than a simple consciousness switch.
- Zahacy et al. (2024) — Mouse experiments showed that claustrum manipulation changes prefrontal activity and large-scale functional connectivity.
- Atlan et al. (2024) — Linked projection-defined claustrum activity to engagement, sensory responsiveness, and NREM sleep in mice, with circuit-specific and strategy-dependent effects.
- Atilgan et al. (2025) — Showed claustrum-dependent modulation of prefrontal variability and population response organization in mice.
- Dasilva et al. (2021) — Showed that cortical complexity can vary within anaesthesia regimes and relates to slow-oscillation dynamics, cautioning against a single static state label.
Model and alternative constraints
- Sanz Perl et al. (2023); Lee et al. (2022); Jang et al. (2024) — Existing compact and multidimensional state-space baselines that CCH must beat or incorporate.
- Proekt & Kelz (2021); Huang et al. (2021) — Neural-inertia, PK/PD, and effect-site constraints on transition asymmetry.
- Bastos et al. (2021); Tasserie et al. (2022); Luppi et al. (2024); Lu et al. (2025) — Thalamocortical, distributed-dynamical, and cross-condition thalamic competitors.
- Allen et al. (2020); Hesse et al. (2020); Cogitate Consortium (2025) — Active-inference, report-related, and theory-comparison constraints.
CCH research stack — AC/RC ontology • CFH origin & ontology bridge • scientific foundations