Symbiotic human-AI architecture for somatic sensing, symbolic reasoning and metacognitive control

Symbiotic human-AI architecture for somatic sensing, symbolic reasoning and metacognitive control† Alexander Mathiesen-Ohman1and Jacek Ma lecki2* 1AMOTHO Research Institute, Vallsj¨on 20, 780 00 R¨orb¨acksn¨as, Sweden. 2Department of Mathematics, Wroc law University of Science and Technology, Wybrze˙ze Wyspia´nskiego 27, 50-370 Wroc law, Poland. *Corresponding author(s). E-mail(s): [email protected]; Abstract We present a tripartite cognitive architecture that unifies somatic grounding, symbolic inference and metacognitive control within an extended SORK-N loop. Cognition is cast as the interaction of a Somatic layer (biophysical and affective inputs), a Symbolic layer (linguistic, logical, and representational processes), and a Metacognitive layer (global coherence estimation and policy adjustment), coordinated by a methodological framework that time-locks and analyzes multimodal physiological, linguistic and self-report data. Mathematical structure is provided by the Mirrored Profile Graph (MPG), an evidence-linked, hierarchical state space, and Rogue Variable (RV) analysis, which together localize structural sources of prediction-observation gaps and support falsifiable tests. This framework enables reproducible tests of intuition, pre-event cognition, and collective coherence, while remaining compatible with empirical scrutiny. We further discuss implications for symbiotic human-AI systems and argue that intentional co-evolution of biological and artificial cognition offers a practical route toward robust, reflective intelligence. Keywords: Cognitive architecture; Human-AI symbiosis; Somatic sensing; Symbolic reasoning; Metacognitive control; SORK-N loop; Mirrored Profile Graph (MPG); Rogue Variables (RV); Minimal Unaware Flip Set (MUFS); Counterfactual analysis; Decision calibration; Graph-based reasoning; Multimodal signals; Hybrid intelligence; H3lix system; LAIZA protocol; process. †Associated IP: U.S. Provisional Utility Patent Application No. 63/910,500, “H3LIX: AI–Human Symbiotic Integration Process,” filed November 3, 2025 (process patent). 1 1 Introduction: Toward an Integrative Science of Cognition Contemporary science approaches intelligence from two divergent traditions (see [23,5]). The materialist paradigm explains cognition as an emergent computation of neural and algorithmic processes, while the noetic tradition regards consciousness as an intrinsic, causally active aspect of reality. Both generate valuable insights, yet neither alone can account for the full phenomenology of mind (see [52]). The former struggles with the hard problem of subjective experience (cf. [20]), while the latter lacks a reproducible empirical framework. This epistemic gap limits progress in understanding and engineering systems that genuinely know that they know. We propose that cognition can be modeled as a dual-aspect process within a single informational substrate (following [5]; cf. [20]), in which mind and matter represent complementary expressions of one continuum (cf. variational accounts under the freeenergy principle [28]). The H3LIX architecture operationalizes this premise through three interdependent layers: Somatic, Symbolic, and Noetic (compare to [36]). These layers form a helical tri-structure whose dynamic interactions are governed by feedback according to the SORK-N model derived from behavioral science ([49]). Within this system, the LAIZA Protocol serves as a research methodology designed to correlate subjective experience with objective measurement, thereby translating the study of consciousness from metaphysical speculation into testable empiricism (see [91]). The protocol collects synchronized physiological, linguistic, and phenomenological data streams, allowing for the detection of preconscious markers of decision (see [84]), intuition or nonlocal correlation (cf. [59]; but see [76] for critique). Analyses combine information-theoretic summaries (e.g. transfer entropy [75]) with causal reasoning over structured representations (see [68] and Granger-based effective connectivity [78]) to support falsifiable claims. By combining Bayesian and information-theoretic analyses with human-reported introspective states [26,88], the model aims to determine whether specific configurations of consciousness can measurably influence or anticipate physical and informational systems. The ultimate objective is not to replace existing paradigms, but to integrate them by establishing a unified, falsifiable framework where biological and artificial cognition co-evolve. In doing so, the H3LIX system positions itself as both a theoretical and practical instrument for exploring the frontier between neuroscience, information theory, and the emerging field of noetic science. 2 Theoretical Background Our approach is intentionally anchored in well-established scientific traditions rather than ad hoc speculation. We synthesize classical behaviorism (S-O-R-K), contemporary predictive processing and embodied cognition, and dual-aspect information perspectives into a single, testable framework, and we operationalize these links with rigorous mathematical tools (evidence-linked Mirrored Profile Graphs and Rogue Variable analysis). This grounding ensures that each later claim about coherence, 2 intuition, and meta-control rests on reproducible constructs and standard statistical methodology. 2.1 Behavioral Foundations: The SORK Model and Adaptive Feedback The classical S-O-R-K framework (Stimulus - Organism - Response - Kontingenz/Consequence) expanded behaviorist models by emphasizing organismic variables (perception, memory, motivation) and the reinforcement contingencies that shape future behavior (see [48]). We reference the SORK tradition as a canonical formalization of these contingencies in clinical and experimental settings (cf. [47,35]). H3LIX extends this logic into the digital and cognitive domain through SORK-N, where N denotes Noetic Integration. This fifth phase formalizes reflective modeling of the behavioral loop-supporting both first-order (behavioral) and second-order (metacognitive) adaptation (following [25]; see also [7]). 2.2 Predictive Processing and Embodied Cognition Predictive processing frames the brain as a hierarchical inference engine minimizing surprise by continuously updating predictions of sensory input (see [21,70]). This aligns with the free-energy formulation of cortical inference and active sensing (cf. [28, 67]). In parallel, embodied cognition views mind as the dynamic enactment of a body in a world (see [12]). Together they imply that any model approaching genuine situational awareness must integrate motor, affective, and sensory feedback, the Somatic Layer of H3LIX (cf. [77,1]). 2.3 Dual-Aspect Information Monism Interdisciplinary work in philosophy, neuroscience, and information theory proposes that matter and mind are complementary manifestations of a single informational substrate (see [89,65,4]). Convergent strands include information-theoretic models of consciousness and empirically tractable dual-aspect proposals. H3LIX operationalizes this stance by treating symbolic computation (objective information flow) and noetic resonance (subjective correlation) as coupled domains of one process, with the LAIZA Protocol measuring cross-correlations between them. 2.4 Noetic Research and the Empirical Challenge Psi/noetic research has explored extrasensory perception, telepathy, and precognition (e.g., [14]). Reported effects are typically small and contested (see [73]; cf. [93]). Meta-analytic evidence and high-powered replication attempts together motivate stringent, preregistered tests with balanced interpretation (e.g. [59,19]; and methodological guidance in [64]). H3LIX does not assume affirmative answers, but it provides a neutral empirical framework for testing them, embedding noetic hypotheses in predictive-processing and standard statistics. 3 2.5 Synthesis From behaviorism, H3LIX inherits reinforcement precision, from predictive neuroscience, error minimization, from dual-aspect theory, an ontological bridge, from noetic studies, the courage to test beyond current empiricism. Cognition becomes a recursive informational process. The LAIZA protocol is an experimental philosophy of mind rendered as method. 3 The H3LIX Architecture As we stated before, H3lix is a tripartite cognitive architecture (Somatic, Symbolic, Noetic) interacting via a unifying Mirror Core based on Mirror Profile Graph together with AI agentic system. The design captures cognition as a closed-loop information ecology. Layers may be analyzed separately, but they operate as one process. 3.1 The Somatic Layer The Somatic Layer provides the biophysical grounding of H3LIX. It aggregates lowlatency signals that co-vary with arousal, affect, motor preparation, and autonomic regulation (see [2,22]; cf. [81]), then converts them into a calibrated state vector suitable for predictive coding and cross-layer alignment (compare to [83]; cf. [67]). The general operational scheme is as follows 1. Signals and sensors. The catalog of signals processed by the system is not closed and includes all technically available sensory data comprising, among others, electrodermal activity (EDA/GSR), ECG or PPG for HR/HRV (time/frequency and non-linear indices), respiration, facial EM, oculometrics (blink, fixation, pupil), speech prosody, posture/micro-motion, EEG micro-potentials and narrow-band rhythms. 2. Acquisition and synchronization. Channels are time-locked to a master clock with event markers for stimuli, responses, and introspective reports. Short, overlapping windows yield robust features with per-channel uncertainty. 3. State estimation. Preprocessed streams x(t) produce standardized features z(t) and uncertainties σ(t). A light Bayesian filter yields a smoothed somatic state ˆ s(t) with innovation ε(t) (Somatic contribution to prediction error). This follows a lightweight Kalman family update, with physiological priors anchored in established HRV and pupillometry markers of arousal and uncertainty [46,79,43]. 4. Anticipatory markers. Event-locked windows preceding action/report are scanned for readiness-like trends and cross-modal phase-locking; markers are forwarded to LAIZA for preconscious analyses. 5. Outputs. As the result of the process, we obtain time-aligned vectors ˆ s(t) with given covariance, event-locked summaries, change-points/anomaly flags, all consumable by the system to be analyzed in the view of Mirrored Profile Graph. 4 3.2 The Symbolic Layer The Symbolic Layer is the architecture’s representational and inferential workspace. Operationally, it is realized by the LAIZA protocol, which coordinates three things: the storage and alignment of multiple data streams, the conversion of language and behavior into structured representations (see [10]), and a set of probabilistic reasoning routines that fuse somatic evidence with discrete symbols to produce beliefs, explanations, and policies (see [53,34]; cf. [44]). This layer translates observations into linguistic, mathematical, and logical forms, runs counterfactual simulations and “what-if” analyses (following [68]) and turns conclusions into candidate actions (compare to [29]). It provides top-down predictions that guide perception and behavior (cf. [70]), and it surfaces uncertainty so other parts of the system know when to rely on, question, or override its output (see [51]). 1. LAIZA keeps time-aligned records of text, speech, and behavior alongside the somatic summaries coming from sensors. Language is parsed into entities, events, and relations. Behaviors are captured as discrete acts and sequences and both are embedded into a shared representational space so that bodily cues and symbolic content can be compared directly. The layer then maintains a running “belief state” over this space-essentially a compact snapshot of what the system currently thinks is true, relevant, or likely together with a trace of how confident it is in each part of that belief. 2. The Symbolic layer generates predictions about upcoming words, events and outcomes. It compares those predictions with what actually happens and adjusts its internal representations and weights to reduce future mistakes. When errors persist or contradictions appear, it marks those areas as uncertain and invites either more evidence (from the Somatic Layer) or structural guidance (from the Noetic Layer). 3. Typical outputs include a machine-readable account of the current situation (entities, relations, intents), predictions with calibrated confidence, alternative plans with expected utility and risk, and concise indicators of where the model is unsure or misaligned with the data. 3.3 The Noetic Layer This layer monitors the whole system for signs of global coherence, i.e. those moments when bodily rhythms, internal narratives, and external outcomes line up. It detects states commonly described as intuition, insight, or anticipatory awareness by tracking how stable, synchronized, and mutually reinforcing the signals are across layers (cf. timeand phase-based coherence measures [54,92,32]). When such a state emerges, the layer treats it as a system-level configuration rather than a single cue, and it adjusts priors, attention, and decision cadence accordingly. The layer operationalizes introspection by turning “felt coherence” into testable, system-wide indicators. We summarize multistream regularities using complexity/entropy tools suited to short windows and non-Gaussian dynamics (see [11,72, 5 94]). The noetic layer asks whether particular mental or emotional configurations correspond to measurable reductions in noise or prediction error and evaluates this claim repeatedly across tasks and contexts. In doing so, it reframes “noetic” phenomena as patterns of non-local information resonance: distributed alignments that arise across sensors, symbols and outcomes. Instead of taking intuition at face value, it tests whether coherent states reliably precede more accurate judgments, faster convergence, or safer actions and under which boundary conditions they do not (see conditions for valid intuition [45]). The outputs of the layer can be described as 1. Correlation matrices: compact summaries of how strongly key signals co-vary (e.g., somatic markers with narrative consistency and performance), used to identify stable alignments versus brittle coincidences. 2. Entropy-change coefficients: indicators of whether the system is moving toward order or disorder, computed over relevant streams (physiology, language, policy choices) to flag emerging stability or fragmentation (cf. [11,72,94]). 3. Coherence spectra: timeand frequency-aware portraits of alignment that reveal when and at what scales coherence appears (e.g., sustained, transient, or oscillatory episodes) (see [54,32,92]). 4. Probabilistic measures of intuitive accuracy: calibrated estimates of how likely a detected coherent state will translate into better outcomes for the current task, derived from the system’s own history rather than a fixed prior (assessed with proper scoring rules [31]). Together, these outputs let the architecture use “intuition” as a disciplined control signal: detectable, auditable, and actionable. 3.4 Mirror Core and Feedback Dynamics The Mirror Core is the coordination hub that instantiates SORK-N at runtime. Building on S-O-R-K (see [48]), it adds Noetic Integration (N) to turn each behavioral loop into a reflective one, where we combine the first-order action and second-order to perform metacognitive adjustment (following [25]). Consequently, the structure of the cycle is S (Stimulus): ingest and time-align somatic and contextual signals. O (Organism): update the Symbolic Layer’s beliefs, expectations, and candidate actions. R (Response): execute/schedule actions and record precise outcomes (cf. [87]). K (Kontingenz): attach consequences and external feedback to the same clock (see [87]). 6 N (Noetic): assess global coherence, attribute anomalies, and set meta-adjustments (priors, thresholds, attention) (cf. adaptive uncertainty/attention control [95]; learning-rate/priors adaptation [60,13]). S’ (Write-back): apply those adjustments to the next stimulus epoch (adjusting decision thresholds/cadence as needed [71]). The following parameters are subject to regulation: attention and gating of inputs (cf. [95]), decision thresholds and cadence (see [71]), learning rates and priors (cf. [60,13]), and brief probes or inhibitions during periods of low coherence. In this manner, the Noetic Layer (N) continuously conditions S’, thereby closing the loop and enabling both behavioral and metacognitive adaptation. 4 The LAIZA Protocol The LAIZA protocol is a method used to study the dependencies between biological, symbolic, and noetic processes, based on AI models and mathematical modeling. However, this model is not rigidly defined. Instead, it is created by a system of cooperating artificial intelligence tools which, within the framework of the proposed mathematical and algorithmic mechanisms, form a structure reflecting the user’s cognitive construct (see [9,41]). In addition, the system is in a continuous process of creation and transformation, which itself is also a subject of analysis (cf. continual learning [66]). A very important element of the system is a vast stream of input data encompassing all available data related to the user, which ensures the possibility of in-depth analysis and the creation of a model that takes into account all required aspects (see digital phenotyping and safety considerations [40]). The immense progress in the field of artificial intelligence algorithms and the increased availability of technological tools related to data collection and processing finally enables the LAIZA protocol to become an empirically valid method that was not previously available (cf. wearable sensing advances [3] and foundation models [57]). We begin with presenting two main mathematical models used in the protocol, which are Mirrored Profile Graph (MPG) and Rogue Variable (RV) analysis. 4.1 Mirrored Profile Graph (MPG): A compact formalism MPG is an evidence-linked, hierarchical graph model of a person’s profile that plugs into the LAIZA protocol. It turns data and multimodal traces into a structured object that supports inference, feedback, and symbiotic AI co-adaptation (see knowledge graph formalisms [38,42] and LLM-powered construction [17]). It complements the Somatic–Symbolic–Noetic loop and SORK-N feedback by providing a personal, auditable state space over which those dynamics can operate (cf. provenance standards [58]). Concretely, each node captures psychologically meaningful constructs (traits, values, coping routines, cues, pivotal events) with provenance, while typed, directed edges encode relations such as causes,triggers,buffers, and moderators, yielding a compact causal-contextual scaffold (cf. [86,69]). 7 Beyond the flat view, MPG is hierarchical: any node may denote a segment (a subgraph) with its own evidence, metrics, and internal edges. Edges can target segments as wholes or project to specific subnodes via boundary interfaces, enabling horizontal paths within a level and vertical paths that traverse levels (node ↔segment; segment ↔segment). Treating segments as nodes at a higher level induces a recursive lift to meta-graphs of “thoughts about thoughts,” where segment-level importance and confidence are rolled up from their contents (see multi-scale/ hierarchical organization parallels [16,8]). Node/edge weights still separate importance (behavioral centrality) from confidence (evidence support), enabling principled updates as new traces arrive at any layer (e.g. probabilistic weighting [6], network centrality foundations [61]). The result is a human-interpretable, machine-legible profile: small edits to nodes, segments, edges, or weights propagate transparently through predictive models and AI-assistance strategies across levels (cf. relational/learning methods on KGs [63]). We begin the description of the mathematical model by introducing the layer-type graph and its structure, which then will be propagated into higher levels by the lift operator. Definition 1 ALayer-Type Graph is a typed, directed multigraph equipped with attribute maps: G = V, E, ΣE;τ, m, w, E,Conf,Imp, R, where •Nodes and edges. Vis the set of nodes; Eisamultiset of directed edges (parallel edges allowed). Thus, for u, v ∈Vthere may exist several distinct edges ek∈E(e.g., different relation types). •Layers. Lis an open label set, system-generated family of layers’ names (e.g., Psychological,Social,Historical,Family,Professional,Hobby) inferred from data. The node-to-layer map is λ:V→ P(L) (multi-labels allowed). •Edge types. ΣEis an extensible alphabet of relation types, initially seeded with causes, triggers,amplifies,buffers,moderates,enables,aligns,is part of,contradicts,correlates, etc. The type map is τ:E→ΣE. •Node metrics. m:V→[−1,1] ×[0,1]3returns (valence,intensity,recency,stability), capturing affective sign/magnitude,strength/frequency,freshness, and temporal consistency. •Edge strength. w:E→[0,1] gives the effect strength of each directed relation. •Evidence. E: (V∪E)→2Massigns each node/edge a set of evidence items ei∈ M. Each eistores: a short description and provenance class (selects quality multiplier qifor confidence), a resolvable pointer/identifier, a short verbatim or lightly quoted snippet supporting the claim, and a timestamp (for timeliness factor ti). Four numerical parameters accompany each item: item-level support ci∈[0,1] (how strongly this item supports the specific node/edge, independent of qi), source-quality multiplier qi>0, diversity bonus ui∈[1,1.15], and timeliness factor ti∈(0,1]. 8 •Confidence. Conf : (V∪E)→[0,1] is computed from evidence via S(x) = X ei∈E(x) ciqiuiti,Conf(x) = 1 −exp −αS(x),(1) with adaptable calibration parameter α∈(0,∞). •Importance. Imp : V→[0,1] returns the importance of every node based on five factors: valence,intensity,recency,stability, and centrality. Each factor comes from simple inputs and the current graph structure. All are scaled to [0,1] for easy comparison, except valence, which is bipolar [−1,1]. In particular, valence comes from pleasantness ratings, emotion/affect labels, and sentiment of the data; intensity from strength ratings and short-window frequency counts; recency from the age of the latest supporting episode or artifact; stability from temporal consistency and source agreement; and centrality from the node’s position in the current graph. The final computation uses a learnable linear model. Node importance Impvand Imputogether with edge strength are then used to measure the contextual edge priority for the edge u→v, which is one of the key elements in the system. •Reasoning provenance. R: (V∪E)→Text stores a short, human-readable justification for why each node or edge was created (e.g., the rule-of-thumb, abductive step, or mapping from an interview answer), distinct from raw evidence. This enables transparent audits and model critique. We denote by GLT the family of all Layer-Type Graphs. The next construction step is to describe the segmentation mechanism and the process of building the higher-level graph structure. The proposed procedure is universal and is an algorithmic method for moving from a given level to a higher level (cf. multilevel partitioning and hierarchical coarsening [50,18]; see also flow-based segmentation [74] and spectral methods [62]). The starting point is a graph with the structure described above. The characteristics of the individual vertices and edges allow for the identification of segments, which are subgraphs consisting of a set of vertices and edges with specific properties (topological, semantic, functional, thematic, etc.) (see [27,15]). Definition 2 Given a level-kgraph of Layer-Type Graph G(k)= (V(k), E(k),ΣE;τ(k), m(k), w(k),E(k),Conf(k),Imp(k), R(k)) asegment is any subgraph S= (VS, ES) with VS⊆V(k),ES⊆E(k). Let S(k)= {S1,...,SNk}denote a chosen (possibly overlapping) family of segments at level k, where Nkis the number of selected segments for given level k∈ {0,1,2, . . .}. The selected segments become the vertices for the graph at the higher level. Note that a selected segment might be just one node, which allows us to consider also segment to node and node to segment relations. Additionally, each segment has a specified boundary interface, which are groups of vertices that connect with other segments 9 •Learning and re-coherence. Outcomes from interventions update edge strengths, importances, and interface selections; summaries roll upward across levels and confirmed higher-level relations project back down so coherence is restored as prediction error subsides. •Governance and impact accounting. All changes are validated, versioned, and narrated; the system tracks which events or chains most shaped structure and policy so future decisions become faster, safer, and more explainable. 5 MPG intuition model Many concepts included in the broadly conceived noetic grasp are considered unscientific primarily because they lack a well-established definition and a set of operational guidelines within which this definition can be applied and tested experimentally. The mathematical models presented in the previous chapters allow for a very specific description of phenomena that have previously been completely elusive. In this chapter, we will present a strictly mathematical definition of intuition within the MPG model (MPG-Intuition), which, thanks to digital technology and advanced artificial intelligence models, can be tested experimentally. We will begin by defining intuition within the MPG model as a process occurring in our mind (modeled by the system), where there is a difference between all the data and structures utilized by the brain (system) during inference and prediction, and those that are consciously perceived (cf.[37], [82]). This difference must be sufficiently significant, which is also formally defined, so as to cause a change in the final decision made. Setup. Our starting point is the Mirrored Profile Graph MPG = {G(0),G(1) . . . , G(N)} Additionally, we have a defined set of decisions that must be taken, which we denote by H0. This could be a simple binary decision (H0={0,1}), a classification problem (H0={h1, . . . , hn}), or any continuous problem (H0⊆Rd). Furthermore, we have a set of observations, which we denote by O={o1, . . . , om}. Our goal is to make a decision based on the observed data Oand the MPG graph, a process which we model with the decision rule ˆ h=ˆ h(O, MPG, R(H)) and utility rates R:H→Rd, which in general might be multidimensional. Unawareness factors. As we have already formulated, the key concept in the definition of intuition is the notion of a lack of full awareness. We propose two types of unawareness that need to be considered. •Input Unawareness (IU). The first is the simplest and concerns the fact that we lack full awareness regarding the observed data related to the decision being made. This means that our awareness pertains to the set ˜ O, which is different from the set O; specifically, ˜ O⊊O. Consequently, we define the set of unaware inputs as Uin =O\˜ O. 16 •Process Unawareness (PU). The second type of unawareness concerns the lack of cognitive knowledge regarding certain parts of the MPG graph that are significant in the decision-making process we are discussing. Similarly as before, we define Uproc ⊆MPG as unaware part of the graph. The next step is to indicate a measure that will determine whether the level of unawareness is large enough to have a significant impact on the decision-making process. We propose that this be done by pointing to the existence of a set of unaware states or unaware objective goals that flip or change the decision. Accordingly, we define the total set of unaware states as U=Uin ∪ Uproc. Definition 7 A set U⊂ U is called Minimal Unaware Flip Set (MUFS) iff removing all elements in Uflips or changes ˆ h, while removing any proper subset of Udoes not. Based on that definition we can proceed directly to the definition of intuition, more precisely intuitive decision in the MPG based system. Definition 8 (MPG - Intuition) A decision is said to be MPG-Intuitive iff there exists at least one nonempty Minimal Unaware Flip Set U⊆ U. This is not the only possible definition that can be considered in the context of a mathematical model; however, it serves as a starting point and is sufficiently comprehensive yet mathematically precise to form the basis for experimental research. Subsequent possible generalizations and applications of the entire system, including the mathematical model, are not only related to more subtle definitions of intuitionwhich include, among other things, the actual difference between the established prediction target and the true prediction target, or a difference in the loss function or the estimator itself. The mathematical foundation of the system also allows us to provide strictly formal definitions of other phenomena related to cognition and metacognition, which are collectively termed noetic (such as insight, for example). All these concepts will be the subject of further research. 6 Experimental Strategy for Testing MPG - Intuition We propose a compact program of experiments that probes MPG - Intuition at three levels: system-only validation inside H3LIX, human decision replication with the system running in parallel, and a link to self-reported intuition. In all cases, the core manipulation contrasts a full awareness condition, where all inputs and model processes are available with a restricted awareness condition, created either by masking inputs (Input Unawareness, IU) or by ablating limited graph structure in the MPG (Process Unawareness, PU). The same tasks are used across levels with difficulty adapted to maintain comparable performance envelopes. 17 System-only (internal validation). Within H3LIX we run matched blocks under full and restricted awareness. IU is induced by withholding subtle cues and microevents from the Symbolic/Noetic layers while retaining their somatic effects. PU is induced by temporarily disabling selected MPG segments or cross-level pathways identified a priori by RV potency. For each trial LAIZA searches counterfactually for a Minimal Unaware Flip Set (MUFS), the smallest set of masked inputs or ablated structures, whose restoration reverses the decision. Primary readouts are: the rate at which a nonempty MUFS exists, the probability of decision flip relative to full, and task accuracy/calibration as a function of Noetic coherence. MPG - Intuition is supported if restricted blocks yield more MUFS and more flips than full, yet trials marked as highly coherent preserve (or even improve) accuracy and confidence calibration despite restriction. Human decisions (external confirmation). The very same stimuli and timing are presented to human participants while the system operates in parallel. Human IU is produced with standard psychophysical methods (brief masking, peripheral microcues) verified by awareness checks; PU remains a silent, system-internal ablation so that the information available to the participant is unchanged. We compare human choices to system choices across full and restricted blocks, focusing on the incidence of flips between block types and on their alignment with trials in which the system detected a MUFS. Convergent evidence is obtained if human decisions flip more often under restricted than full, and if these flips co-occur with system MUFS trials on the same items, indicating shared dependence on minimal unaware factors. Intuition self-report linkage. Immediately after each restricted trial, participants provide a brief introspective rating accompanied by confidence. These reports are related to contemporaneous Noetic measures, to MUFS presence and size and to behavioral outcomes (accuracy, calibration, and latency). The key prediction is that self-labeled intuitive trials exhibit higher Noetic coherence, a greater likelihood of containing a MUFS, and superior calibration/accuracy relative to nonintuitive trials-specifically under restricted awareness, where deliberation is limited and meta-cognitive integration should matter most. 7 Conclusions We have presented H3LIX, a tripartite cognitive architecture that unifies somatic grounding, symbolic inference, and noetic metacognition within a single, timesynchronized control loop (SORK-N). Coupled with the LAIZA protocol, the system translates traditionally elusive phenomena-intuition, insight, anticipatory awareness—into operational quantities that can be measured, analyzed, and used for closed-loop control. At the center of this framework stands the Mirrored Profile Graph (MPG), an evidence-linked, hierarchical state space that supports transparent updates, cross-level reasoning, and human-auditable provenance. A central contribution of the work is a formalization of surprise and deviation via Rogue Variables (RVs), which identify structural patterns-segments and interlevel pathways with outsized explanatory power for prediction-observation gaps. This provides a principled way to prioritize interventions and relate model dynamics to 18 interpretable mechanisms. Building on this substrate, we defined MPG - Intuition as the existence of a nonempty Minimal Unaware Flip Set (MUFS): a minimal set of inputs or graph structures outside conscious access whose removal flips a decision. This definition shifts “intuition” from anecdote to criterion, linking meta-cognitive coherence to concrete, testable counterfactuals. Methodologically, we outlined an experimental program that evaluates MPG - Intuition under full versus restricted awareness, both within the system and with human participants, including brief self-reports of intuitive experience. The proposed readouts-coherence spectra, entropy-change coefficients, MUFS incidence, calibration, and accuracy-allow us to test whether coherent states preserve or improve decision quality when deliberation is limited, and whether human “intuitive” choices align with system-detected MUFS on the same items. Together, these studies close the loop between theory, mathematical formalism, and reproducible evidence. Conceptually, H3LIX integrates multiple traditions without collapsing their distinctions. From behaviorism it inherits precise reinforcement contingencies, from predictive processing, error minimization and uncertainty management, from dualaspect information monism, an ontological bridge between matter and mind and from noetic studies, a willingness to test hypotheses about non-local coherence under falsifiable designs. The Mirror Core operationalizes this synthesis by writing noetic meta-feedback back into stimulus processing (N→S′), enabling both first-order (behavioral) and second-order (meta-cognitive) adaptation. There are, however, clear limitations. Our formalizations do not entail claims about the metaphysical status of consciousness; they specify decision-level criteria and control signals within a bounded architecture. Empirical effects related to pre-event or non-local coherence, if present, are likely small and context-dependent; rigorous preregistration, device-level controls, and multi-lab replications will be crucial. Finally, ethical considerations-privacy of multimodal traces, interpretability of structural edits in the MPG, and human oversight of meta-policy changes and must accompany any deployment beyond the lab. Despite these caveats, the framework offers a tractable route toward integrative cognition. By binding physiological signals, symbolic structure, and metacognitive coherence into a single, auditable loop, H3LIX provides both a scientific instrument for studying mind and a practical blueprint for human-AI symbiosis. We expect that refining RV potency measures, expanding the family of coherence indicators, and scaling the experimental program to collaborative settings (multi-agent MPG echoes) will further clarify when and how “intuition” functions as a reliable control signal. In this sense, the path forward is iterative: instrument, test, revise, and re-instrument-allowing biological and artificial cognition to co-evolve under a shared, falsifiable methodology. References [1] Misha Ahissar and Elad Assa. Perception as a closed-loop convergence process. Current Opinion in Neurobiology, 40:58–65, 2016. [2] Gary Aston-Jones and Jonathan D. Cohen. An integrative theory of locus coeruleus–norepinephrine function: Adaptive gain and optimal performance. 19 Annual Review of Neuroscience, 28:403–450, 2005. [3] Hacer C. Ates, Bader Alizadeh, Karin Dincer, Aysu M. Elma, Elias M. W¨urthner, Can Dincer, Gerald A. Urban, and Can Dincer. End-to-end design of wearable sensors. Nature Reviews Materials, 7(10):887–907, 2022. [4] Harald Atmanspacher. Dual-aspect monism `a la pauli and jung. In Paul Davies and Niels Henrik Gregersen, editors, Information and the Nature of Reality: From Physics to Metaphysics, pages 260–274. Cambridge University Press, Cambridge, 2010. [5] Harald Atmanspacher. A brief primer for dual-aspect monism. Synthese, 203(9):1–24, 2024. [6] Stephen H. Bach, Matthias Broecheler, Bert Huang, and Lise Getoor. Hinge-loss markov random fields and probabilistic soft logic. Journal of Machine Learning Research, 18(109):1–67, 2017. [7] David Badre. Cognitive control. Annual Review of Psychology, 75:1–28, 2024. [8] David Badre and Derek E. Nee. Frontal cortex and the hierarchical control of behavior. Neuron, 99(5):993–1004, 2018. [9] Tadas Baltruˇsaitis, Chaitanya Ahuja, and Louis-Philippe Morency. Multimodal machine learning: A survey and taxonomy. IEEE Transactions on Pattern Analysis and Machine Intelligence, 41(2):423–443, 2019. [10] Laura Banarescu, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. Abstract meaning representation for sembanking. In Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse, pages 178–186, 2013. [11] Christoph Bandt and Bernd Pompe. Permutation entropy: A natural complexity measure for time series. Physical Review Letters, 88(17):174102, 2002. [12] Lawrence W. Barsalou. Grounded cognition. Annual Review of Psychology, 59:617–645, 2008. [13] Timothy E. Behrens, Mark W. Woolrich, Mark E. Walton, and Matthew F. S. Rushworth. Learning the value of information in an uncertain world. Nature Neuroscience, 10(9):1214–1221, 2007. [14] Daryl J. Bem. Feeling the future: Experimental evidence for anomalous retroactive influences on cognition and affect. Journal of Personality and Social Psychology, 100(3):407–425, 2011. [15] Austin R. Benson, David F. Gleich, and Jure Leskovec. Higher-order organization of complex networks. Science, 353(6295):163–166, 2016. [16] Richard F. Betzel and Danielle S. Bassett. Multi-scale brain networks. Trends in Cognitive Sciences, 21(8):589–602, 2017. [17] Haonan Bian. Llm-empowered knowledge graph construction: A survey. arXiv, 2025. [18] Vincent D. Blondel, Jean-Loup Guillaume, Renaud Lambiotte, and Etienne Lefebvre. Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, (10):P10008, 2008. [19] Etzel Carde˜na. The experimental evidence for parapsychological phenomena: A review. American Psychologist, 73(5):663–677, 2018. 20 [20] David J. Chalmers. Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3):200–219, 1995. [21] Andy Clark. Whatever next? predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3):181–204, 2013. [22] Hugo D. Critchley and Neil A. Harrison. Visceral influences on brain and behavior. Nature Reviews Neuroscience, 14(6):394–406, 2013. [23] Stanislas Dehaene, Hakwan Lau, and Sid Kouider. What is consciousness, and could machines have it? Science, 358(6362):486–492, 2017. [24] Jean-Charles Delvenne, Sophia N. Yaliraki, and Mauricio Barahona. Stability of graph communities across time scales. Proceedings of the National Academy of Sciences, 107(29):12755–12760, 2010. [25] Stephen M. Fleming and Nathaniel D. Daw. Self-evaluation of decision-making: A general bayesian framework for metacognitive computation. Psychological Review, 124(1):91–114, 2017. [26] Stephen M. Fleming and Hakwan C. Lau. How to measure metacognition. Frontiers in Human Neuroscience, 8:443, 2014. [27] Santo Fortunato and Darko Hric. Community detection in networks: A user guide. Physics Reports, 659:1–44, 2016. [28] Karl J. Friston. The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11:127–138, 2010. [29] Malik Ghallab, Dana Nau, and Paolo Traverso. Automated Planning: Theory and Practice. Morgan Kaufmann, San Francisco, 2004. [30] John C. Gittins, Kevin Glazebrook, and Richard Weber. Multi-Armed Bandit Allocation Indices. Wiley, Chichester, 2 edition, 2011. [31] Tilmann Gneiting and Adrian E. Raftery. Strictly proper scoring rules, prediction, and estimation. Journal of the Royal Statistical Society: Series B, 69(2):243–268, 2007. [32] Aslak Grinsted, John C. Moore, and Svetlana Jevrejeva. Application of the cross wavelet transform and wavelet coherence to geophysical time series. Nonlinear Processes in Geophysics, 11(5/6):561–566, 2004. [33] Aditya Grover and Jure Leskovec. node2vec: Scalable feature learning for networks. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 855–864, 2016. [34] David L. Hall and John Llinas. An introduction to multisensor data fusion. Proceedings of the IEEE, 85(1):6–23, 1997. [35] Gregory P. Hanley, Brian A. Iwata, and Brandon E. McCord. Functional analysis of problem behavior: A review. Journal of Applied Behavior Analysis, 36(2):147– 185, 2003. [36] Stevan Harnad. The symbol grounding problem. Physica D: Nonlinear Phenomena, 42(1–3):335–346, 1990. [37] Aviad Heifetz, Martin Meier, and Burkhard C. Schipper. Interactive unawareness. Econometrica, 74(6):1525–1556, 2006. [38] Aidan Hogan, Eva Blomqvist, Michael Cochez, Claudia d’Amato, Gerard de Melo, Claudio Guti´errez, Jos´e Emilio Labra Gayo, Sabrina Kirrane, Sebastian Neumaier, Axel Polleres, Roberto Navigli, Axel-Cyrille Ngonga Ngomo, Sebastian 21 Rudolph, Anna Tordai, and Rapha¨el Troncy. Knowledge graphs. ACM Computing Surveys, 54(4):1–37, 2021. [39] Petter Holme and Jari Saram¨aki. Temporal networks. Physics Reports, 519(3):97– 125, 2012. [40] Kit Huckvale, Svetha Venkatesh, and Helen Christensen. Toward clinical digital phenotyping: a timely opportunity to consider purpose, quality, and safety. npj Digital Medicine, 2:88, 2019. [41] Frank Hutter, Lars Kotthoff, and Joaquin Vanschoren, editors. Automated Machine Learning: Methods, Systems, Challenges. Springer, Cham, 2019. [42] Shaoxiong Ji, Shirui Pan, Erik Cambria, Pekka Marttinen, and S Yu Philip. A survey on knowledge graphs: Representation, acquisition, and applications. IEEE Transactions on Knowledge and Data Engineering, 33(5):312–333, 2021. [43] Sagar Joshi and Joshua I. Gold. Pupil size as a window on neural substrates of cognition. Trends in Cognitive Sciences, 24(6):466–480, 2020. [44] Leslie Pack Kaelbling, Michael L. Littman, and Anthony R. Cassandra. Planning and acting in partially observable stochastic domains. Artificial Intelligence, 101(1–2):99–134, 1998. [45] Daniel Kahneman and Gary Klein. Conditions for intuitive expertise: a failure to disagree. American Psychologist, 64(6):515–526, 2009. [46] Rudolf E. Kalman. A new approach to linear filtering and prediction problems. Journal of Basic Engineering, 82(1):35–45, 1960. [47] Frederick H. Kanfer, Hans Reinecker, and Dieter Schmelzer. SelbstmanagementTherapie: Ein Lehrbuch f¨ur die klinische Praxis. Springer, Berlin, 5 edition, 2012. [48] Frederick H. Kanfer and Gordon Saslow. Behavioral analysis: An alternative to diagnostic classification. In P. H. Hoch and J. Zubin, editors, The Application of Learning Principles to Clinical Problems, pages 211–238. Grune & Stratton, New York, 1965. SORCK model (Stimulus–Organism–Response–Contingency–Kontingenz). [49] Frederick H. Kanfer and Laurence G. Saslow. Behavioral analysis. Behavior Modification, 1(2):159–191, 1977. [50] George Karypis and Vipin Kumar. A fast and high quality multilevel scheme for partitioning irregular graphs. SIAM Journal on Scientific Computing, 20(1):359– 392, 1998. [51] David C. Knill and Alexandre Pouget. The bayesian brain: the role of uncertainty in neural coding and computation. Trends in Neurosciences, 27(12):712–719, 2004. [52] Christof Koch, Marcello Massimini, Melanie Boly, and Giulio Tononi. Neural correlates of consciousness: progress and problems. Nature Reviews Neuroscience, 17:307–321, 2016. [53] Daphne Koller and Nir Friedman. Probabilistic Graphical Models: Principles and Techniques. MIT Press, Cambridge, MA, 2009. [54] Jean-Philippe Lachaux, Eugenio Rodriguez, Jacques Martinerie, and Francisco J. Varela. Measuring phase synchrony in brain signals. Human Brain Mapping, 8(4):194–208, 1999. 22 [55] Scott M. Lundberg and Su-In Lee. A unified approach to interpreting model predictions. In Advances in Neural Information Processing Systems 30, pages 4765–4774, 2017. [56] Ron Milo, Shalev Itzkovitz, Nadav Kashtan, Reuven Levitt, Shai Shen-Orr, Ilan Ayzenshtat, Michal Sheffer, and Uri Alon. Network motifs: simple building blocks of complex networks. Science, 298(5594):824–827, 2002. [57] Michael Moor, P´eter Riba, Max Welling, et al. Foundation models for generalist medical artificial intelligence. Nature, 616:259–265, 2023. [58] Luc Moreau, Paolo Missier, and the W3C PROV Working Group. Prov-dm: The prov data model. W3C Recommendation, 2013. https://www.w3.org/TR/provdm/. [59] Julia Mossbridge, Patrizio E. Tressoldi, and Jessica Utts. Predictive physiological anticipation preceding seemingly unpredictable stimuli: a meta-analysis. Frontiers in Psychology, 3:390, 2012. [60] Matthew R. Nassar, Kevin T. Wilson, William F. Heasly, and Joshua I. Gold. An approximately bayesian delta-rule model explains the dynamics of belief updating in a changing environment. Proceedings of the National Academy of Sciences, 107(37):15990–15995, 2010. [61] M. E. J. Newman. Networks: An Introduction. Oxford University Press, Oxford, 2010. [62] Andrew Y. Ng, Michael I. Jordan, and Yair Weiss. On spectral clustering: Analysis and an algorithm. In Advances in Neural Information Processing Systems, volume 14, pages 849–856, 2002. [63] Maximilian Nickel, Kevin Murphy, Volker Tresp, and Evgeniy Gabrilovich. A review of relational machine learning for knowledge graphs. Proceedings of the IEEE, 104(1):11–33, 2016. [64] Brian A. Nosek, Charles R. Ebersole, Alexander C. DeHaven, and David T. Mellor. The preregistration revolution. Proceedings of the National Academy of Sciences, 115(11):2600–2606, 2018. [65] Masafumi Oizumi, Larissa Albantakis, and Giulio Tononi. From the phenomenology to the mechanisms of consciousness: Integrated information theory 3.0. PLOS Computational Biology, 10(5):e1003588, 2014. [66] German I. Parisi, Ronald Kemker, Jose L. Part, Christopher Kanan, and Stefan Wermter. Continual lifelong learning with neural networks: A review. Neural Networks, 113:54–71, 2019. [67] Thomas Parr, Giovanni Pezzulo, and Karl J. Friston. Active inference: twenty years on. Nature Reviews Neuroscience, 2022. [68] Judea Pearl. Causality: Models, Reasoning, and Inference. Cambridge University Press, Cambridge, 2 edition, 2009. [69] Jonas Peters, Dominik Janzing, and Bernhard Sch¨olkopf. Elements of Causal Inference: Foundations and Learning Algorithms. MIT Press, Cambridge, MA, 2017. [70] Rajesh P. N. Rao and Dana H. Ballard. Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects. Nature Neuroscience, 2(1):79–87, 1999. 23 [71] Roger Ratcliff and Gail McKoon. The diffusion decision model: theory and data for two-choice decision tasks. Neural Computation, 20(4):873–922, 2008. [72] Joshua S. Richman and J. Randall Moorman. Physiological time-series analysis using approximate entropy and sample entropy. American Journal of PhysiologyHeart and Circulatory Physiology, 278(6):H2039–H2049, 2000. [73] Stuart J. Ritchie, Richard Wiseman, and Christopher C. French. Failing the future: Three unsuccessful attempts to replicate bem’s ’retroactive facilitation of recall’ effect. PLOS ONE, 7(3):e33423, 2012. [74] Martin Rosvall and Carl T. Bergstrom. Maps of random walks on complex networks reveal community structure. Proceedings of the National Academy of Sciences, 105(4):1118–1123, 2008. [75] Thomas Schreiber. Measuring information transfer. Physical Review Letters, 85(2):461–464, 2000. [76] D. Samuel Schwarzkopf. We should have seen this coming. Frontiers in Human Neuroscience, 8:332, 2014. [77] Anil K. Seth. Interoceptive inference, emotion, and the embodied self. Trends in Cognitive Sciences, 17(11):565–573, 2013. [78] Anil K. Seth, Adam B. Barrett, and Lionel Barnett. Granger causality analysis in neuroscience and neuroimaging. The Journal of Neuroscience, 35(8):3293–3297, 2015. [79] Fred Shaffer and J. P. Ginsberg. An overview of heart rate variability metrics and norms. Frontiers in Public Health, 5:258, 2017. [80] Lloyd S. Shapley. A value for n-person games. Contributions to the Theory of Games, 2:307–317, 1953. [81] Hiroshi Shibasaki and Mark Hallett. What is the bereitschaftspotential? Clinical Neurophysiology, 117(11):2341–2356, 2006. [82] Christopher A. Sims. Implications of rational inattention. Journal of Monetary Economics, 50(3):665–690, 2003. [83] Anne C. Smith and Emery N. Brown. Estimating a state-space model from point process observations. Neural Computation, 15(5):965–991, 2003. [84] Chun Siong Soon, Marcel Brass, Hans-Jochen Heinze, and John-Dylan Haynes. Unconscious determinants of free decisions in the human brain. Nature Neuroscience, 11(5):543–545, 2008. [85] Erik ˇ Strumbelj and Igor Kononenko. Explaining prediction models and individual predictions with feature contributions. Knowledge and Information Systems, 41(3):647–665, 2014. [86] Yizhou Sun and Jiawei Han. Mining heterogeneous information networks: A structural analysis. IEEE Transactions on Knowledge and Data Engineering, 24(12):2507–2530, 2012. [87] Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction. MIT Press, Cambridge, MA, 2 edition, 2018. [88] Christopher Timmermann, Leor Roseman, Robin L. Carhart-Harris, et al. A neurophenomenological approach to non-ordinary states of consciousness: Integrating first-person examinations and neurophysiology. Trends in Cognitive Sciences, 26(12):1078–1091, 2022. 24 [89] Giulio Tononi, Melanie Boly, Marcello Massimini, and Christof Koch. Integrated information theory: From consciousness to its physical substrate. Nature Reviews Neuroscience, 17(7):450–461, 2016. [90] Charles Truong, Laurent Oudre, and Nicolas Vayatis. Selective review of offline change point detection methods. Signal Processing, 167:107299, 2020. [91] Francisco J. Varela. Neurophenomenology: A methodological remedy for the hard problem. Journal of Consciousness Studies, 3(4):330–349, 1996. [92] Martin Vinck, Robert Oostenveld, Marijn van Wingerden, Franscesco Battaglia, and Cyriel M. A. Pennartz. An improved index of phase-synchronization for electrophysiological data in the presence of volume-conduction, noise and samplesize bias. NeuroImage, 55(4):1548–1565, 2011. [93] Eric-Jan Wagenmakers, Ruud Wetzels, Denny Borsboom, and Han L. J. van der Maas. Why psychologists must change the way they analyze their data: The case of psi (comment on bem, 2011). Journal of Personality and Social Psychology, 100(3):426–432, 2011. [94] Shuen-De Wu, Chia-Chi Wu, Kuan-Lin Lin, and Chan-Cheng Wang. Time series analysis using refined composite multiscale entropy. Physica A: Statistical Mechanics and its Applications, 392(24):5865–5873, 2013. [95] Angela J. Yu and Peter Dayan. Uncertainty, neuromodulation, and attention. Neuron, 46(4):681–692, 2005. 25