The Internal State and Agency Layer

The missing half of the Commonsent model

A person never receives information as a neutral processor. The body and mind constitute the operating conditions under which every external signal is interpreted.

Two people may receive the same employment message, political alert, medical recommendation, family request, or financial decision and respond in radically different ways. The difference may not reflect values or intelligence. One person may be rested, metabolically stable, socially secure, and able to concentrate. The other may be sleep-deprived, in pain, physiologically activated, interrupted repeatedly, or primed by prior conflict. The external content is identical; the decision environment is not.

Commonsent therefore needs to model the individual as a dynamic system embedded in an environment rather than as a stable preference profile. The relevant question is not simply, “What does this person believe?” It is also, “What is the person’s present capacity to understand, deliberate, decide, communicate, resist manipulation, and act?” WHO defines mental well-being partly through the ability to cope, learn, work, and contribute to community, and emphasizes that individual, family, community, and structural factors interact to strengthen or undermine it.^[1][2]

Commonsent should not attempt to make the person permanently optimized. It should help the person recognize the conditions under which their agency expands or contracts.

The proposed layer has four functions. First, it observes physiological and behavioral signals with the least intrusive means available. Second, it builds a personalized model that separates direct measurements from psychological inference. Third, it links state transitions to external events and internal vulnerability without claiming certainty where only association exists. Fourth, it adapts information, timing, and coordination support so that the system reduces rather than adds to cognitive burden.

Why this is a coordination layer, not merely a health module

Physiological and psychological state influences whether a person can participate in collective life. Exhaustion affects whether someone reads a policy summary. Overload affects whether they speak in a meeting. Social threat affects whether they dissent. Low perceived control affects whether they believe action is worth attempting. Acute arousal can make urgency framing more persuasive. Chronic strain can narrow attention toward immediate survival and away from long-term civic interests.

The layer therefore connects personal health, cognitive ergonomics, anti-manipulation, family coordination, workplace functioning, and democracy. It also creates a feedback path that current social systems largely lack: institutions can measure whether a policy was enacted, but they rarely observe whether the process systematically overloads, alienates, frightens, or excludes the people expected to use it.

Measurement, inference, and causal humility

Any system that interprets internal state can become harmful if it presents uncertain inferences as psychological facts. Current sensors can measure movement, heartbeat timing, skin conductance, temperature, pupil diameter, gaze direction, and other observable signals. They can derive variables such as HRV, respiration, sleep estimates, and glucose trends. They can support probabilistic inferences about workload or arousal. They cannot directly read a complex emotion, value, intention, or cause.

Research on wearable stress detection typically combines HRV, EDA, PPG, motion, respiration, sleep, and sometimes EEG. Reviews report promising classification results but repeatedly identify small samples, controlled laboratory settings, inconsistent protocols, and weak real-world generalization as limitations.^[3][4] Eye tracking can continuously reflect workload through pupil, blink, fixation, and gaze patterns, but lighting, task type, expertise, medication, and individual differences complicate interpretation.^[5]

Digital phenotyping shows a similar pattern. Mobility, sleep regularity, phone use, social interaction proxies, and activity can correlate with stress, anxiety, or depression, but the meaning of the same behavior changes across roles and contexts. Reduced movement may indicate withdrawal in one setting and productive focus in another.^[6] The Commonsent model must therefore be predominantly within-person. Population models can initialize the system, but the user’s own baseline, corrections, and repeated patterns should progressively dominate.

This formulation makes an important conceptual change. Commonsent does not ask a binary question such as whether the user is stressed. It asks how the present state differs from the user’s relevant baseline, what evidence supports the change, what competing explanations remain, and whether an intervention is likely to help.

Endorphins and the temptation of false biological precision

Commonsent should not claim to measure “endorphin level,” “dopamine level,” or “serotonin level” continuously with present consumer technology. Central neurochemical activity cannot be inferred reliably from a watch, camera, voice sample, or ordinary peripheral blood reading. Even future biochemical wearables will initially measure accessible molecules in sweat or interstitial fluid, whose relationship to brain function may be indirect, delayed, context-dependent, or insufficiently validated.

A defensible alternative is to model functional constructs: reward responsiveness, motivation, initiation latency, positive affect, activity, sleep stability, social engagement, and response to known restorative activities. These remain inferences, but they correspond to observable patterns and can be corrected by the user.

Present capabilities: what can be built now

A useful first version does not require futuristic glasses or brain-reading. It requires disciplined integration of sensors that already exist, brief user confirmation, and a personal model that respects uncertainty.

3.1 Current sensing stack

Current consumer wearables already combine heart rate, ECG, respiratory rate, wrist temperature, sleep, blood oxygen, and motion-derived features, while eye-tracked headsets provide high-frequency gaze input.^[22][24] Over-the-counter CGMs are now available for adults, and in June 2026 the FDA cleared an OTC CGM for children aged two and older under adult caregiver supervision.^[7][8] That expansion makes glucose a realistic optional input for family and metabolic-learning use cases, although Commonsent should avoid turning normal variation into anxiety or prescriptive medical advice.

Cuffless blood-pressure estimation should remain outside the primary state model except as an experimental trend input. The American Heart Association’s 2025 scientific statement concluded that many cuffless devices have not demonstrated sufficient real-world accuracy for diagnosis or treatment and remain sensitive to calibration, posture, motion, skin characteristics, and context.^[9]

3.2 Present software architecture

A present-day Commonsent MVP should combine passive sensing with low-burden ecological momentary assessment. Instead of asking users to complete long questionnaires, the system would detect meaningful transitions and ask a two-to-five-second question: “Your activation rose after this interaction. Did it feel threatening, exciting, overloaded, physically strenuous, or unclear?” Such confirmations provide labels that sensors cannot supply and teach the model how this particular person maps physiology to experience.

The software stack should include rolling baselines conditioned on time of day, recent sleep, location, activity, and typical task. Change-point detection identifies when a pattern shifted. Event alignment checks what happened immediately before and after. A state-space model estimates persistence and lag. A causal-hypothesis layer compares internal, external, and interactive explanations. A receptivity model determines whether support should be offered at all.

3.3 Present-day outputs

Just-in-time adaptive interventions already provide a research framework for using momentary context to decide whether and how to intervene. Micro-randomized trials can test which prompt works, for whom, and under what conditions, while detecting notification burden and habituation.^[16] Early wearable-triggered trials demonstrate feasibility and some proximal benefits, but effects remain use-case-specific and often come from small studies.^[17]

The next two to three years: convergent wearable intelligence

The most important near-term advance is not a single new sensor. It is the convergence of environmental perception, eye tracking, ear-worn physiology, on-device AI, and personalized causal learning.

Current everyday AI glasses generally emphasize camera, microphone, speakers, voice interaction, translation, and contextual assistance rather than physiological sensing.^[25] Research glasses already integrate eye tracking, cameras, inertial sensing, spatial mapping, and first-person context; Meta’s Aria Gen 2, for example, includes camera-based eye tracking and six-degree-of-freedom spatial tracking as a research platform.^[23] Eye-tracked headsets such as Apple Vision Pro demonstrate that high-speed gaze input can operate reliably as an interface, although headset weight, battery life, social acceptability, and privacy make them different from all-day glasses.^[22]

4.1 Likely capability gains by 2028–2029

4.2 What should still remain out of scope

Even with better glasses and ear-worn systems, detailed belief decoding, truth detection, personality inference, and continuous emotion classification should remain outside Commonsent’s claims. EEG has limited spatial resolution and is vulnerable to motion and muscle artifacts. Pupillometry reflects several processes simultaneously. Voice and facial behavior are culturally and individually variable. Multimodal systems can become more useful without becoming omniscient.

4.3 Well-being gains in the two-to-three-year horizon

The three-to-five-year horizon: selective biological depth

The medium-term opportunity is not universal biochemical surveillance. It is the careful addition of a small number of validated biomarkers where they materially improve explanation or intervention.

5.1 Biochemical wearables

Wearable sweat and interstitial-fluid sensing is advancing quickly. Research devices can monitor lactate, glucose, drugs, inflammatory markers, and multiple analytes through microneedles or microfluidic platforms. Reviews describe interstitial fluid as a clinically relevant source that can be accessed through microneedles, reverse iontophoresis, and related methods.^[14] Integrated microneedle arrays have demonstrated continuous multianalyte monitoring in research settings.^[15]

Yet translation is difficult. Sweat concentration depends on secretion rate, collection, contamination, temperature, skin location, and the uncertain relationship between sweat and systemic concentration. A major 2024 perspective concluded that the physiological relevance of many sweat analytes remains to be established outside narrow use cases.^[13] Therefore, Commonsent should assume that only a limited set of biochemical measures will become reliable enough for broad use within five years.

5.2 A plausible 2029–2031 capability set

The most powerful advance may be causal rather than sensory. By 2031 a user may have years of synchronized context, physiology, self-report, and intervention outcomes. With careful experimental design, the agent can estimate which changes reliably improve capacity for that person. This requires micro-randomization, natural experiments, interrupted time-series analysis, and explicit accounting for confounding. It also requires the humility to say when the evidence remains ambiguous.

5.3 The danger of biological overreach

More biomarkers can make a system less truthful if they encourage mechanistic stories unsupported by evidence. A cortisol-related signal, for example, may reflect circadian rhythm, exercise, anticipation, illness, or chronic strain. It does not independently reveal why someone feels threatened or whether a social situation is unjust. Commonsent must preserve the social and institutional model rather than reducing human experience to chemistry.

How each capability horizon improves individual well-being

6.1 Present: self-knowledge and cognitive protection

The present version can help people distinguish identity from temporary condition. A person who repeatedly experiences diminished confidence after poor sleep, rapid task switching, or unresolved conflict can learn that the state is real but not a stable measure of competence. That insight can reduce shame and learned helplessness while making interventions more concrete.

The system can also protect attention. When cognitive load rises, it can summarize, defer nonessential alerts, preserve task context, and ask one question at a time. It can recognize that an urgent message arrived during a moment of vulnerability and recommend later review. It can help schedule demanding work or civic participation during periods of higher capacity.

6.2 Two to three years: real-time context and better intervention fit

Eye- and context-aware systems can identify which specific interaction, visual task, or conversational pattern preceded overload. They can detect repeated rereading, rapidly shifting gaze, interruption density, or drowsiness and change the interface before failure occurs. AR can provide subtle support in the environment rather than requiring the user to stop, unlock a phone, and formulate a request.

Near-term systems can also distinguish risk from receptivity. When the user is distressed but unreceptive, the agent may suppress intervention, preserve a note, and revisit later. When the user is calm but approaching a known risk pattern, the agent can offer preventive support.

6.3 Three to five years: longitudinal self-science

Selective biochemical sensing and richer brain-body models could help explain why apparently similar days produce different outcomes. The user may discover that a difficult interaction is far more destabilizing under sleep debt, inflammation, medication changes, or metabolic volatility. Personal experiments can test whether a walk, meal timing, breathing practice, reduced notification load, conversation structure, or clinical intervention actually changes the outcome.

Well-being should be measured by improved functioning and agency rather than by the amount of data collected. The key outcomes are fewer avoidable overload episodes, better recovery, more accurate self-understanding, improved decision quality, increased perceived control, and greater ability to pursue chosen goals.

How each capability horizon improves family well-being

Families are regulatory systems. Members influence one another’s arousal, attention, safety, and recovery. Reviews of parent-child synchrony find behavioral and physiological coordination across development, with outcomes that depend strongly on context, relationship quality, stress, trauma, and the physiological system measured.^[19][20] Synchrony can support connection and regulation, but it can also represent co-dysregulation. A family layer must therefore help members coordinate capacity without ranking, diagnosing, or policing one another.

7.1 Present: timing, routines, and caregiver load

A present version can identify household patterns such as recurring conflict during rushed mornings, overloaded transitions after work, sleep-related irritability, or communication failures when several people need attention simultaneously. The intervention may be operational rather than psychological: move a task, reduce demands, create a quiet interval, simplify instructions, or delay a difficult conversation.

For caregivers, the system can recognize cumulative load across sleep, work, appointments, and repeated interruptions. It can help allocate tasks and signal that support is needed without exposing why. For children, sensing should be narrower, age-appropriate, and under strong protections. Manual check-ins and routine-level signals may be more appropriate than continuous psychological inference.

7.2 Two to three years: consent-based co-regulation

With better wearables and context models, family members could share a small set of voluntary states such as available, concentrating, needs quiet, open to talk, or caregiver support needed. The system could suggest when family members are most likely to communicate successfully, or identify that a conflict is escalating and offer a pause. It should never reveal raw physiology or label one member as “the problem.”

AR and audio interfaces may help children and adults who struggle with rapid verbal input by presenting one instruction at a time, preserving a visual task sequence, or reducing competing notifications. This can improve family functioning without framing ordinary differences in processing style as pathology.

7.3 Three to five years: household learning and preventive support

A mature family layer can identify combinations of factors that predict difficult periods: poor sleep across several members, schedule compression, illness, financial stress, extreme heat, or an unusually demanding school week. It can recommend preventive changes before conflict peaks. It can also evaluate whether the changes worked, allowing the household to build an evidence-based operating rhythm.

The family benefit depends on governance. A controlling partner could use state signals coercively. Parents could use them to demand compliance rather than support regulation. Employers or schools could pressure families to share data. Commonsent must therefore support asymmetric privacy, age-gated consent, emergency escape, hidden-safe modes for abuse risk, and strict limits on guardians’ access to older children’s private psychological content.

How each capability horizon improves social well-being

The collective value of internal-state data is not that institutions can inspect citizens. It is that citizens can produce protected evidence about the environments and systems that shape human capacity.

Social and economic conditions profoundly influence health and mental well-being. WHO emphasizes that access to power, money, resources, housing, education, work, and social protection contributes to avoidable health inequalities.^[2] A Commonsent internal-state layer could help communities observe the lived effects of those conditions, but only through one-way privacy-preserving aggregation.

8.1 Present: process burden and participation quality

Current devices can already help determine whether a civic process creates avoidable overload. A public meeting may require four hours of attention, rapid topic switching, dense documents, and late-night decisions. A benefits application may repeatedly trigger confusion and abandonment. A transit route may create chronic uncertainty and stress. A school communication system may deliver fragmented, contradictory messages to parents.

Commonsent could measure completion, interruption, self-reported control, time demand, and aggregated state changes. The result would not be “citizens were anxious.” It would be an actionable process diagnosis: where people disengaged, which information caused confusion, and what redesign improved comprehension.

8.2 Two to three years: environmental and institutional feedback

Context-aware glasses, phones, and wearables could help communities map how heat, noise, unsafe crossings, crowding, service delays, hostile interfaces, or workplace scheduling affect daily capacity. Aggregated patterns could support better public-space design, worker protections, school schedules, and resource allocation.

Group emotions and expressions influence group cognition and behavior through contagion, inference, and shared affective tone.^[21] Commonsent could adapt deliberative settings when aggregate overload or polarization rises: slow turn-taking, provide a factual reset, separate questions, or introduce a break. The system must never expose which participant is aroused or use group-state data to manipulate voting.

8.3 Three to five years: resilience and policy evaluation

Longitudinal aggregate data could reveal whether a policy improves or degrades lived capacity across neighborhoods and social groups. For example, did a transit redesign reduce commute volatility and evening household conflict? Did a benefits simplification reduce abandonment and overload? Did a heat mitigation program improve sleep and daytime functioning? Did a new work-scheduling rule reduce chronic physiological strain?

This creates a new class of civic metric: not happiness scoring, but **agency conditions**. The aim is to observe whether people have adequate recovery, comprehensible systems, perceived control, social safety, and time to participate. These indicators can complement economic output and service counts by measuring whether institutional arrangements support human functioning.

Distinguishing internal and external causes

The user’s original insight requires more than correlation. Commonsent should estimate whether a state change was driven mainly by internal vulnerability, an external exposure, or their interaction. That is feasible probabilistically, but only with personal baselines, event timing, lag models, self-report, and safe experimentation.

9.1 The attribution pipeline

Baseline. The model learns expected ranges by time, sleep, activity, location, medication, meals, and social context. A heart rate of 90 may be normal during a walk and unusual while sitting at a desk.

Transition detection. The system looks for coordinated change across channels: HRV, EDA, respiration, pupil, movement, voice, interaction behavior, and self-report.

Event alignment. It checks messages, meetings, locations, meals, task switches, noise, people, and internal-thought reports near the transition.

Lag modeling. Different mechanisms unfold at different speeds. Pupils and EDA change in seconds; glucose, cortisol-related processes, fatigue, and chronic stress operate on longer horizons.

Competing hypotheses. The model maintains several explanations with confidence ranges instead of selecting one story prematurely.

User confirmation. The user can identify an internal thought, pain, social threat, physical exertion, or other cause the sensors could not observe.

Micro-experiment. When safe, the agent tests whether a change in notification timing, break, food, hydration, movement, breathing, or interaction structure alters the outcome.

9.2 Example

Suppose a user opens an employment-related message at 10:14 a.m. The system observes a rapid EDA response, reduced HRV, increased pupil diameter after controlling for light, repeated rereading, and no physical exertion. The user confirms feeling threatened and overloaded. This supports a strong external trigger association.

However, the user also slept four hours, had elevated resting heart rate since waking, showed low activity, and reacted unusually strongly to neutral messages. Commonsent should explain: “The message appears to have triggered the immediate response, while reduced recovery likely increased susceptibility.” It should not claim that either the message or sleep alone caused the experience.

9.3 Why causal modeling matters for agency

Without attribution, self-tracking can become fatalistic. The user sees a stress score but does not know what to change. Causal hypotheses turn measurement into options. If state changes are consistently associated with a particular meeting structure, location, schedule, or media pattern, the person can redesign the environment, set boundaries, coordinate with others, or challenge the institution producing the burden.

Cognitive liberty, self-transparency, and prohibited uses

Internal-state technology can either expand autonomy or create a new infrastructure of behavioral control. UNESCO’s 2025 Recommendation on the Ethics of Neurotechnology explicitly addresses technologies that infer mental states and emphasizes autonomy, freedom of thought, mental privacy, proportionality, non-discrimination, human oversight, and prohibition of coercive social control or arbitrary mental-state surveillance.^[18]

10.1 Constitutional protections

10.2 Children and vulnerable users

Children should not be treated as smaller adult users. The system should default to minimal sensing, age-appropriate explanation, time-limited data, and gradual transfer of control as the child matures. Parents may need operational alerts for safety or care, but should not receive a continuous dashboard of inferred emotion, sexuality, beliefs, social loyalties, or private thought patterns.

Similarly, people in abusive relationships, precarious employment, immigration proceedings, or coercive institutions require safe modes that hide participation, suppress shared signals, and allow rapid deletion. The architecture should assume that someone with legitimate device access may still be an unsafe observer.

10.3 Clinical boundary

Commonsent should distinguish wellness guidance, research features, and regulated medical functions. Abnormal patterns may justify suggesting professional evaluation, but the platform should not diagnose from consumer sensors. Clinically validated modules should be separated, reviewed, and presented with their intended use and limitations.

Implementation roadmap

11.1 Phase 1: foundation, 0–12 months

Create the Internal State Data Constitution, define prohibited uses, establish local storage, and integrate phone and wearable data through user-authorized APIs. Implement short manual check-ins and a state-context timeline. The first model should identify change and association, not emotion or diagnosis. A small adult pilot should test burden, comprehension, perceived control, and privacy expectations.

11.2 Phase 2: personalized intervention, 12–24 months

Add multimodal fusion, change-point detection, causal hypotheses, and adaptive presentation. Use micro-randomized trials to test quiet mode, message simplification, task preservation, breaks, and timing suggestions. Create user-visible correction tools and independent scientific review.

11.3 Phase 3: relational layer, 18–36 months

Introduce voluntary family capacity signals, caregiver support modes, and age-gated child design. Test conflict de-escalation and shared routine planning. Conduct adversarial review for coercive partners, parental overreach, and household surveillance.

11.4 Phase 4: civic and public-health integration, 30–60 months

Deploy secure aggregation, process-burden metrics, community stress indicators, and policy-impact evaluation. Partner with public-health institutions, universities, and community organizations. Federated learning should improve models without collecting raw internal-state data centrally.

11.5 Recommended MVP

Evaluation, risks, and research agenda

12.1 Primary evaluation outcomes

The principal outcome should be increased agency, operationalized through perceived control, decision quality, successful completion of chosen tasks, ability to recover, willingness to participate, and reduced avoidable overload. Secondary outcomes include sleep regularity, self-efficacy, family conflict, caregiver burden, and public-process completion.

Model evaluation should emphasize within-person calibration, false alerts, stability across contexts, subgroup performance, and the frequency with which users correct the system. A model that appears accurate but causes anxiety, compulsive checking, or family conflict is not successful.

12.2 Failure modes

12.3 Research questions

State validity: Which combinations of sensors and self-report best predict personally meaningful states in daily life?

Causal attribution: How accurately can the system distinguish a trigger from a background vulnerability, and when should it abstain?

Intervention timing: Which forms of support help under which states, and which increase burden?

Family outcomes: Can capacity signaling reduce conflict and caregiver load without creating surveillance or coercion?

Collective outcomes: Can privacy-preserving aggregate signals improve public-service design and democratic participation?

Equity: Do expensive sensors create a two-tier system, and can Commonsent provide meaningful benefits from phones and manual input alone?

Governance: Which technical and institutional controls remain effective when commercial or governmental actors seek access?

Conclusion: from quantified self to protected agency

The Internal State and Agency Layer should not turn Commonsent into a system that knows the user better than the user knows themselves. It should help the user become the most authoritative interpreter of their own changing condition.

Present technology is sufficient to build a meaningful first version. Phones, watches, rings, optional EDA devices, and CGMs can observe behavior, cardiovascular dynamics, sleep, temperature, movement, environment, and metabolism. Brief self-report can supply meaning. Personal baselines and event alignment can reveal recurring relationships among body, mind, environment, and institutions.

Over the next two to three years, lighter eye-tracked glasses, ear-worn electrophysiology, better on-device models, and more capable context sensing can improve workload estimation and intervention timing. Over three to five years, selectively validated biochemical and brain-body sensing may deepen explanation for specific use cases. None of these capabilities should be treated as inevitable, universal, or automatically beneficial.

At the individual level, the layer can separate identity from temporary condition, protect attention, improve decision timing, and create personal evidence about recovery. At the family level, it can support co-regulation, caregiver planning, and better timing without exposing private psychological content. At the social level, it can reveal how services, environments, work structures, and civic processes affect human capacity while keeping individual states inaccessible.

The decisive issue is ownership. Psychological transparency can be emancipatory only when it flows inward to the person and outward only under their deliberate control. If it becomes a source of employer scoring, insurance pricing, political targeting, family coercion, or state surveillance, the same layer would reproduce the asymmetry Commonsent is meant to correct.

The architectural objective is not maximum legibility. It is maximum agency under conditions of minimum necessary observation.

With that constraint, the Internal State and Agency Layer becomes more than a health feature. It completes the Commonsent loop: the world is interpreted in relation to the person, the person understands their changing capacity, assistance adapts to that capacity, and communities gain protected evidence about the systems that strengthen or weaken human life.

References

1. World Health Organization. Mental health: strengthening our response (2025).
2. World Health Organization. World report on social determinants of health equity (2025).
3. Sanchez-Comas A, et al. Smart Devices and Wearable Technologies to Detect and Monitor Mental Health Conditions and Stress: A Systematic Review (2021).
4. Basnet N, Zahabi M. Real-time cognitive workload assessment using non-intrusive methods: a systematic review (2025).
5. Wilbanks BA, et al. Using Eye Tracking for Measuring Cognitive Workload During Clinical Simulations (2021).
6. Choi A, et al. Digital Phenotyping for Stress, Anxiety, and Mild Depression: Systematic Review (2024).
7. FDA. First over-the-counter continuous glucose monitor clearance (2024).
8. FDA. First over-the-counter continuous glucose monitor for children (2026).
9. American Heart Association. Cuffless Devices for the Measurement of Blood Pressure: Scientific Statement summary (2025).
10. Kaveh R, et al. The future of wearable EEG: a review of ear-EEG technology and its applications (2023).
11. Zanetti R, et al. EEG glasses for real-time brain electrical activity monitoring (2025).
12. Xu Y, et al. In-ear integrated sensor array for continuous monitoring of brain activity and lactate in sweat (2023).
13. Davis N, et al. The challenges and promise of sweat sensing (2024).
14. Wu Z, et al. Interstitial fluid-based wearable biosensors for minimally invasive healthcare (2024).
15. Tehrani F, et al. Integrated wearable microneedle array for continuous monitoring of multiple biomarkers (2022).
16. Qian T, et al. Microrandomized Trials: Developing Just-in-Time Adaptive Interventions for Better Public Health (2023).
17. Vinci C, et al. Proximal Effects of a Just-in-Time Adaptive Intervention With Wearable Sensors (2025).
18. UNESCO. Recommendation on the Ethics of Neurotechnology (2025).
19. Ravindran N, et al. Parent-Child Synchrony After Early Childhood: A Systematic Review (2022).
20. Davis M, et al. A systematic review of parent-child physiological synchrony (2018).
21. Van Kleef GA, et al. Emotional influence in groups: affect, cognition, and behavior (2017).
22. Apple. Apple Vision Pro hardware and eye-tracking overview.
23. Meta. Aria Gen 2 research glasses sensor overview.
24. Apple. Apple Watch health capabilities and current sensor-derived features.
25. Meta. Ray-Ban Meta smart glasses camera, microphone, and AI capabilities.
26. Jeffri NFS, Awang Rambli DR. Augmented reality systems, mental workload, and task performance: review (2021).
27. Guo R, et al. Development and application of emotion recognition technology: systematic review (2024).
28. Leaning IE, et al. Smartphone data to clinically relevant predictions in depression: systematic review (2024).
29. WHO. Stress: questions and answers (2026).
30. DePasquale CE. Caregiver-child physiological synchrony across systems: systematic review (2020).