Over the past several weeks, I mapped over 100 labs working on this question across affective computing, mobile sensing, behavioral science, sleep research, social dynamics, and human-computer interaction. Together, they form the scientific infrastructure behind any serious attempt to build coherence-based systems.

By coherence I mean internal alignment across physiology, attention, emotion, and behavior such that actions reflect intent rather than compulsion. Operationally, coherence shows up as reduced internal conflict and increased consistency between stated intent and observed behavior under stress. I offer this as an operational definition, not a philosophical claim. It can be measured indirectly, modeled probabilistically, and optimized only when a user chooses it as their objective.

How to Use This Map

If you are a founder: This map shows which wedges have worked and which remain research-grade. Historically, narrow constructs with validated ground truth have been easier to commercialize. General state inference has mostly stayed research-grade, except in constrained environments with strong context. Pick a tractable target, find a lab partner who has published validation studies, and resist the temptation to claim more than your interpretation layer can support.

If you are an investor: The six traditions cluster risk differently. Measurement companies face commoditization as sensors get cheaper. Interpretation companies face accuracy and liability questions. Influence companies face ethics and consent scrutiny. Diligence should focus on ground truth quality, personalization requirements, and whether the company’s claims exceed its validation.

If you are a researcher: The translation paths that worked share a pattern. Picard, Patel, and Pentland all founded companies around specific, defensible inference problems, not general platforms. Open-source tools dominate research infrastructure. Commercial success required proprietary interpretation or a novel sensing modality.

The Coherence Stack

These labs contribute to three layers. Each has a boundary test:

Measurement: Raw signals captured. PPG, accelerometer, audio, screen events, GPS, skin conductance. If you cannot name the sensor and what it directly captures, you are not in measurement. Sampling rate and preprocessing determine quality.

Interpretation: Probabilistic claims about latent state with explicit uncertainty. Interpretation means probabilistic inference plus context plus calibration. Causality is the hard mode. Good interpretation surfaces confidence, context features used, and known failure modes, not just a single score. Interpretation outputs should look like a forecast, not a verdict. If your claim would change with more context, you need interpretation. If you are asserting cause, you need experimental validation.

Influence: Intervention delivered. Reflection, suggestion, adaptation, automation. If your product changes behavior, you are in influence, even if you call it “insights.”

Validation Ladder

Claims in this space vary widely in rigor. A rough hierarchy:

1. Face validity and user self-report alignment

2. Correlation to a validated instrument

3. Longitudinal stability within a person

4. Generalization across cohorts and contexts

5. Intervention effect shown in trials or natural experiments

Most commercial products operate at levels one and two. Levels four and five are rare outside clinical settings.

Quick Index

• Affective computing: signals are rich, labels are messy, context needed

• Ubiquitous sensing: capture is feasible, meaning is not, cold start is brutal

• Behavior design: interventions work, ethics drift risk, requires upstream truth

• Sleep and psychophysiology: validated constructs, but wearables still error-prone

• Social dynamics: high value, high power asymmetry, governance required

• HCI: delivery layer, can mask weak inference, must show uncertainty

Affective Computing

What they measure: Facial expressions, vocal prosody, skin conductance, heart rate variability, and other physiological signals. A distinction matters here: emotion classification assigns discrete labels (anger, joy, fear), while affect dimensions model continuous variables (arousal, valence). Most commercial systems use dimensions because they are more robust, but marketing often implies discrete recognition.

What they proved: Affect leaks through the body. Rosalind Picard’s Affective Computing Group at MIT established that multimodal signals can detect arousal and valence under controlled conditions. Hatice Gunes’ AFAR Lab at Cambridge extended detection into naturalistic environments. The datasets they created became the training ground for machine learning models, though label quality remains a bottleneck. Self-report, observer ratings, and physiological proxies often disagree.

What got commercialized: The pattern was narrow targets with clear buyers. Affectiva focused on automotive safety (driver monitoring) and advertising (audience response), selling to automakers and agencies. Empatica focused on seizure detection, selling to clinicians and caregivers. Both wedges had ground truth: a driver looked away, a seizure occurred. Commercial affect succeeds when ground truth is external and observable, not internal and self-interpreted. General “emotion AI” without a specific use case stayed research-grade.

Stack position: Primarily Measurement, with partial Interpretation for specific constructs.

Unsolved constraint: The affective ambiguity problem. The same physiological signal means different things in different contexts. Without rich situational data, inference fails. Label disagreement compounds the problem. Affective models perform best when paired with context sensors and user baseline calibration.

Ubiquitous Sensing

What they measure: Behavioral and contextual signals captured passively through smartphones, wearables, and ambient sensors. Movement patterns, app usage, location, sleep proxies, social interaction frequency.

What they proved: Collection at scale is feasible. Shwetak Patel’s UbiComp Lab at University of Washington and Tanzeem Choudhury’s People-Aware Computing Lab at Cornell demonstrated that phones and wearables can continuously capture health-relevant signals. The StudentLife study showed correlations between passive data and mental health outcomes. Inference at scale is not solved. Correlations do not generalize reliably across individuals.

What got commercialized: Commercial wins came from single-variable inference with clear ROI. Patel’s exits (Zensi to Belkin, SNUPI to Sears, Senosis to Google) each solved one well-defined problem: energy disaggregation, leak detection, specific biomarker screening. Choudhury’s HealthRhythms focused on mental health monitoring for clinical populations, not general wellness. The constraint that made these companies viable was narrowness.

Stack position: Strong Measurement of behavioral and contextual signals. Context is where coherence companies should start, not emotion labels.

Unsolved constraint: The cold-start problem is severe. Personalization requires longitudinal data, which requires sustained engagement, which requires value delivery before calibration is complete. Successful products deliver immediate value from measurement alone, then unlock personalization as data accumulates.

Behavior Design

What they measure: These labs rarely generate new sensing modalities. They optimize interventions and measure behavioral outcomes: adherence, engagement, symptom reduction, habit formation.

What they proved: Behavior change follows predictable patterns. B.J. Fogg’s Behavior Design Lab at Stanford produced the Fogg Behavior Model. Fogg’s methods became foundational for consumer product growth. David Mohr’s Center for Behavioral Intervention Technologies at Northwestern developed IntelliCare, evidence-based apps for depression and anxiety with published clinical trials. Susan Michie’s UCL Centre created the Behavior Change Technique Taxonomy, classifying 93 intervention mechanisms.

What got commercialized: Validated intervention protocols licensed to digital health companies. Mohr’s IntelliCare became Adaptive Health. Kevin Volpp’s CHIBE at Penn spun out VAL Health for enterprise behavioral economics consulting. In both cases, the buyer was health systems or employers seeking evidence-based programs.

Stack position: Primarily Influence. These labs specialize in what to do once you know something.

Unsolved constraint: Behavior design assumes sensing and interpretation are solved upstream. The same mechanisms can be used for agency or addiction. Coherence products must pick a side. If incentives reward time spent rather than intent achieved, the product will drift toward compulsion.

Sleep and Psychophysiology

What they measure: Sleep architecture, heart rate variability, autonomic nervous system function, and emotion regulation through validated physiological protocols.

What they proved: Some internal states have clear biological signatures with established ground truth. Matthew Walker’s Center for Human Sleep Science at Berkeley works with polysomnography-validated constructs. James Gross’s Stanford Psychophysiology Laboratory developed the process model of emotion regulation with measurable physiological correlates. The HeartMath Research Center built an ecosystem around HRV biofeedback, though some claims in the literature remain disputed.

What got commercialized: Clinical-grade validation translated to consumer or enterprise products. Walker co-founded Somnee for sleep enhancement. Ki Chon’s lab spun out Mobile Sense Technologies for cardiac monitoring. Tractability was the constraint: sleep stages and arrhythmias have clear definitions. Mood and stress do not.

Stack position: Deep Measurement with validated Interpretation for specific clinical constructs.

Unsolved constraint: Success here depends on working with tractable targets. Consumer wearables often infer sleep stages with meaningful error relative to polysomnography. The construct is tractable, but the measurement pipeline still matters. For coherence systems, start with constructs that have clear physiological grounding and validated measurement protocols before attempting higher-order inference.

Social Dynamics

What they measure: Voice patterns, physical proximity, interaction frequency, turn-taking, and other signals of interpersonal behavior.

What they proved: Unconscious behavioral cues predict group outcomes. Sandy Pentland’s Human Dynamics Lab at MIT showed that tone of voice, movement patterns, and interaction dynamics predict team performance and negotiation outcomes. His “honest signals” framework became the theoretical basis for workplace sensing.

What got commercialized: Pentland’s lab produced three major companies: Cogito (voice coaching for contact centers, acquired by Verint), Humanyze (workplace analytics), and Ginger (on-demand mental health, merged with Headspace). The wedge in each case was organizational effectiveness, not individual monitoring. The buyer was enterprise HR or operations. Framing mattered: companies that positioned as “employee surveillance” struggled to scale and faced adoption barriers. Companies that positioned as “team effectiveness” or “customer experience” found traction.

Stack position: Measurement of Relationships domain with emerging Interpretation.

Unsolved constraint: Relationship sensing has asymmetric power risks. Consent is not a checkbox when employers are involved. Interpretation mistakes can become managerial weapons. This area will face strong norms even without legal regulation. Approaches that minimize raw data retention and keep inference on device will matter here. Enterprise products should assume adversarial use cases and design accordingly. Coherence companies working in the Relationships domain need governance architectures that prevent misuse by design, not policy.

Human-Computer Interaction

What they measure: User behavior, input patterns, gaze, gesture, and interaction quality.

What they proved: The interface between system and user can be radically expanded. Carnegie Mellon’s HCII produced multiple spinouts from Chris Harrison’s Future Interfaces Group. Ehsan Hoque’s Rochester HCI Lab built communication coaching systems that analyze speech in real time.

What got commercialized: Novel input modalities or real-time feedback loops. Harrison’s Qeexo (acquired by TDK) focused on touch intelligence for device manufacturers. Hoque’s Yoodli focused on speech coaching for professionals. The buyer was enterprises seeking training tools or device OEMs seeking differentiation.

Stack position: Primarily Influence. These labs excel at the last mile.

Unsolved constraint: HCI labs depend on other layers to provide the signal. A polished UX can make weak interpretation feel persuasive, which is a coherence risk. A good coherence UX makes uncertainty legible rather than hiding it. Show confidence bands, show what data was used, show what would change the recommendation.

Why Interpretation Is the Bottleneck

Across all six traditions, the same asymmetry appears. Measurement keeps getting cheaper, smaller, and more continuous. Interpretation remains brittle.

Context-dependence. The same signal means different things depending on situation, time of day, and recent history. Elevated heart rate during a meeting could indicate engagement, anxiety, or recent stair climbing. Without rich context, inference fails.

Baseline variance. People differ widely in their physiological and behavioral signatures. Population-level models mislead individuals. Personalization requires longitudinal data, which requires sustained engagement, which requires value delivery before calibration is complete.

Causality. Correlation masquerades as insight. Observing that patterns correlate with outcomes does not establish what causes what. Causal claims require experimental designs that most sensing systems cannot support.

What the Landscape Reveals

Commercial successes cluster where the target is narrow and the ground truth is clearer: seizure detection, arrhythmia screening, sleep staging, communication coaching. Full-stack coherence systems that close the loop from sensing to inference to action remain rare, because interpretation is where uncertainty lives.

The white spaces are now visible:

1. Interpretation layer companies that can reliably contextualize physiological and behavioral signals

2. Longitudinal personalization systems that improve with individual data over time

3. Integration plays that connect measurement across traditions without collapsing into surveillance

The academic infrastructure exists. We have abundant sensors and cheap continuous capture. What we do not have is reliable meaning at the individual level.

If You Are Building in This Space

• Pick a tractable construct with ground truth. External and observable beats internal and self-interpreted. Seizure, arrhythmia, sleep stage, communication pattern. Not mood, not wellness, not coherence itself until you have validation and calibration.

• Design for calibration and cold start from day one. Your product must deliver value before personalization kicks in. Then it must improve as individual data accumulates. If you cannot explain both phases, you do not have a product.

• Make uncertainty and consent visible in the product. If your interface hides confidence levels, you are building score theater. If your consent flow is a checkbox, you will lose trust when it matters. Governance is a feature, not a constraint.

Scope and Limits

This map is not exhaustive. It reflects about 100 labs selected for relevance to coherence-based technology, not a census of all research in these fields. The six traditions are a lens for organizing the landscape, not a taxonomy of truth. Some labs span multiple traditions. Some important work does not fit cleanly.

The focus here is commercialization paths, not scientific contribution. Labs that have stayed academic may be doing work that matters more in the long run. Commercial success is one signal of feasibility, not a measure of importance.

I will update this map as I find omissions, new labs, and better categorizations.

Most confusion in happens when those layers blend together. A company says it “measures stress” when it actually measures heart rate variability and infers stress. A product claims to “optimize wellbeing” without specifying who decided what wellbeing means. A research paper treats a model output as ground truth.

This post introduces the taxonomy I’ll use throughout this research project. It has two parts: the Coherence Stack (how a system works) and the Application Domains (where it applies). Together, they let me map any company, paper, or idea into a consistent structure.

Part 1: The Coherence Stack

To support coherence as choice architecture, a system needs a full loop: Measure → Interpret → Influence. The discipline is simple: if you cannot answer “what was directly measured?” in one sentence, you are already in inference.

Layer 1: Measure

Measurement is the sensor layer. It produces observable data. It does not produce meaning.

• Physiological signals: Heart rate, heart rate variability, respiration, temperature, sleep stage estimates, electrodermal activity. These reflect autonomic dynamics and load—but they are context-dependent and confounded. They are proxies, not direct reads of emotion or meaning.

• Behavioral signals: Movement patterns, interaction tempo, approach and avoidance, dwell time, return frequency. Behavior often expresses internal state, but it can also be strategic, culturally shaped, or constrained by circumstance.

• Affective signals: Voice tone and pacing, language patterns, stress signatures in speech, facial expression where appropriate. These can leak internal state, but they vary widely by person and setting and are easy to misread without calibration.

• Context and environment signals: Time, location, calendar context, ambient noise and light, temperature, air quality, device usage. Context doesn’t explain meaning, but it shapes how the same signal should be interpreted.

One note: longitudinal evaluation is not a fifth modality. It’s how you learn whether any of this matters. Without it, the stack becomes a dashboard that feels insightful but cannot demonstrate downstream impact.

Layer 2: Interpret

Interpretation maps measured proxies plus context into probabilistic hypotheses about internal state. This is where most overreach happens. The layer must earn trust, which means it cannot pretend to be certain.

• Uncertainty management: Confidence is not a footnote. It is part of the product. Interpretation should make it clear when multiple readings fit the data.

• Contextual integration: The same physiological signal can correspond to different states depending on sleep, exertion, caffeine, social setting, and temperature. Good interpretation controls for context rather than ignoring it.

• Subjective calibration: People are not averages. The system must learn a user’s personal baseline and patterns over time. Population models can mislead individuals.

• Causal attribution: This is the bottleneck. Correlation is cheap. Causation is hard. Strong attribution usually requires experimental structure—even lightweight personal experiments. Claims that skip this step should be treated with suspicion.

I will treat interpretation claims carefully throughout this research. If a product claims it can read your emotions from a single signal with high accuracy in the wild, the burden of proof is on them.

Layer 3: Influence

Influence is where the system changes something. Highest leverage, highest risk.

• Reflection: Mirroring patterns back to the user so they can notice what was previously invisible. Descriptive, not prescriptive.

• Suggestion: Nudges the user can accept or ignore. “Consider a walk.” “You might delay that meeting.” The user retains choice.

• Adaptation: Automation—where the system changes the environment on the user’s behalf, within explicit permissions. Adaptive lighting, automatic focus modes, schedule adjustments.

This is also where governance matters most. Influence can support agency or quietly replace it. The line between coaching and manipulation is thinner than people like to admit.

Your job: Ensure influence without explicit user-stated goals is identified as a red flag. If the system is optimizing for something, the user should know what it is.

Part 2: Application Domains

The same stack applies differently depending on context. Incentives, consent models, and failure modes shift by domain.

1. Self: Personal energy, focus, emotional regulation, health. Today these tools are fragmented—sleep in one app, calendar in another, exercise in a third. The opportunity is integration. The risk is obsession, anxiety, and false certainty.

2. Relationships: Teams, communities, partnerships. This domain is powerful and ethically radioactive. The only viable version is consent-based and user-controlled. If it becomes a tool for employers to score individuals, it will fail on trust, adoption, and regulation—regardless of technical merit.

3. Things: Environments, experiences, consumption choices. Most recommendation systems optimize for what similar people clicked. A coherence-aware system would optimize for what actually works in your life. The product value could be enormous, but so is the data intimacy required.

Quick Reference

• Measurement: What happened (signals, timestamps, observable data).

• Interpretation: What we think it means (with uncertainty).

• Influence: What we do about it (with consent).

• Values: What “better” means (chosen by the user, not assumed).

That is not inherently evil. It is a sane strategy when the only thing you can reliably observe is external behavior. But it also produces a predictable outcome: products compete to capture attention and shape action, not to improve the quality of a person’s inner life over time.

AI makes a different move possible. For the first time, systems can start incorporating internal context—signals that reflect how an experience lands in your nervous system, your attention, your emotional dynamics. If we can sense internal context, we can build choice architecture oriented around coherence rather than conversion.

This essay opens a research project. Not a manifesto. Not an investment pitch. A genuine attempt to figure out whether this idea has substance.

What I Mean by Coherence

Coherence is a working construct, not a single settled variable. In plain language, it points to patterns of internal organization that make a person more stable, less fragmented, and more able to act intentionally. It is something you can sometimes feel and sometimes measure indirectly. It becomes a goal only when someone explicitly chooses it.

To keep this practical, I separate coherence into layers that can be studied before we pretend we know what we’re doing:

• Physiological regulation: Autonomic stability, recovery capacity, resilience to stressors.

• Cognitive coherence: Attention stability, reduced fragmentation, the ability to sustain focus without compulsive switching.

• Affective coherence: Emotion dynamics like recovery speed and flexibility, not the absence of negative emotion.

• Values coherence: Behavior matches stated aims; tradeoffs are chosen rather than drifted into.

A boundary worth stating early: coherence is not moral superiority, perpetual calm, productivity at all costs, or compliance with an external agenda.

The Basic Hypothesis

Today, most choice architecture optimizes external stimuli to drive external behavior. AI allows the loop to include internal context. Not mind reading. Something more modest and more testable:

If we can measure proxies of internal state over time, then we can build tools that help people notice patterns, form better hypotheses about what supports them, and make better choices.

Why This Might Be a Real Category

Three forces are converging.

First, measurement is no longer scarce. Wearables and phones produce continuous streams of physiological, behavioral, and context data. Apple, Oura, Whoop, Garmin, and others have made sensing normal. The bottleneck is not inputs. The bottleneck is interpretation and responsible influence.

Second, foundation models are finally capable of integrating messy context. Humans live in noisy, confounded reality. We drink coffee, sleep badly, travel, argue, exercise—then try to infer what any one signal “means.” That is exactly the kind of ambiguity that modern AI is built to handle, at least probabilistically.

Third, the costs of the attention economy are becoming visible. People feel scattered. Teams feel brittle. Culture feels reactive. Whether technology can help without becoming surveillance or manipulation is an open question, but the demand signal is clear.

So the project is not “coherence is the future.” The project is: does a coherence-oriented architecture produce products that people want, trust, and pay for while staying inside acceptable ethical constraints?

Why I’m Doing This

I have spent most of my career watching how narratives organize people in startups, in families, in institutions. I’ve built a company around trust and data. I’ve worked with hundreds of founders. I’ve written about meaning-making and consciousness for years.

The coherence economy, if it exists, lives where these threads converge.

What Comes Next

Over the coming months, I’ll be mapping the landscape of companies and labs working in this space. I’ll be studying what’s technically feasible in measurement, interpretation, and influence. I’ll be building a conversation with researchers, founders, investors and people who share a curiosity for this space.