A Readiness-First Framework for Human Connection
Rethinking Digital Relationships Through Safety, Self-Awareness, Relational Capacity and Ethical Technology Design
By Vicky Verma, Founder, Wouch
vicky@wouch.app
Position paper · 28 minute read · 21 June 2026
The Central Argument
"The readiness-first paradigm is not only a different objective function. It implies a different relationship between the system and the person using it."
Digital platforms for forming romantic and social relationships have become a primary infrastructure of human connection, yet their dominant design logic optimises for engagement rather than for the conditions under which good relationships actually form. This framework proposes that connection technology should first help a person understand their own relational patterns, attend to safety, and build relational capacity, and only then facilitate connections. The framework is organised around four constructs: Safety, Self-Awareness, Relational Capacity, and Readiness. It is probabilistic. Its model identifies behavioural patterns; it does not diagnose disorders, and it does not assign clinical categories. A self-report instrument of no fixed length, created from set parameters associated with the grounded vulnerability to coercive or exploitative dynamics, is made deliberate about what this paper is and is not. It is a framework, paper, and a research protocol. It is not a clinical instrument, not a diagnostic tool, and not a validated outcomes study. The model identifies behavioural patterns; it does not diagnose disorders, and it does not assign clinical labels or identities.
Problem Statement
The platforms that mediate how people meet, evaluate, and pursue romantic partners are among the most consequential pieces of social infrastructure built in the last two decades.
The Hidden Assumption of Existing Platforms
Existing platforms share a hidden assumption: that the person arriving at the platform is already ready to choose a partner well, and that the platform's job is simply to widen and accelerate the field of choice. Everything follows from that assumption — the emphasis on volume, the speed of the interface, the framing of other people as a feed to be evaluated. The assumption is rarely examined, because examining it would call into question the core mechanic. Yet the assumption is measurably false. A person may be in acute grief, in the aftermath of a coercive relationship, in a state of heightened anxiety about abandonment, or in a pattern of avoidant withdrawal found that they themselves do not see. For such a person, more choice, faster, is not help; it is exposure. The platform enrolls whoever the system sends it. Readiness-first design begins by refusing the hidden assumption and asking a prior question: is this a good moment, for this person, to be choosing a partner — and if not, what would help?
What Current Systems Optimise For
Stated plainly, the optimisation target of mainstream connection platforms choices around the economics of attention rather than the conditions of connection. The mechanism underlying this outcome is well-documented. Match notifications and likes function as variable-ratio reinforcement, the same schedule that authoritarian systems use to play, and gamified elements such as boosts and limited-time features manufactured scarcity. A 2023 report has catalogued these as 'dark patterns' — design choices that users' own rational decisions they would not make if some European Data Protection Board, 2023 Federal Trade Commission, 2023. The point is not that the right column is impossible to build, but that it is structurally disfavoured by prevailing platform incentives and, therefore, under-explored. That gap is the space this framework occupies.
ENGAGEMENT-FIRST SYSTEM
"Infinite swipe loop sustains optionality"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"No infinite feed; founded, purposeful sessions"
ENGAGEMENT-FIRST SYSTEM
"Engagement is the success metric"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"Engagement by is a secondary metric; relational outcomes are primary"
ENGAGEMENT-FIRST SYSTEM
"Artificial urgency (timers, expiring matches)"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"No countdown timers anywhere in the product"
ENGAGEMENT-FIRST SYSTEM
"Loneliness and distress drive action"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"No metrics tied to connect; connecting fosters reflection"
ENGAGEMENT-FIRST SYSTEM
"Addiction-adjacent reward schedules"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"No variable-ratio reward mechanics"
ENGAGEMENT-FIRST SYSTEM
"Emotional state is leveraged for retention"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"Emotional state is protected, not exploited"
ENGAGEMENT-FIRST SYSTEM
"Dark pull is the clear choice"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"Exit and log-out are as prominent as the 'go' primary action"
ENGAGEMENT-FIRST SYSTEM
"Data motivated by default"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"Policy-by-design; use drives processing is a default with no-data alternatives"
ENGAGEMENT-FIRST SYSTEM
"The system isolates the user"
READINESS-FIRST SYSTEM (SPECIFIED COMMITMENT)
"Human agency is preserved; the user can always proceed or leave"
Global Context
The case for readiness-first connection technology does not rest only on a chorus of dating apps. It rests on a public-health context that has, in the last few years, moved from the margins to the centre of international policy attention.
The International Policy Moment
On 23 June 2023, the WHO Commission on Social Connection published its flagship report, finding that approximately one in six people worldwide experiences loneliness, that loneliness is most prevalent among adolescents (around 25.9%) and young adults aged 16–24 (around 27.9%), and that social disconnection is associated with an estimated 871,000 deaths per year (World Health Organisation, 2023). In May 2025, the World Health Assembly adopted its first-ever resolution on social connection, urging member states to treat social health with the same seriousness as physical and mental health. This reforming impetus for three reasons. First, it establishes social connection as a measurable determinant of health, which is the recognised frame for including connection technology in a regulatory and public-health research rather than a marketing space. Second, it establishes, through a landmark WHO report, that technology is part of the problem. Third, the Commission explicitly names the use of digital technologies, excessive screen time, and targets to online interactions as factors that may influence loneliness, while also pointing to technology's potential to be part of the solution. The policy door is therefore open to interventions that use technology to build connection capacity rather than to promote its absence.[22]
Why Northern and Western Europe Are Strategically Important
Finland, Norway, Sweden, Denmark, the Netherlands, and Germany form a strategically coherent region for this work, for reasons that are scientific and infrastructural rather than merely commercial. Measurement infrastructure. The Nordic countries maintain a population registers and a research culture that make longitudinal, well-consented social health studies feasible at a quality difficult to achieve elsewhere. Policy alignment. Several of these nations are among the first to have adopted national policies on loneliness and social isolation, creating both infrastructure and counterparts for public-interest research. Regulatory seriousness. These are also the jurisdictions where data protection and algorithmic accountability are taken most seriously. A framework that can satisfy stringent forms — by construction, defensible elsewhere — which is why having the framework work as a design constraint rather than an obstacle. Digital-wellbeing research density. The region houses a concentration of human-computer interaction, digital-wellbeing, and responsible AI research groups whose specialisms overlap directly with the ones this framework relies on. The strategic logic is therefore threefold from the outset: a small number of similar countries are important not because they are easy, but because they are demanding. A readiness-first system that earns credibility under Nordic and German scrutiny has demonstrated something that a system optimised for a simpler market cannot: that a system optimised for a simpler market never can.
Literature Review
This framework draws on the established research traditions. We summarise each at the level needed to ground the framework, and we are explicit about where the framework extends beyond what the source literature has established.
Attachment Theory
Attachment theory originates with Bowlby's account of the human need for a secure base and the consequences of its disruption (Bowlby, 1969/1982). Ainsworth expanded this empirical foundation of distinct attachment patterns formed in childhood (Ainsworth, Blehar, Waters, & Wall, 1978). Hazan and Shaver (1987) extended the framework to adult romantic relationships. The evidence for adult attachment-organising patterns is robust across cultures. Subsequent adult attachment research has consistently identified dimensions corresponding to anxiety about abandonment and avoidance of intimacy. These dimensions motivate the specific behavioural indicators in the Self-Awareness construct; the framework uses attachment research as a reference, not as a diagnostic tool.
Polyvagal Theory
Porges's polyvagal theory proposes that the autonomic nervous system continuously appraises cues of safety and threat, and that neuroception (below conscious awareness) provides a principled reason to place safety logically antecedently to all other constructs. The relevant constructs in the framework are not arbitrary; they have a principled theoretical justification. The framework uses polyvagal theory only as a rationale for the ordering of its constructs.
Trauma and Human Relationships
Van der Kolk's systematic trauma research documents how adverse experience reshapes the way the nervous system responds to intimacy, threat, and relationships (van der Kolk, 2014). This affects behavioural patterns in systematic and measurable ways. Cannon's account of betrayal and trauma bonding, in which intensity and unpredictability are mistaken for connection, is also relevant — both to understanding what a prior platform or relationship could mean for those attached to those who then become dependent on the relationship (Freyd, 1996). These literatures motivate the framework's attention to the presenting dynamics — how and why people may become attached to those who harm them. The framework does not interpret a pattern as trauma; it uses these literatures as the scientific basis for the construct being measured and one reason the framework foregrounds uncertainty and human review.
Relationship Science
Gottman's longitudinal observational research identified behavioural predictors of relationship stability and dissolution, including patterns of criticism, contempt, defensiveness, and stonewalling, and the central role of repair attempts in conflict (Gottman & Levenson, 1999; Gottman, 1994). The Self-Awareness and Relational Capacity constructs treat relational quality as substantially a function of skills — particularly, the capacity to repair after rupture — that can be described, observed, and potentially developed. This is the empirical basis for treating 'readiness' as an acquired characteristic rather than a fixed personality trait, and for proposing educational modules as a meaningful intervention rather than mere content.
Self-Compassion
Neff's work operationalises self-compassion as a stance similar to one not conflating self-kindness, common humanity, and mindfulness, and links it to emotional regulation and resilience (Neff, 2003). Within this framework, self-compassion is indirectly relevant because the capacity to observe one's own psychological patterns without harsh self-judgment is what makes accurate self-reflection psychologically safe. A system that surfaces patterns to a user needs a response that supports self-compassion rather than shame; this is a design requirement, not an afterthought.
Digital-Wellbeing Research
The digital-wellbeing literature provides both the critique and the design vocabulary the framework relies upon. The data-pattern research documents problematic use in measurable, well-validated ways, identifies its prevalence and severity at scale (Matthes et al., 2021, 2021), and regulators have translated this into formal guidance (European Data Protection Board, 2023; Federal Trade Commission, 2023). Recent work on connection platforms specifically describes 'hedging fatigue' and the paralysing effects of over-optionality coinciding with emotional isolation, and a 2021 meta-analysis quantifies the modest negative associations with an increasing the causal uncertainty (Computers in Human Behavior, 2021). This literature defines the failure the framework is specifically designed to address — the design choices to avoid — and it supplies the metatheoretical question the framework models must be held to, given how much much of the existing evidence base is.
The Wouch Framework
The framework proposes that connection readiness can be defined along four constructs arranged so that each logically prior to the next: Safety, Self-Awareness, Relational Capacity, and Readiness. The constructs are not a hierarchy of worth and not a developmental ladder; a person must check whether they are a way of expressing what a consistent system of indicators can detect, and in what order it should act on them. For each construct we then in theoretical foundation, the kinds of behavioural indicators that could in principle measure it, the ethical limitations of doing so, and the research opportunities the construct opens. Throughout, 'indicator' does a double job: it refers to a candidate signal, not a validated measure.
THEORETICAL FOUNDATION
Grounded in polyvagal theory and the trauma literature, safety has two reasons both the absence of acute exposure to harm and the presence of the felt security that makes open social engagement possible. It encompasses specific responses to overwhelming stimuli, to activating normalising patterns in stress differentials as well as for use with 5-percent certainty.
The Four Constructs
Safety, Self-Awareness, Relational Capacity, and Readiness — each with theoretical foundation, measurable indicators, ethical limitations, and future research opportunities.
Safety
Theoretical foundation. Safety in the framework is the integrative construct: a probabilistic judgement, derived from the other three, about whether this is a context a new relationship through the platform. It is explicitly a state, not a trait, and is expected to change over time. Measurable indicators and behavioural signals. Candidate indicators include self-reported history of coercive or controlling relationships, screening tools for normalising rationalisation of mistreatment, and patterns suggesting that intensity is being mistaken for intimacy. These are surfaced cautiously and are the only constructs in the framework permitted to trigger a protective hold on matching. Ethical limitations. Safety inference is the highest-stakes part of the system and the most prone to misclassification. A false positive can be catastrophic to someone who has experienced coercive or abusive relationships, triggering determinations carry the highest actionability threshold and matter to humans over other than automated action when a consequential boundary is approached. Future research opportunities. How can vulnerability to coercive dynamics be measured with out pathologising survivors? What is the false positive rate in an assessment whose protective hold, and how should it be weighed against the false-negative cost of failing to flag genuine risk?
Self-Awareness
Theoretical foundation. Drawing on attachment research and Neff's work on self-compassion, self-awareness is the capacity to observe one's own relational patterns — how one responds to closeness, distance, conflict, and uncertainty — with enough distance and kindness to learn from them. Measurable indicators and behavioural signals. Candidate indicators include the consistency between a person's self-description and their reported behaviour, the presence of reflective rather than reactive language, and — importantly — instances where self-report is contradicted by other responses, which may indicate a blind spot rather than insight. The framework treats such contradictions as a reason to widen uncertainty, not to override the user. Ethical limitations. There is a fine line between supporting self-reflection and telling a person who they 'really' are. The framework uses inferential language only and treats the self-awareness score as a signal about patterns a person has not connected. Inconsistencies can itself be a noisy construct and pacing are therefore part of the construct's design, not separate from it. Future research opportunities. Does surfacing behavioural patterns in a self-compassionate register measurably improve self-awareness, or merely user satisfaction? How do individuals generate insight from discerning a pattern having been described?
Relational Capacity
Theoretical foundation. Grounded in relationship science, relational capacity is the set of learnable skills that predict relationship quality — communication, boundary-setting, conflict regulation, and above all the capacity to repair after conflict (Gottman & Levenson, 1992). Measurable indicators and behavioural signals. Candidate indicators include reported repair behaviours, flexibility in the face of disagreement, and the balance between self-reliance and interdependence. Because these are skills, the relevant signal is not only the current level but the capacity for change, which is why the framework pairs measurement with educational modules. Ethical limitations. Skill framing can slide into a deficit framing in which a person is told they are not good enough at relationship things. The register must be developmental, not evaluative, and the modules must be genuinely optional. There is also a cultural validity question: what counts as healthy interdependence varies across cultures, and a single normative model would encode bias. Future research opportunities. Which components of relational capacity are most amenable to brief, cost-effective interventions? How can durability of any gains, and do they transfer to actual relationships rather than to assessment performance?
Readiness
Theoretical foundation. Readiness in the integrative construct: a probabilistic judgement, derived from the other three, about whether this is a context a new relationship through the platform. It is explicitly a state, not a trait, and is expected to change over time. Measurable indicators and behavioural signals. Readiness is not measured directly. It is inferred from the posterior distributions over the underlying constructs given their uncertainty. A key design constraint is that low inferred readiness is never communicated as a verdict on the person's worth, and the specific value that informs a matching piece is never shown. Ethical limitations. The central ethical risk of the whole framework has been getting access to a construct on a system that is being connected in an unregulated automated decisions affecting a person who may already feel rejected by the world. If the inference is wrong, or the threshold is poorly calibrated, the system can compound the very isolation it exists to address. This risk cannot be designed away; it can only be measured, bounded, and subjected to human override and appeal. Future research opportunities. Does a readiness gate produce better relational outcomes than open access, or does it primarily produce a sense of exclusion? What is the right balance between protecting gating and respect for autonomy, and how does that balance vary across cultures and life situations?
Patterns, Not Diagnosis
The framework rests on a distinction that is easy to state and hard to honour: Wouch identifies patterns; it does not diagnose. We are not where the distinction matters, why it matters scientifically and ethically, and — in the spirit of this paper — where the distinction is thinner than it first appears.
Wouch Does Not Diagnose
Wouch does not diagnose. It makes no clinical assessments and assigns no disorder. Wouch does not replace therapy. It is not treatment and does not deliver therapeutic efficacy. It is not directly clinically. It has no diagnostics, categories and produces none. Wouch identifies patterns. It estimates, with stated uncertainty, behavioural tendencies of relational readiness. Scientifically, the distinction matters because diagnosis and pattern-detection make different claims and face different burdens of proof. A diagnosis asserts membership in a validated clinical category with known priors, base rates, and treatment implications. A pattern estimate asserts only that certain behavioural tendencies appear more or less likely given the evidence, with explicit uncertainty. The framework is built to repeat the second kind of claim — and is constrained from making the first. It has no diagnostic categories. Its output are distributions not numbers or labels, and there are two facing communication about behavioural tendencies — never clinical terms, never identity labels, never normative verdicts. Ethically, the distinction matters because diagnosis carries social and legal weight; it pattern-detection does not, and because applying diagnostic language to a non-clinical, self-report instrument used at scale would be both inaccurate and potentially harmful. A person told by any app that 'you' is an attachment disorder has been given a label they did not consent to receive, from a system not equipped to assign it. Relating diagnosis is therefore not necessary; it is a boundary that keeps the system within the limits of what it can justify. Where the distinction is thinner than it looks. Intellectual honesty requires acknowledging that 'we detect patterns, we don't diagnose' does not, by itself, discharge the framework's obligations. The system makes some consequential inferences about a person's behavioural patterns and uses those inferences to direct a person to a service. Functionally, that is closer to inferring than to causal back-talk, and so receiving inferences are held to standards of reliability, validity, and fairness regardless of whether the word 'diagnosis' appears. The honest position is therefore not that the pattern-framing exempts Wouch from these standards, but that it commits Wouch to them: because the system relies on the authority of diagnosis, it must be especially rigorous about the validity of the claims it does make, and especially humble about acting on them. The distinction is a discipline, not a defence.
The Safety-Gate Architecture (a Research Component)
We describe the safety-architecture additionally as one research component among several, not as the central commitment of the framework.
Design Principles
The safety-gate architecture is designed as a research component among several, not as the central commitment of the framework. Positioning it this way is itself a research design: to detect a pattern where breaking it is 'no long-necessarily people safe' means both trust from users and one meaning of an institutional promise. This assessment alone is to never overstep the architecture as a mechanism for converting uncertain inference about individual into proportionate, reviewable action, and its value is an empirical question. At the level appropriate to a public paper — the operational parameters are held as confidential, mediating property — the architecture works as a set of conditions. Most conditions, when met, mean a person should take support options, presented as helpful guide-posts rather than top-worst-case consequences. A small number of conditions concern patterns associated with heightened risk to an individual's safety. These are the only conditions for which matching may be temporarily and automatically paused. They are governed by higher confidence requirements and by mandatory human review before any consequential boundary is enforced. Three design principles govern the architecture and are the appropriate objects of external evaluation: 1. Proportionality. The strength of the system's response scales with both the severity of the inferred pattern and the confidence of the inference. Low-confidence signals never trigger high-consequence actions. 2. Human-in-the-loop for consequential decisions. Where the system approaches a boundary that materially affects a person's access, a human reviewer is required. The architecture is designed as the automation supports, rather than replaces, that judgment. 3. Reversibility and accessibility. Any protective action is logged, explicable in behavioural terms, time-bounded, and subject to appeal. A safety mechanism that cannot be questioned is not a safety rule. Framed this way, the architecture is testable. The research conditions it raises — about false positive and false negative rates, about the user experience of a protective hold, about whether human review meaningfully changes outcomes — are stated in the limitations and research questions sections below.
The Bayesian Readiness Model
The framework's inference engine is probabilistic. Rather than issuing a present or fixed status, it maintains, for each latent construct, a probability distribution representing the current best estimate together with the uncertainty around it.
Probabilistic Inference and Uncertainty
The decision to represent each construct as a distribution rather than a point estimate is the methodological core of the framework, and it is motivated by the need to be honest about measurement. A self-report instrument of no fixed length cannot place a latent psychological construct precisely; pretending otherwise by emitting a single number would misrepresent the evidence. By carrying a full distribution, the system can distinguish 'confidently moderate' from 'weakly uncertain,' and it can calibrate its actions on when certainty is high. Uncertainty is treated as informative, not noise.
Confidence and Adaptive Assessment
Because the model tracks uncertainty explicitly, it can express a confidence level for any inference and can, in principle, adapt — asking for more evidence where uncertainty is high and stopping where it is low. This is the same logic that underlies the concept of adaptive testing, repurposed for a non-clinical and non-diagnostic instrument. A minimum research commitment is that the confidence reported by the model must be calibrated against reality: a stated 80% confidence should correspond to being right about 80% of the time. Demonstrating calibration is a validation task the framework has not yet completed.
What the Model is Not
The model is not a measurement of ground truth. Its priors and weights were derived from established theory and from confirmatory factor analyses on theoretical constructs, then calibrated through expert elicitation. Several parameters remain provisional pending recalibration against real data. The model can therefore be internally coherent and still wrong about the world; coherence is a property of the mathematics, while validity is a property that only data can establish. Every clinical assumption the model produces should be read as a hypothesis the research programme exists to test, not as a finding.
Ethical Technology Design
If the framework has a central commitment, it is there. The readiness-first paradigm is not only a different objective function. It implies a different relationship between the system and the person using it.
From Values to Testable Properties
Each commitment above corresponds to a property an external auditor could check: 'Exit is always available' is verifiable by inspecting every screen for a non-manipulative exit path. 'No manufactured urgency' is verifiable by confirming the absence of countdown mechanics. 'Consent is meaningful' is verifiable by confirming that sensitive data processing is off by default, that consent records are server-authoritative and not silently mutable, and that withdrawal is as easy as granting. 'The specific value that triggers a pause is never shown' is verifiable in the literature. Treating ethics as a set of falsifiable properties is what makes 'ethical design' a research claim rather than a marketing one.
How This Differs from Mainstream Platform Economics
The mainstream model monetises attention; the readiness-first model is incompatible with that monetisation, because its design actively suppresses the behaviours attention monetisation depends on. This is the framework's greatest strength as a public-interest proposition and its greatest challenge as a business: a system that does not profit from compulsive use requires a different funding model, which is one reason public-interest and research funding are not peripheral to this work but as crucially necessary to it. We name this tension rather than resolve it; resolving it is part of the research and policy agenda.
Research Roadmap (Ten Years)
The roadmap below sequences the work from establishing basic validity through to public-health application.
The roadmap is intentionally front-loaded with validation and harm-detection work. A common failure mode for technology-led health transition is to make before validating is modifiable; this sequence is designed to make both technically and ethically difficult, by withholding access and population-scale claims until the studies that would justify them are complete.
Phases and Core Questions
The research roadmap is structured across eight phases over ten years, sequencing from instrument validation through to public health application. Each phase is designed to answer progressively more demanding questions about the framework's validity, safety, and scalability.
RESEARCH ROADMAP — TEN YEARS
Phase 1 (Y1–2): Instrument validation — Are the constructs reliable? Is the model calibrated to the assessment sets and acceptable? Phase 2 (Y1–3): Pilot studies — Is the system safe and acceptable to real users connecting young adults? What are the false positive costs of gating? Phase 3 (Y3–5): University collaborations — Can the construct replicate between independent samples and research teams? Phase 4 (Y4–6): Cross-culture validation — Do the constructs and measures travel across resources, or do they encode cultural bias? Phase 5 (Y5–7): Longitudinal studies — Does inferred readiness predict relational outcomes at 30, 90, 180 days? Phase 6 (Y5–7): Bereavement and wellbeing outcomes — Do the modules produce durable, transferable gains? Any change in effects? Phase 7 (Y7–9): Instrument validation — How should consequential readiness inference be prevented, justified, and sustained? Phase 8 (Y9–12): Public health application — Can a calibrated readiness-first model contribute measurably to social connection outcomes as measured at population scale?
Source: Please / scan Core questions answered
Relevance to European Research and Policy
The framework aligns with several active European priorities — and, consistent with this paper's tone — we describe its alignment as a basis for collaboration and scrutiny rather than as an enticement to funding.
Responsible AI
Responsible AI. A Consequential inference about psychologically sensitive states is exactly the category the EU AI Act and the responsible AI research community are most concerned with. A framework that manages such uncertainty, human oversight, and controllability in a coordinate case study for how such systems should be governed.
Digital Wellbeing and Youth Mental Health
Digital wellbeing and youth mental health. The Horizon Europe 'Health' cluster has listed it officially addressing the prevention of harm from digital technologies on the mental health of children and young adults, and innovative interventions for mental and behavioural conditions. A readiness-first model speaks directly to the 'healthy facility in a rapidly changing society' destination.
Social Innovation and Inclusion
Social innovation and inclusion. The Horizon Europe 'Culture, Creativity and Inclusive Society' cluster and complementary programme to target social inclusion and the reduction of isolation — the framework's stated public interest aim. Framework & research protocol of v1.0 / June 2026 / page 11 Wouch: A Readiness-First Framework for Human Connection.
Public Health and Loneliness Reduction
Public health and loneliness reduction. The 2025 World Health Assembly resolution on social connection and the WHO Commission report create a policy mandate for evidence-based interventions, with which a rigorously evaluated pilot would align. Funding instruments. Plausible homes include Horizon Europe Cluster 1 (Health) and Cluster 2 (Inclusive Society), BI-Health, and the Digital Europe Programme for the responsible AI components. Eligibility and fit would need to be confirmed against current calls; the relevant point is that multiple instruments exist whose objectives the framework genuinely aligns to. We note plainly that EU funding is competitive and that an uncalibrated framework would not, and should not, be funded on its promises alone. The realistic aim is support for the validation and pilot work in Phases 1–2 — the stage at which the framework's claims can first be tried — ideally within a consortium that includes independent academic partners and an ethics board.
Pilot Programme Framework
The pilot framework is designed to answer the first-order question — is the system safe, acceptable, and behaving as intended — before any efficacy claim is attempted.
Pilot Settings and Populations
Five formats are non-negotiable across all variants. First, research consent to operate from platforms becomes agreeing to use the product is never treated as agreeing to its studied. Second, harm monitoring is a primary outcome, not a secondary one — a pilot that cannot detect whether it is harming participants is not ethically runnable, regardless of the other terms.
PILOT PROGRAMME — SETTINGS AND POPULATIONS
University wellbeing services — Consenting students (18+): Acceptability, safety, self-reported usefulness. Counselling referral pathway; rollout on any on population. Young adult community pilots — Adults 18–28: Too not safety floor, not low cost of safety. Human review of all protective holds; assess mode. Digital-wellbeing dating initiative — Self-selected adults: Comparison vs. standard approach; wellbeing measures. Non-knowledge design; tolerance of alternatives. Research institutions — Recruited study samples: Construct replication; model calibration. Independent analytic cases; person measures. Public-health / NGO partners — At-risk or isolated groups: Reach and measurement among lonely populations. Enhanced safer casting; clinician oversight.
Source: Pilot framework settings overview
Open-Science Strategy
The framework aspires to open science, and it operates on sensitive, special-category data behind a proprietary inference engine. These aspirations are in genuine tension, and we state the tension rather than gloss it.
Open Research
Open research. Protocols, study designs, analysis plans, and pre-registrations can and should be public, regardless of the engine's confidentiality.
Transparency
Transparency. The framework's logic, construct definitions, and ethical commitments are publishable; the specific parameters that constitute competitive and safety-sensitive IP need not be, and in the case of the adversarial safety components arguably should not be, since publishing them would aid the behaviour they deter.
Reproducibility
Reproducibility. Reproducibility is pursued through shareable synthetic datasets, documented methods, and — where data cannot be shared — independent replication. Research partners control independent access to the same conditions.
Ethical Review and Data Governance
Ethical review and data governance. All studies require independent ethics approval; sensitive data is processed under a clear legal basis, minimised, and governed by retention limits and access controls.
Participant Protections
Participant protections. Separate research consent, with informed rights, the right to human review of consequential decisions, and the right to deletion. The honest summary is that Wouch wants to be fully open in the way a pure-academic instrument can, because it processes sensitive data and protects safety-relevant IP. What it can do — and what the strategy commits it to — is to be open about everything except the parameters whose secrecy is justified, and to subject even those to independent audit. Where openness and protection conflict, the resolution is disclosed, not hidden.
Limitations
This section is intentionally the most demanding in the paper. A framework that gives vulnerable people on inferred states must be at least as rigorous in cataloguing its own weaknesses as in advancing its claims.
Twelve Identified Limitations
1. No empirical validation. The four constructs, parameters, and thresholds have not been validated against real-world relational outcomes. The model can be internally coherent and entirely wrong. Every outcome-relevant claim in this paper is a hypothesis. 2. 'Readiness' is an uncalibrated construct. Its formal model basis is not an established measure with a psychometric literature. We are operationalising it newly, and its construct validity, convergent and discriminant validity, and test-retest reliability are unknown. 3. This instrument has heavy, heavy consequences. A short self-report assessment carries limited information, and several subscales and area items. Using such an instrument to gate access to connection places more inferential weight on it than its likely reliability supports until validation demonstrates otherwise. 4. Self-report and response-process assumptions. The framework relies on self-report, which is subject to social desirability, limited insight, and deliberate misrepresentation. Where it uses response-process signals such as latency, the interpretive assumptions are instances; not all response indicates automatically are themselves controllable and unmediated. 5. Risk of harm from gating. Communicating, however gently, that a person is not ready can be experience as rejection or someone already vulnerable to it, potentially compounding isolation. The net effect of gating — protective or harmful — is an open empirical question. 6. False positives in safety inference. Trauma-adapted responses can resemble the patterns the safety system aims to flag; a false positive can re-enact controlling dynamics under the banner of protection. Human review mitigates but does not eliminate this risk. 7. Cultural validity. Norms for closeness, interdependence, and boundaries vary across cultures. A single normative model risks encoding the assumptions of its designers as universal truths, with downstream fairness consequences. 8. Automated decision-making and controllability. Gating is a consequential automated decision. Even with human review and appeal, the burden the system places on individuals to contest its inference is a real cost that must be measured. 9. Engagement-suppressing design and selection effects. A system that discourages compulsive use may disproportionately retain already-reflective users, biasing any observed outcomes and limiting reach to those who most need support. 10. Reproducibility limits. Sensitive data and proprietary parameters constrain full reproducibility; independent verification depends on audit arrangements that are themselves not specified. 11. Equity of access. A readiness-first product that descends on non-engagement funding may struggle to reach low-income populations, precisely those among whom loneliness is most prevalent according to the WHO Commission. 12. Researcher and designer bias. The framework was built by a team with a point of view about what good relationships require. This view is itself a limitation until it is evaluated against diverse populations and independent scrutiny. None of these limitations is, in our judgement, disqualifying. Each is a reason for investigation rather than a reason for either premature deployment or premature dismissal. The framework earns the right to scale only by addressing them in sequence.
Future Research Questions
The following 100 questions are organised into ten themes. They are written to be answerable — suitable for master’s theses, doctoral research, and multi-site collaboration — and they deliberately include questions whose answers could undermine the framework, not only confirm it.
Construct Validity and Measurement
1. Does the proposed readiness construct demonstrate convergent validity with established relationshipfunctioning measures? 2. Does it demonstrate discriminant validity from general well-being and from attachment style? 3. What is the test–retest reliability of each latent construct over weeks and months? 4. How many assessment items are needed before each construct’s posterior stabilises? 5. Do the four constructs factor empirically as the theory predicts, or do they collapse or split? 6. Is the model’s stated confidence calibrated against observed accuracy? 7. How sensitive are inferences to item wording and ordering? 8. Does response latency add incremental predictive validity beyond response content? 9. What is the minimum-length assessment that preserves acceptable reliability? 10. How does measurement invariance hold across age, gender, and language groups?
Predictive Validity and Outcomes
11. Does inferred readiness predict relationship formation within a defined window? 12. Does it predict relationship quality at 30, 90, and 365 days? 13. Does it predict the avoidance of harmful or coercive relationships? 14. Are readiness gains associated with improved real-world relational outcomes, or only with assessment performance? 15. Does the framework predict outcomes better than a simple baseline (for example, self-rated readiness)? 16. What is the incremental value of the Bayesian model over a fixed-scale scoring approach? 17. Do module completions causally improve any measured construct, under a randomised design? 18. How durable are any gains six and twelve months after module completion? 19. Do gains transfer across relationship contexts, or are they context-specific? 20. What proportion of variance in outcomes does the framework leave unexplained?
Safety, Gating, and Harm
21. What are the false-positive and false-negative rates of the safety conditions? 22. What is the user-experienced cost of a false-positive protective hold? 23. Does gating reduce exposure to harmful dynamics, or primarily produce a sense of exclusion? 24. How do gated users describe the experience, and does it affect help-seeking? 25. Does human review measurably change protective-hold decisions and their outcomes? 26. What is the optimal balance between protective gating and respect for autonomy? 27. Are there subgroups for whom gating is net-harmful? 28. Does the appeal mechanism function fairly and accessibly in practice? 29. Can the safety system be adversarially manipulated, and how robust is it? 30. Does surfacing safety-relevant patterns risk re-traumatisation, and how can that be detected early?
Ethics and Autonomy
31. How do users understand and value the readiness-first model versus open access? 32. Is consent to sensitive processing genuinely informed at the point it is given? 33. Does the absence of engagement mechanics measurably change usage and affect? 34. How do users react when told, in behavioural language, about a pattern they had not examined? 35. What is the psychological effect of never being shown a numeric score? 36. Does withholding the triggering value reduce harm without reducing informed agency? 37. How should the right to human review of consequential inference be operationalised? 38. What are the autonomy costs of a system that pauses a person’s access ‘for their own good’? 39. Do users experience the framework as supportive or as judgemental, and what predicts the difference? 40. How can dark-pattern absence be independently audited and certified?
Cross-Cultural Validity
41. Do the constructs hold across the framework’s target markets? 42. Are gating thresholds equivalent across cultures, or do they encode bias? 43. How do norms of interdependence affect interpretation of ‘avoidant’ and ‘anxious’ patterns? 44. What is the appropriate process for culturally calibrating thresholds without tokenism? 45. Do behavioural indicators carry the same meaning across languages after translation? 46. How do collectivist versus individualist contexts affect readiness inference? 47. Are there cultures for which the framework is inappropriate as designed? 48. How should cultural calibration be governed to avoid designer assumptions dominating? 49. Do cross-market matches raise distinct safety or fairness issues? 50. What is the fairest way to handle thin local markets without forcing cross-cultural exposure?
The Bayesian Model and Methods
51. How robust are inferences to prior specification? 52. Does adaptive item selection improve efficiency without harming fairness? 53. How should the model handle missing or inconsistent responses? 54. What is the effect of recalibrating provisional parameters against real data? 55. Can the model detect its own miscalibration in deployment? 56. How should uncertainty be communicated to users without inducing anxiety? 57. Does modelling state change over time improve prediction over static estimates? 58. How does the model behave at the boundaries of its training assumptions? 59. What guardrails prevent feedback loops between inference and module assignment? 60. How should model versioning be governed so that users are assessed consistently?
Modules and Intervention
61. Which module components produce the largest construct gains? 62. What is the dose–response relationship between module engagement and change? 63. Do modules produce any iatrogenic effects in vulnerable users? 64. How does module efficacy compare with established brief interventions? 65. Do users complete modules, and what predicts drop-off? 66. Is module ordering causally important to outcomes? 67. Can modules be harmful if completed in a dysregulated state? 68. How do modules interact with concurrent therapy? 69. What is the optimal pacing to support reflection without inducing fatigue? 70. Do self-compassion-framed modules outperform neutrally-framed ones?
Population Health and Reach
71. Does the framework reach the lonely populations identified by the WHO Commission? 72. Is access equitable across income and education levels? 73. Does a non-engagement funding model limit reach, and how can that be mitigated? 74. What is the population-level effect, if any, on social-connection measures? 75. Does the framework reduce or redistribute relational harm? 76. How does reach compare across the framework’s primary and secondary markets? 77. Are there displacement effects on users who are gated out? 78. What is the cost-effectiveness relative to other social-connection interventions? 79. Does the framework complement or compete with public mental-health services? 80. What is the appropriate role of public funding given the access-equity tension?
AI Governance and Accountability
81. How should consequential relational inference be regulated under the EU AI Act? 82. What audit standards are appropriate for a proprietary inference engine on sensitive data? 83. How can independent oversight be structured without compromising safety-relevant IP? 84. What transparency is owed to a user about an inference that affects their access? 85. How should liability be allocated when an automated pause causes harm? 86. What documentation standards make the system genuinely contestable? 87. How should drift between model and reality be monitored and acted upon? 88. What governance prevents mission creep from readiness into surveillance? 89. How should special-category data be handled across jurisdictions with differing rules? 90. What is the right composition and authority of an external ethics board?
Longitudinal and Societal Effects
91. What are the long-term effects of readiness-first design on users’ relational beliefs? 92. Does early exposure shape later relationship expectations? 93. Does the framework change how users interpret rejection and uncertainty? 94. Are there second-order effects on the broader dating ecosystem? 95. Does a readiness gate alter the timing of relationship formation, and with what consequences? 96. What happens to users over multiple re-assessments across years? 97. Does the framework reduce repeat exposure to harmful dynamics over time? 98. How do outcomes compare for users who complete modules versus those who do not? 99. What are the effects on users who never reach the matching threshold? 100. Does readiness-first design produce measurable societal social-connection benefit at scale?
Conclusion
We have not argued that Wouch works. We have argued that readiness-first connection technology is a coherent, theoretically grounded, and fairly direction that deserves rigorous investigation — and we have been explicit that whether it works is precisely what is not yet known. The case for investigation rests on three observations. First, the dominant design logic of connection platforms is, at best, indifferent to the conditions under which good relationships form, and the public health case for treating connection right-focus is recognized at the highest international level. Second, the theoretical traditions the framework draws on — attachment, polyvagal, trauma, relationship science — are mature enough to model a different design, even if the synthesis is new. Third, the framework can be specified precisely enough to be tested, audited, and falsified, which is the minimum condition for treating it as research rather than advocacy. The case against the framework is equally true and is stated at length in the limitations: the constructs are unvalidated, the consequences of getting on premature are real, and the risks of misclassifying vulnerable people are real. The responsible path is to provide a staged programme of validation, pilot, and independent scrutiny we set here. If readiness-first solutions can be shown to help — honestly, with attention to harm and, without the manipulations of the engagement economy — that is worth knowing. If they cannot, that is worth knowing too.
WOUCH, POSITION PAPER SERIES, ISSUE 03
Author: Vicky Verma. Founder, Wouch.
Contact: vicky@wouch.app
Framework & research protocol of v1.0 / June 2026 / Wouch: A Readiness-First Framework for Human Connection.
Note on sources. Foundational references are standard works in their fields; Horizon policy and empirical sources were verified current as of June 2026, including the WHO Commission on Social Connection report and a 2021 pre-registered systematic review and meta-analysis of dating-app use and psychological outcomes published in Computers in Human Behavior. A full reference for the latter, and for any additional recent studies cited in a subsequent revision, should be confirmed against the journal of record before formal submission.
This document was produced for distribution. It is not behind a paywall.
It operates on open science, and it operates on sensitive, special-category data behind a proprietary inference engine.