Why the test has the shape it has, what evidence sits under each dimension, and how the scoring works. Written to be checked, not just believed.
The established premarital instruments (PREPARE/ENRICH, RELATE, FOCCUS) are good. Published follow-up studies report roughly 80% accuracy predicting marital outcomes over five years. But they share three structural limitations:
The assessment's three-layer architecture maps directly onto Karney and Bradbury's (1995) vulnerability-stress-adaptation (VSA) model, the most-cited integrative framework in marital research. All three components were independently confirmed as predictors of relationship satisfaction in a 2022 replication.
| VSA component | Assessment layer | What it captures |
|---|---|---|
| Enduring vulnerabilities | Layer 1 | Attachment, personality, family-of-origin, relationship narrative |
| Stressful events | Layer 2 | Financial alignment, life transitions, digital strain |
| Adaptive processes | Layer 3 | Conflict conduct, repair, commitment structure |
The model's central claim, and ours: outcomes are decided by the interaction, not by any single factor. Vulnerable people under high stress with weak adaptation decline; the same vulnerabilities with strong adaptation and manageable stress do fine. This is why the report never reduces to a single score.
Attachment security (weight 2). Insecure attachment, on both the anxiety and avoidance dimensions, negatively predicts relationship satisfaction across a 132-study meta-analysis, with both actor and partner effects. Items are modeled on the Experiences in Close Relationships constructs. A validated deployment would license or adopt the full ECR-R short form.
Emotional stability (weight 2). Malouff et al. (2010; 19 samples, N=3,848) found four Big Five traits predict intimate-partner satisfaction: low neuroticism, high agreeableness, high conscientiousness, high extraversion, robust across gender and marital status. Neuroticism is the most consistent negative predictor in the longitudinal literature, so v0.1 indexes emotional stability specifically. Openness does not predict satisfaction and is omitted deliberately.
Family-of-origin awareness (weight 1). RELATE (Busby, Holman and Taniguchi, 2001) established family background as one of four distinct predictive contexts. Our items measure awareness and processing of family patterns rather than the raw history itself. The working hypothesis, consistent with clinical literature but needing our own validation, is that examined patterns do less damage than unexamined ones.
Relationship narrative (weight 1). Huston's PAIR project (168 couples, 13 years) found courtship dynamics predict divorce versus stability, satisfaction among those who stay, and speed of dissolution among those who split. Our items probe whether the couple's origin story is one of active choice versus drift.
Financial alignment (weight 2). Financial strain is the most common chronic external stressor, and the VSA model shows stress exposure independently predicts outcomes. Items measure transparency, ease of money conversations, shared goals, and compatible instincts. The gap between partners is scored, not just the levels.
Life transition readiness (weight 1). Measures whether dealbreaker-class futures (children, geography, career priority, eldercare) have been explicitly discussed. Low scores are treated as an unmapped-future risk factor, because unanticipated stressors hit adaptation hardest.
Digital boundaries (weight 1). Partner phubbing shows significant negative associations with relationship satisfaction (correlations around -.22 to -.27) in a 52-study meta-analysis with 19,698 participants. No legacy instrument measures this.
Conflict competence (weight 2). Gottman and Levenson (1992) distinguished regulated from nonregulated couples by the balance of positive to negative affect during conflict; nonregulated couples had roughly triple the divorce rate (19% versus 7%). Items index soft startup, regret-speech, staying in the room, and perspective gain.
Repair capacity (weight 3, top tier). What distinguishes stable couples is not the absence of negativity but whether de-escalation attempts are made and accepted. Measured twice: by self-report items here and behaviorally in the scenario reflections.
Commitment structure (weight 3, top tier). Karney and Bradbury's review of 115 longitudinal studies found the satisfaction-to-stability link is well replicated but modest; stably happy couples still divorce at around 14% over ten years. Social exchange theory holds that stability is governed by dedication plus barriers and alternatives, so we measure the commitment architecture directly rather than inferring it from satisfaction.
Mutuality and appreciation (weight 3, top tier; added in v0.3). The largest machine-learning study of couples ever run (Joel et al., 43 longitudinal datasets, 11,196 couples) found that relationship-specific perceptions, above all perceived partner commitment, appreciation, and perceived partner satisfaction, out-predicted every individual-difference variable including personality. This dimension measures exactly those: whether each partner believes the other is all-in, believes the other is happy, and both gives and receives appreciation.
The scenario protocol exists because of a methodological finding: observed behavior during disagreement predicts outcomes beyond self-report. Gottman's lab got its predictive results by watching couples discuss real disagreements and coding the interaction. That method needs trained coders and lab time, so it does not scale. Part two approximates it in a self-administered format:
The six scenarios each target a domain with documented predictive weight (money, values and autonomy, in-law boundaries, children timeline, career sacrifice, digital attention), and roles alternate so each partner initiates three and responds to three, revealing whether conflict behavior depends on position. The known weakness of self-coded behavior is covered on the limitations page.
Step 1: dimension scores. Each item is a 1-5 rating, reverse-scored where marked, averaged within its dimension, and rescaled to 0-100.
Step 2: gap analysis. For every dimension we compute the couple average and the gap between partners. Gaps over 30 points on high-weight dimensions flag as alerts; gaps over 15 flag for conversation. On structural dimensions like money and commitment, the disagreement itself is the risk factor, independent of levels.
Step 3: behavioral tallies. Across 6 scenarios and 2 partners (12 reflection reports) we count each marker. The composite conflict index weights contempt three times as heavily, reflecting its status as the single strongest behavioral divorce predictor. We also compute the repair-acceptance rate and the resolution rate (conversations that ended with a concrete next step).
Step 4: contextual calibration. v0.1 collects financial situation, cohabitation, duration, and children-intent, and uses them lightly. The validated version will implement the socioeconomic moderation finding directly, scoring demand/withdraw-like patterns as less harmful for resource-constrained couples. The hook ships before the full model because the evidence says context-blind scoring is wrong.
Step 5: classification. Three outputs. The partnership type is one of seven profiles grounded in Olson and Fowers' cluster analysis of 8,385 couples, assigned in v0.1 by transparent threshold rules approximating the cluster profiles; the validated version re-derives clusters from our own data. The conflict trajectory (low, middle, high) mirrors the finding that 20-year conflict trajectories form three stable classes of roughly 17%, 61%, and 23%. The satisfaction forecast (stable, declining, at-risk) comes from a risk-point model over contempt, alert-level gaps, weak commitment, low emotional stability, low repair acceptance, and low resolution rate, framed against the finding that 50-90% of couples maintain stable satisfaction.
Step 6: action plan. Findings map to a playbook of interventions, each tied to an evidence-based practice: soft startup training, repair rituals, the appreciation habit as the contempt antidote, structured future-mapping, financial-therapy referral. Contempt, if present, always comes first. Maximum three actions, because behavior change fails when overloaded.
The Enneagram is the inspiration for the format (a memorable identity, not a score) but deliberately not the foundation. A systematic review of 104 independent samples (Hook et al., 2021, Journal of Clinical Psychology) found that factor analyses of major Enneagram instruments generally fail to recover nine factors, and no peer-reviewed study has ever derived the nine types through clustering. Our seven types come from the only large empirical cluster analysis of couples in the literature. The principle: typologies must be derived from data, not imposed on it. The same standard will be applied to our own types, which will be re-derived from our own couple data and revised if the data disagrees.
Claims that failed adversarial source verification were excluded from the design, including the popularized "over 90% prediction accuracy" figure and the two-trajectory divorce timeline (5.6 versus 16.2 years).