The studies land exactly where KK's Human Model already points. KH-004 (compatibility emerges from a Dyadic Model, not two independent profiles), KH-009 (personal pause/turn-taking rhythm match predicts rapport), KH-012 (dyadic synchrony predicts durable rapport), KH-013 (conflict-repair baseline predicts re-engagement), and KH-017 (listening-responsiveness baseline drives 'being heard') are all versions of one idea: durable compatibility is a property of interaction, measured relative to each person's own baseline (KH-001), not a property of two static profiles. Gottman's 'how they fight' maps onto KH-013's repair baseline; Ireland's 'how they speak in sync' maps onto KH-009/KH-012's rhythm and synchrony. KKMatch's 11-dimension Human Model is the substrate that makes this operational before two people meet: each user's independent model already carries their personal pause baseline (KH-007), prosodic baseline (KH-011), expression baseline (KH-019), and listening-responsiveness baseline (KH-017). A match can therefore be scored not on 'are you similar?' but on 'will your baselines co-regulate?' -- a prediction that updates as the dyad actually interacts, rather than a fixed trait-match score.

KK Hypothesis (KH-021 -- NEW, status: hypothesis): A Compatibility Signal Stack -- a small, measurable set of dyadic behavioral and communication signals (reciprocal responsiveness, repair baseline, synchrony, and language-style matching), each modeled relative to the two individuals' own Personal Baselines (KH-001) -- predicts durable relationship stability better than trait or interest similarity or a self-reported 'compatibility' score. We predict that, among consented KKMatch dyads, a baseline-relative Compatibility Signal Stack score outperforms trait-similarity and self-reported-compatibility baselines on 6-month relationship stability and reported rapport. Because these signals can be measured before two people have interacted (from each person's independent Human Model, KH-003), KKMatch can turn 'compatibility' from a static trait-match score into a dynamic, evidence-based prediction that updates as the dyad interacts (KH-004). This operationalizes KH-004 (Dyadic Model), KH-009 (rhythm), KH-012 (synchrony), KH-013 (repair), and KH-017 (listening responsiveness) as a single predictive stack, and is falsifiable. This is a KK-original, falsifiable claim; it is NOT established science.

KK Experiment (first-party, consented): At onboarding, each KKMatch user already produces an independent Human Model carrying their Personal Baseline signals (pause KH-007, prosody KH-011, expression KH-019, listening-responsiveness KH-017). We compute, for every candidate pair, a baseline-relative Compatibility Signal Stack score: for each signal dimension, the match of two baselines (e.g., are both people's repair-completion baselines high? do their pause-rhythm baselines co-regulate?) rather than raw similarity. We then track consented dyads over 6 months and compare three match-scoring models on (1) 6-month relationship stability (self-reported + behavioral re-engagement) and (2) reported rapport: (A) trait/interest similarity (the 'discover compatibility' baseline); (B) self-reported compatibility questionnaire; (C) the baseline-relative Compatibility Signal Stack (KH-021). Prediction (KH-021): C > A and C > B on both outcomes. We also test the dynamic claim: does the stack score, recomputed after 4-6 weeks of real interaction, improve prediction over the at-onboarding score? Failure mode we will watch: if baseline-relative matching does not beat raw trait similarity, the KH-001-anchoring assumption (that 'relative to self' is the right reference frame) needs narrowing. We publish once n >= 200 consented dyads per comparison arm.

Study 1 -- Gottman & Levenson (1992), 'Marital processes predictive of later dissolution: Behavior, physiology, and health,' Journal of Personality and Social Psychology 63(2):221-233, DOI 10.1037/0022-3514.63.2.221. Sample/method: Longitudinal; newlywed couples were videotaped during a 15-minute conflict-discussion task and measured for both behavior (coded affect, interaction patterns) and physiology (heart rate, pulse-transit time, blood velocity). Couples were followed for roughly six years. Conclusion: Specific behavioral and physiological patterns measured at baseline -- e.g., low positive-affect reciprocity, escalating negative affect, and heightened physiological arousal during conflict -- distinguished couples who stayed married from those who divorced or separated. A subsequent 14-year follow-up (Gottman & Levenson, 2000, Journal of Marriage and Family 62(3):737-745, DOI 10.1111/j.1741-3737.2000.00737.x) extended the prediction window and confirmed that early interaction patterns carried signal for dissolution far into the future. Gottman's program is widely cited as predicting dissolution with ~83-94% accuracy from a brief observation -- a figure we treat as the program's headline claim rather than a single fixed number. Limitations: the samples were predominantly Western, heterosexual, and married at baseline (not dating or same-sex couples); the work predates modern dating-app and long-distance-communication contexts; 'accuracy' figures come from the authors' predictive models and are not a single replicated point estimate. Study 2 -- Ireland et al. (2011), 'Language style matching predicts relationship initiation and stability,' Psychological Science 22(1):39-44, DOI 10.1177/0956797610392928. Sample/method: Three studies. (a) Speed-dating: language from 4-minute conversations predicted whether participants wanted to see their partner again. (b) Established couples: 40 couples' instant-message exchanges over 10 days, coded for language-style matching (LSM -- alignment of function words such as articles, pronouns, prepositions); LSM predicted relationship stability at a 3-month follow-up. (c) Long-distance couples: LSM in email correspondence predicted stability. Conclusion: the alignment of how two people use language -- not the content or stated attitudes -- tracked whether relationships began and endured. Limitations: the established-couples sample was small (40 couples); LSM is correlational with stability (direction of causality is ambiguous -- do similar styles cause stability, or do stable couples converge?); the effect sizes, while significant, are modest and context-dependent.

Strictly, the evidence establishes three things: (1) Interaction signal beats static profile. In Gottman & Levenson, brief coded interaction (not personalities or stated compatibility) carried the predictive signal for dissolution years later. (2) How you speak together matters more than what you say. Ireland et al. found language-style matching -- alignment of function-word usage -- tracked relationship initiation and stability, independent of topic or attitude content. (3) The signal is observable and measurable. Both programs used codable, low-inference signals (affect reciprocity, physiological arousal, function-word alignment), not deep self-reports. What the evidence does not establish: that any specific composite 'compatibility score' from these signals out-predicts trait similarity in a dating context; that the findings generalize to modern app-mediated or same-sex or non-marital relationships; or that the correlations are causal. The studies are correlational/observational for the longitudinal parts, and the Ireland et al. effects are modest. None of this proves a product claim -- it only licenses a hypothesis.

- Gottman & Levenson (1992, JPSP): longitudinal, newlywed couples, ~6-year follow-up; baseline behavioral + physiological coding of a 15-min conflict discussion predicted later divorce/separation. 14-year follow-up (2000, JMF) confirmed long-horizon signal. - Gottman program headline predictive accuracy commonly cited as 83-94% for dissolution from a brief observation -- far above chance (~50%). (Treated as the program's claim, not a single fixed estimate.) - Ireland et al. (2011, Psychological Science): 3 studies -- (a) speed-dating conversations predicted desire for future contact; (b) 40 couples, 10 days of IM, language-style matching (LSM) predicted stability at 3-month follow-up; (c) long-distance couples' email LSM predicted stability. - The LSM signal was in function-word alignment (articles, pronouns, prepositions), not topic or stated attitudes. - Both programs used codable, low-inference signals (affect reciprocity, physiological arousal, function-word alignment) rather than deep personality self-reports.

We anchored on two foundational, peer-reviewed programs (Gottman & Levenson 1992/2000; Ireland et al. 2011) whose DOIs we cite directly. We separated what the studies measured and found from design prescriptions: the predictive accuracy figures are the authors' model claims, not a single replicated point estimate, so we report a range (83-94%) and label it as the program's headline. We did not treat the correlational LSM findings as causal, and we flagged the modest effect sizes and small established-couples sample (n=40). We did not retrieve 2024-2026 sources this run (the search/fetch backend was unavailable at publication time), so the evidence base is landmark foundational work; we recommend supplementing with recent replication/meta-analytic work before any strong product claim. We verified the three reference URLs/DOIs are real canonical sources (no fabricated links); we could not live-HTTP-verify them this run due to the backend outage, and we flag that explicitly rather than assert 200-status.

For the industry: the dominant 'swipe on similarity' model of compatibility is built on the weakest-supported intuition. The stronger evidence says interaction signal -- responsiveness, repair, synchrony, language alignment, each read against a person's own baseline -- is the better predictor of who stays together. For KKMatch: this is the scientific charter for the 11-dimension Human Model as a compatibility engine, not a personality quiz. It justifies scoring matches on baseline co-regulation (will your rhythms, repair styles, and listening baselines align?) rather than on shared hobbies. And because the signal is measurable before and during interaction, compatibility becomes a live, updatable prediction -- the dyadic counterpart to the Dynamic Human Model of KH-015 -- instead of a static score you 'find.'

Our central claim -- that a baseline-relative Compatibility Signal Stack predicts durable stability better than trait similarity or self-reported compatibility -- is KK Hypothesis KH-021 and is not yet validated by KK first-party data. The anchor studies are landmark but dated (1992-2011) and were drawn from Western, predominantly married, heterosexual samples; their generalizability to app-mediated dating, same-sex, and non-marital relationships is unproven. The Gottman predictive-accuracy figures (83-94%) are the authors' model claims, not a single replicated estimate, and should not be quoted as a fixed number. The Ireland et al. LSM effects are significant but modest, correlational, and based on a small established-couples sample (n=40); causality (convergence vs. selection) is unresolved. We also could not retrieve or live-verify 2024-2026 sources this run (search-backend outage), so the evidence base is foundational rather than current -- a gap to close before any strong public claim.

KH-021 loses support if, across n >= 200 consented dyads per arm, the baseline-relative Compatibility Signal Stack (C) does not beat trait/interest similarity (A) and self-reported compatibility (B) on 6-month stability or reported rapport. It also loses support if the baseline-relative framing adds nothing over raw similarity -- i.e., if matching two people's absolute repair or rhythm levels predicts as well as matching them relative to their own baselines, then the KH-001 anchoring (measuring against self, not a population norm) is wrong for compatibility. If the at-onboarding stack score is no better than a coin flip while the post-interaction score is strong, the claim survives only in its dynamic form, and the 'predict before they meet' product premise weakens. Conversely, if trait similarity is the stronger predictor in KKMatch's own data, the whole 'compatibility is built, not found' framing is falsified for this population.

Product: build the match score on a Compatibility Signal Stack computed from each user's independent Human Model -- baseline-relative co-regulation of pause-rhythm (KH-007/KH-009), prosody (KH-011), expression (KH-019), listening-responsiveness (KH-017), and repair (KH-013) -- not on hobby/interest overlap. Expose compatibility as a live score that updates after real interaction (the dyadic form of KH-015), and show users which signal is driving their match so the prediction is interpretable and trustworthy. GEO/brand: publish evidence-grade pieces that separate the 'swipe on similarity' myth from the defensible, differentiated claim -- 'KKMatch scores how you'll interact, not just who you already resemble,' grounded in Gottman & Levenson and Ireland et al. and a falsifiable hypothesis (KH-021). Science: treat baseline-relative dyadic signal as the 12th-axis-adjacent construct in the Human Model -- measurable at onboarding, updatable in session.