Reviewed for evidence accuracy · July 2026
Most of what you've heard about relationships is either well-measured or merely famous, and the two sets barely overlap.
That sentence is the whole reason this page exists. The research on how couples fight and recover is real, and some of it is very good. But the parts that made it into podcasts, workbooks, and quiz formats are, with unhelpful regularity, the parts that didn't survive checking. So before we ask you to trust a tool we built, here's the ground it stands on: what we take as solid, what we refuse to say, whose relationships the research actually studied, and the places where we're guessing along with everyone else.
We'd rather you finish this page a little more skeptical of us and a lot better equipped than you started.
The field is split, and the split matters
There are two traditions in relationship science, and they have wildly different relationships with evidence.
One is a quiet, rigorous, mostly unglamorous body of work: observational studies where couples were wired to heart-rate monitors and their fights coded second by second, randomized trials of couples therapy, longitudinal follow-ups years later. It produced findings that replicate. Very few of them are exciting.
The other is a tradition of brilliant clinicians who wrote beloved books. Their frameworks are often clinically rich and genuinely useful in a therapist's hands. Most of them have never been tested in a controlled trial, and a few of their most quotable claims have been tested and didn't hold.
The uncomfortable fact underneath: the most evidence-backed ideas in this field are not the most famous ones, and several of the most famous claims are the least defensible. A product that quietly launders the second tradition into the first is lying to you at scale, one confident sentence at a time. We label every idea we hand you with where it actually stands. There's a legend for that, and it applies here too.
What the evidence agrees on
Cross-school agreement in this field is narrow, and the narrow part is worth more than the wide claims. Four things hold up.
Your body decides before you do. Under conflict, arousal climbs. Somewhere around 100 beats per minute (closer to 85 for athletic people), the parts of the brain that listen, take another perspective, and solve problems start going offline. Gottman called this state flooding. The word sounds like a feeling; it was read off a heart-rate monitor. Above that line, a person cannot take in what their partner is saying no matter how badly they want to, and adding logic to the conversation makes things worse, because a threat-detection system reads every new input as more threat. The stress chemistry that put you there takes twenty minutes or more to clear, and only if you actually stop chewing on the argument. Rumination resets the clock. In one of Gottman's own studies, couples who were interrupted for half an hour and handed magazines came back with lower heart rates and a better conversation than the ones who kept going. (Research-backed)
Most of what you fight about will never be solved. In Gottman's data, roughly 69% of a couple's conflict is perpetual: rooted in stable differences in temperament, values, and need. The part that startles people is the follow-on finding. Thriving couples and divorcing couples have the same proportion of unsolvable problems. What differs is whether those problems get talked about or barricaded. So a tool that measures its success by whether the fight got resolved is measuring the wrong thing about two-thirds of the time, and it will push you toward resolution in exactly the moments when resolution isn't available. (Research-backed)
Repair is the variable. Rupture is ordinary. Every close bond breaks contact and re-establishes it, and the research lineage on this runs from infant studies through decades of couples work: a well-repaired rupture can leave a bond stronger than it was before the break. Fight count doesn't sort the flourishing from the failing. Whether the repair attempt gets made, and whether the other person can receive it, does. (Research-backed)
No school owns the truth. Across dozens of studies, couples therapy works, with a large average improvement in relationship satisfaction, while couples left on a waitlist don't improve. But when the major approaches are compared head to head, none of them decisively wins. The best available reading is that what does the work is common to all of them: a structure, a way to slow down reactivity, some safety, and the experience of being understood. This is why we don't sell you a brand of relationship therapy. We built to the consensus. (Research-backed)
Then a fifth thing, which is less a finding than an indictment: only about a third of distressed couples ever get to couples therapy. The best-supported intervention in the field is one most people never receive. That gap is the entire reason tools like ours exist, and it doesn't make us therapy.
What we build on directly
The most-tested couples approach is Integrative Behavioral Couple Therapy, which pairs behavior change with the harder, stranger idea of accepting what won't change. It is also the one framework that has been successfully translated to a screen. Brian Doss adapted it into a self-guided web program, ran it through randomized trials with thousands of couples, and got relationship-satisfaction effects in the range you'd hope for from in-person therapy, holding up about a year later. The finding that mattered most to us: the versions with almost no human coach performed about as well as the versions with more support. The content did the work. (Research-backed)
Sitting inside that program is the most portable idea we found anywhere in the literature, an analysis its authors call DEEP. When you're stuck, four forces are usually doing the damage: plain personality differences, each person's emotional sensitivities, external stress that has nothing to do with either of you, and the pattern the two of you fall into when the first three collide. It converts "you're the problem" into "we're caught in something." That move is structured enough to be built, and it is the spine of what our tool does with a conflict once you're calm enough to look at it.
And one clinical idea, which we take seriously while being honest that it hasn't been trialed. Dan Wile's insight was that a fight is often two people failing to say the vulnerable thing underneath the hostile thing, and that the intervention is to say it for them: to take "you never listen to me" and offer back "I felt invisible in that room, and it scared me." That's a sentence a flooded person cannot generate about themselves, and a machine can offer it in the moment when no one else is around to. We think this is the highest-leverage thing our tool does. We also think you deserve to know it rests on clinical judgment rather than trial data. (Clinically grounded)
Three famous claims you won't hear from us
That fights predict divorce with 90-something percent accuracy. This is the most-repeated statistic in the field and it should never have left the lab. The models were fitted to outcomes that were already known, which is postdiction wearing prediction's coat. When Richard Heyman built a similarly impressive equation and then tested it on couples it hadn't seen, accuracy fell to about 29%. Gottman's descriptive work is a real contribution and we lean on it throughout this page. The prediction percentages aren't part of it.
That you and your partner have love languages, and matching them makes you happy. Reviewed against the actual evidence in 2024, the three assumptions holding this up all failed: that each person has one primary language, that there are five of them, and that matched couples do better. A separate study found matched couples are no happier than mismatched ones. People appreciate care in all its forms. The framework is a useful conversation starter and a bad theory, and we won't dress it up as a finding.
That your vagus nerve explains what happened to you last night. Polyvagal theory gave the wellness world its vocabulary for safety, shutdown, and co-regulation. In February 2026, a large group of physiologists published a paper in Clinical Neuropsychiatry arguing the theory is untenable on its core anatomical and evolutionary claims. Porges published a rebuttal in the same issue. The dispute is live and we're not going to pretend it's settled either way. What we do is separate the plumbing from the practice: we'll keep using words like activated, settled, and co-regulation, because they describe something you can feel and check against your own experience. We won't tell you which nerve did it. Our own physiology claims rest on ordinary stress-arousal science, which never depended on the contested theory in the first place. (Practice tradition)
Whose relationships did the research describe?
Mostly: heterosexual, married, largely white, mid-twentieth-century American couples. That's the sample the canon was built on, and it's a real constraint rather than a disclaimer we're clearing our throat with.
The consequences are specific. In unhappy same-sex couples, the relationship between physiological arousal and distress can run the other way, which means the arousal model at the center of this entire page is not universal. Autistic and ADHD people frequently can't read their own arousal reliably from the inside, so "notice you're activated" is an instruction that assumes a body that reports. "You always do this" lands as a factual error before it lands as a feeling, and for some people the factual error has to be dealt with first or nothing else can be heard. The frameworks assume a "we" with a costly exit, which is not the situation of someone four months into something undefined. And almost none of it was written with more than two people in the room.
So we hold the canon as a set of strong hypotheses about you rather than facts about you, we let you tell us who you are instead of inferring it, and where the research is silent we say so instead of improvising something that sounds clinical.
What we don't know about our own product
This is the section we argued about, and it's the one that made the page worth writing.
One person, alone, is a genuinely open question. Our tool is built for the person who's in it by themselves at 11pm, because that's who's actually reaching for a phone mid-fight. But the field ran something close to a clean test of solo work: the same proven web program, delivered to one partner instead of two. It improved that person's quality of life, their functioning, and how they felt about their own health. It did not move relationship satisfaction. Understanding, it turns out, doesn't transmit across the kitchen table on its own. What appears to move a relationship from one side is changed behavior, not insight, which is why every session we run ends with something you're actually going to do or say differently rather than a nice realization. It's also why we say what we can honestly say: change your half, influence the rest, get clear either way. We can't promise you a fixed relationship, and anyone who does is selling.
Nobody has proven that any of this works compressed into the moment. Every result cited on this page comes from planned, multi-session, deliberate work. Eight hours of program. Structured training. A therapist's room on a Tuesday. Not a fifteen-minute intervention delivered while you're locked in the bathroom with your heart going. Whether these effects survive that compression is the bet this company is making, and it hasn't been tested. We're designing our own evaluation to find out, and we'll publish what we learn.
We can't reliably tell when you're flooded. The whole loop depends on it, and every available signal is weak. Wearable heart-rate data isn't in the product. Reading escalation from your text is exactly where language models are documented to fail, on sarcasm most of all. Self-report is unreliable, in different ways for different bodies. So we gate conservatively, we'd rather send you back to a break you didn't need than into a conversation you weren't ready for, and when the system sends you back, that's the system working.
What all of that made us build
The evidence didn't just decorate the product. It set its rules, and most of the rules are refusals.
We won't reason with you while you're flooded, because it makes things worse. We won't let you vent to us during a break, because rehearsing the argument is how a twenty-minute time-out becomes an hour of staying angry. We won't write the message and send it for you; we'll offer you a sentence, and you'll edit it until it's true and send it in your own voice. We won't tell you who's right, because there are two realities in every fight and a verdict helps neither. We won't diagnose you or your partner, and we won't guess at what your partner is feeling, because they're not in the room and profiling an absent person from one side is a thing we've decided not to do.
We're a coaching and communication tool. Not therapy, not treatment, not a clinician. That's a legal line and it's also just true.
When this isn't the help you need
If what's happening at home is coercion, fear, or violence, this is the one place the research is unambiguous: couple-style work is contraindicated there, and a tool that tries to mediate it can make you less safe. That's not a limitation we're admitting reluctantly. It's the reason we built the exits before we built the rest.
If you're carrying something heavier than conflict, depression, trauma, an addiction, a crisis tonight, the honest answer is a licensed human being, and we'd rather point you to one than keep you here. There's a short shelf of books we trust in the Reading Room, and a plain-language glossary if the vocabulary is what's in your way. Take those, use nothing else of ours, and we'll consider this page to have done its job.
Something here off, broken, or not sitting right? Tell us — a person reads it.
People and books appear here because we’d recommend them with nothing to gain. Some may partner with us over time; no one is listed because of it.