Recovery has decades of real science behind it. Marrow is built on that science, and this page is where we show our work, including where the evidence is strong and where it is honestly still thin.
Most recovery apps tell you they work. Very few show you why, or admit what they cannot do. We think that is backwards, especially for something as serious as recovery.
So here is every load-bearing design choice in Marrow, the research behind it, and an honest label for how strong that evidence really is: strong when it rests on meta-analyses or randomized trials, good when the base is real but smaller, promising when it is early, and practice when it is widely used but not trial-tested on its own.
Each lantern below is one claim on this page, sized and colored by how strong the evidence behind it honestly is. Hover or tap to read it; follow it to jump to the full entry. The dimmer lanterns stay on the map on purpose: a weak claim is never hidden, only labelled.
Hover, or touch one. Every light here is a claim on this page, not a decoration.
This is the decision everything else grows from, and it is the one most recovery apps get backwards.
The abstinence violation effect describes how a single lapse, when read as proof of personal failure, sets off the guilt and hopelessness that push a lapse into a full return to use. A sober-day counter that drops to zero on a slip is that failure story, drawn as a user interface. Marrow keeps two numbers instead: your days in practice, which never reset, and your current stretch, which begins again without erasing anything.
In a prospective study, recovering people who showed more shame while describing a past slip were more likely to relapse, and to relapse harder, over the following months. Self-reported guilt showed no such link. The takeaway the authors themselves drew: shaming people for a hard-to-control behavior is close to the worst thing you can do. Marrow's return flow is built to lower shame first, separating what happened from who you are, before anything else.
Urge surfing, watching a craving rise and fall without acting on it, is the core skill of Mindfulness-Based Relapse Prevention. In a randomized trial against standard relapse prevention and usual care, MBRP showed added benefit at twelve months, with fewer days of use and heavy drinking. The Pull turns this into something you do: rate the wave, breathe, and watch it come down, then keep a record of every wave you have ridden.
Marrow offers the twelve steps and a secular, CBT-based track side by side, because the evidence points to more than one valid road.
The 2020 Cochrane review of 27 studies found that manualized twelve-step facilitation produced higher rates of continuous abstinence than other active treatments, including standard CBT, and did so at lower cost. Marrow walks the steps in its own plain words, never reproducing copyrighted literature, and always points toward real meetings and a human sponsor.
When researchers took apart how twelve-step programs actually work, the single strongest driver was social: shedding people who use, adding people who do not, and building confidence in hard social moments. It accounted for the majority of the benefit. So Marrow treats the written work as a beginning, and keeps turning you outward, toward one person to tell, a meeting, or a room of people doing the same work.
SAMHSA is explicit that recovery has many pathways, and comparisons across twelve-step, SMART, and other frameworks find broadly comparable outcomes, with the freedom to choose a good fit lowering the odds you drop out. Marrow's SMART track offers the same evidence-based tools, cost-benefit work and catching the thought that gives the pull permission, in the app's own voice.
NIDA is clear that a return to use does not mean treatment has failed. Recurrence rates for substance use disorders, roughly forty to sixty percent, sit right alongside those for asthma and high blood pressure. It is a signal to resume and adjust support, not a verdict. Marrow says this plainly at the moment it matters most, and keeps the door back in one tap away.
After a period away, tolerance drops, so a dose that once felt ordinary can become an overdose, a risk that is especially acute with opioids. Because of this, Marrow's return flow checks on your safety first, surfacing plain harm-reduction guidance and a real human line, before it offers any reflection or reframe. Self-compassion, rather than self-blame, is associated with better outcomes from here. The postures behind this flow are named on the Trust page.
Recovery capital, the sum of the internal and external resources you can draw on, your people, health, meaning, and coping, predicts who initiates and sustains recovery. It is a strengths measure, not a deficit one, and it survives a return untouched. Marrow's Standing reading uses the validated short-form items over these domains, scored only on your device and shown as a picture that grows with your resources, never as a number to protect.
Implementation intentions, plans in the shape of "when this happens, then I will," reliably improve follow-through on a goal. The effect is smaller and more mixed in substance use specifically than in general behavior change, so Marrow offers plans, never demands them, and resurfaces your own words at the moment your trigger tends to arrive.
The helper-therapy principle is well documented: those who support fellow members are meaningfully more likely to stay abstinent and report less depression. It is a genuine active ingredient, not a nicety. Marrow builds simple, unquantified ways to turn outward and give something back, with no service streak to chase.
Professional guidance on AI wellness chatbots calls for persistent transparency that you are talking to an AI, rigorously tested crisis escalation to real human services, firm boundaries against diagnosis and medical advice, and safeguards against dependency. Marrow's guide, Heron, is fenced by all of these. A safety check runs the instant you type, before any model call, and it is tuned to catch even indirect or metaphorical expressions of despair.
The app keeps the instructions short. The mechanism, source, and evidence label stay here when you want to inspect them.
Five minutes of cyclic sighing beat mindfulness meditation for mood and lowered breathing rate in a randomised trial. The second short inhale reinflates collapsed air sacs; the long exhale offloads carbon dioxide and slows the heart on the way out.
The mammalian dive reflex reliably slows heart rate, and cold on the face is used as a front-line distress tolerance skill in dialectical behaviour therapy. It overrules fight-or-flight with something older than thinking.
Urge surfing comes out of Marlatt relapse-prevention work and is a core practice in mindfulness-based relapse prevention, which has been studied in people in treatment with benefit at twelve months.
Labelling an emotion reduced amygdala response and raised prefrontal activity in brain imaging work. The naming itself does the work, which is why a rough word beats a perfect one.
This is long-standing practice in trauma and crisis work rather than a trial-tested protocol. It is included because it is used everywhere and costs nothing, and it is labelled practice rather than rounded up.
Alternating left-right stimulation is used as a resourcing technique inside EMDR. The evidence for the tapping on its own, separated from the rest of the therapy, is early, so it carries the promising label.
Imagining specific future events reduces how steeply people discount future rewards, and lowers craving in laboratory studies. Playing the tape forward is the same move in plain words.
Single bouts of exercise lower craving intensity across a number of trials and reviews, in tobacco and in other substances. The craving comes down while you are busy doing something else.
Slow breathing with a lengthened exhale raises heart rate variability and lowers physiological arousal. The heart slows on every breath out; making the out longer leans on that brake.
Cognitive defusion is a core process in Acceptance and Commitment Therapy, with experimental support for reducing how believable and how gripping a distressing thought feels.
Plans in the shape of "when X, then I will Y" raise follow-through substantially. Pooled across a large body of studies the effect is medium to large. Marrow adds the honest note that the effect is smaller in substance use specifically, and offers plans rather than demanding them.
Refusal skills training is a standard component of behavioural treatments for substance use, and those treatments hold up well in trials.
Slow breathing near six breaths a minute raised heart-rate variability and eased stress and anxiety across controlled studies. Ninety seconds of it is the short, structured version of the long exhale.
Slow, exhale-weighted breathing engages the parasympathetic nervous system. Making room for grief rather than suppressing it is long-standing practice in loss work, and the two sit well together.
Gradual, warmth-first re-entry is standard practice for freeze and shutdown states rather than a trial-tested protocol. It is included because demanding calm from a shut-down body does not work, and it is labelled practice plainly.
Alternating left-right stimulation is used as a resourcing technique inside EMDR. The evidence for the tapping on its own is early, and the extended two-minute version rests on the same early base, so it is labelled promising.
Visuospatial orienting and grounding are long-standing practice for dissociation and scattered states, rather than a trial-tested protocol. It is labelled practice rather than rounded up.
Marrow is informed by these approaches and affiliated with none of them. Nothing here is medical advice, and citing a study is not a claim that Marrow reproduces its intervention. Questions or corrections, from clinicians or anyone else, are genuinely welcome at hello@martori.studio.
Free, private, and built on the science above. No account, nothing to sign up for.
Open Marrow