Grounded in published work

The research behind Even

Published, peer-reviewed, and listed in full. Every entry was checked against PubMed before it was allowed in.

The same list is inside the app, one tap from anywhere it makes a claim. It is here as well because a person deciding whether to trust a stress app should be able to read its bibliography before installing it.

The rule the app is built around

Every claim has to trace back to a computed quantity. That is enforced by the type system rather than by discipline: there is no code path in Even from a written sentence to a statement on screen. A value carries where it came from and how confident it is; a comparison against a population cannot be constructed at all unless a real citation is attached to it; and the sentence a person reads is generated from the value, not stored next to it.

The test suite checks the consequences. Every template has to change its output when its input changes, which is the real test for horoscope writing, and no sentence may appear in more than half of two hundred generated profiles.

The nine

How Even learns what works for one person Klasnja and colleagues. Annals of Behavioral Medicine, 2019. Harvard University and the University of Michigan. doi:10.1093/abm/kay067

The right help at the right moment Nahum-Shani and colleagues. Annals of Behavioral Medicine, 2018. University of Michigan. doi:10.1007/s12160-016-9830-8

The twenty-minute walk Hunter and colleagues. Frontiers in Psychology, 2019. University of Michigan. doi:10.3389/fpsyg.2019.00722

Short sleep and attention Lim and Dinges. Psychological Bulletin, 2010. University of Pennsylvania. doi:10.1037/a0018883

Paced breathing Goessl and colleagues. Psychological Medicine, 2017. Boston University. doi:10.1017/S0033291717001003

The daily day rating (the NASA Task Load Index, adapted for whole days) Hernandez and colleagues. Ergonomics, 2021. University of Southern California. doi:10.1080/00140139.2021.2006317

Sleep regularity Windred and colleagues. Sleep, 2024. Monash University, on UK Biobank. doi:10.1093/sleep/zsad253 · what this one measures

Quick movement Weinstein and colleagues. Psychosomatic Medicine, 2024. George Mason University. doi:10.1097/PSY.0000000000001321

The evening wind-down (a small pilot) Hill and colleagues. Sleep Medicine, 2025. CQUniversity Australia. doi:10.1016/j.sleep.2025.02.043

Two more instruments are used under their own terms rather than as citations: the WHO-5 Well-Being Index, from the World Health Organization, in the first-run questions, and Apple's State of Mind model, which Even both reads and writes.

Even's credibility page, listing Harvard University, the University of Michigan, the University of Pennsylvania, Boston University, the World Health Organization and NASA, with one line each on what Even took from them — The same names, inside the app, on the third page of the first run.

What the evidence does not support

An app that only quotes its best numbers is doing marketing with footnotes. Three limits are worth stating plainly, and they are the reason Even is shaped the way it is.

The effects in this field are small

Across the meta-analytic literature on digital stress interventions, average effects are modest, and they are smaller again when measured physiologically rather than by questionnaire. Paced breathing is the strongest single family in the space and it is still not a large effect for everybody. What the literature does consistently show is that personalised guidance raises the ceiling, which is exactly the gap Even is built into: not another library of sessions, but a system that finds which of them work for one particular person.

Your HRV has no percentile here

Apple Watch samples heart rate variability opportunistically rather than over a fixed overnight window, and no published normative distribution describes that measurement. So Even cannot honestly tell you where your HRV sits among other people, and it does not. The axis speaks about you relative to yourself, and the code physically cannot construct the comparison, because the reference it would need does not exist.

Some things are not in the app

Exercise, daylight, seeing people you like, and professional care are all better supported than anything a phone can deliver in eight minutes. Even is allowed to name them, and does, including saying when a hard week is a see-a-friend problem rather than a breathing problem. It does not pretend to be the treatment.

Citing a study is not a claim that its authors or institutions endorse Even. It is a statement that the work is published, peer-reviewed, and available for you to read: every DOI above is a link.

NextThe Even score, and the five parts behind it AlsoSleep regularity, the flagship measure