The science behind every move Mags makes

The first post in this series covered why Mags is a text thread instead of an app. This one goes feature by feature: what each of Mags’ moves is actually doing, what evidence sits behind it, and (because trust is the whole product) an honest grade of how strong that evidence is. Strong means meta-analyses or multiple controlled trials. Emerging means peer-reviewed but young. First-principles means we derived it from the strong findings and our pilot exists to test it.

“When-ish?”: implementation intentions Strong

When you text Mags a vague intention (“I should call grandma sometime”) she asks exactly one question: when-ish? This is the most evidence-backed sentence in the product.

Psychologists call the underlying move an implementation intention: converting “I will do X” into “when situation Y happens, I will do X.” A meta-analysis of 94 studies found this one reframing improves goal attainment with an effect size of d = 0.65: a medium-to-large effect, which in behavior-change research is enormous for something this cheap (Gollwitzer & Sheeran, 2006). The mechanism matters for ADHD specifically: an if-then plan delegates initiation to the situation rather than to willpower, the same externalizing trick as the rest of Mags.

Body doubling and focus sprints Emerging

“Starting my taxes” arms a session: Mags checks in every twenty minutes until you’re through. This is body doubling (working alongside another presence) which the ADHD community discovered for itself years before research caught up. It’s now being studied seriously: researchers have mapped how and why it works across a continuum of settings (Eagle et al., ASSETS 2023), followed by a larger investigation with neurodivergent participants (Eagle & Ringland, ACM TACCESS) and even early EEG work (ASSETS 2025). The evidence is young but consistent: an accountable presence lowers the cost of starting and staying. Mags’ version adds what a human body double can’t promise: she’s available at 6am, never judges, and never cancels.

Focus sprints (twenty on, five off) ride the same session engine. Timeboxing with a concrete, arriving signal converts abstract duration into felt events (the time-perception findings from post one) rather than relying on a visual timer you have to remember to look at.

Tasks that shrink First-principles, honestly

When something bounces three times, Mags offers the ten-minute version, or a guilt-free toss. The shrinking half is standard clinical practice (micro-steps are a staple of ADHD coaching and CBT protocols for task paralysis). The novel part is automating the noticing: your snooze pattern is data, and it triggers the downshift without you having to confess you’re stuck. We derived this from the initiation research; our pilot measures whether it moves loop-completion rates. We’ll publish what we find either way.

The no-guilt rule: this one isn’t a vibe, it’s a finding Strong

Mags never sends streaks, scores, red badges, or “you missed…”, and this is a scientific position, not just a brand one.

Two findings drive it. First, emotion dysregulation is increasingly understood as a core feature of ADHD, not a comorbidity: present in a substantial share of adults and tied directly to how criticism and failure land (Shaw et al., American Journal of Psychiatry, 2014). Many adults with ADHD describe rejection-sensitive experiences intense enough that one shaming notification ends the relationship with a tool, which is exactly how it should end.

Second (and this is the counterintuitive one) self-criticism doesn’t even work as motivation. Controlled experiments show self-compassion increases self-improvement motivation relative to self-criticism: people who were induced to respond to their own failure kindly worked harder afterward, not less (Breines & Chen, Personality and Social Psychology Bulletin, 2012). The guilt-based design of most productivity software is not tough love. It’s just wrong about how motivation works.

So when Mags says “Tossed. Nest is lighter.” That’s the Breines & Chen finding, shipped as copy.

Waiting-on watch, rabbit-hole parking, “today is too much” First-principles

These are our own designs, derived from the strong findings: externalize the open loop (waiting-on), externalize the intrusive tangent so it stops competing for working memory (parking), and treat overwhelm as a state the system should absorb rather than penalize (load-shedding). Each is a small bet with a measurable target in our pilot: loop-completion rates, capture rates, and retention through high-stress weeks, respectively. The kill criteria are public inside the company: if a feature doesn’t earn its interruptions, it goes.

Why we grade our own homework

Every ADHD product page says “science-backed.” Few say which science, or admit which parts are extrapolation. Our view: this audience has been burned by eleven apps already, and the only durable moat we have is being the tool that never overclaims. Strong, emerging, first-principles: we’ll keep the labels updated as the literature (and our own pilot data) comes in.

Questions about any of this, or research we should be reading? Text Mags, or email the humans at Magpie & Co.