Sioul: an emergency pause for overwhelming moments¶

Research report, started 2026-10-06 13:08. Builds on capacity-budget.md (cited as "capacity report"), wellbeing-gain.md and done-for-the-day.md. Nothing here is built; this note researches an idea. A sibling note, global-pause.md, researches a different pause, taken to enjoy a good moment, which unlike this one moves the end of the working day.

The idea, as given: an "emergency pause", a panic button for overwhelming situations (sensory overstimulation, bad news coming in, and so on). It puts everything on hold; it may send an SMS (Android) or an email (desktop) to a contact chosen in advance, to say the current state or to ask for support; it puts the phone in do-not-disturb, except for people approved in advance; it locks the app's window with a comforting message ("we are holding the fort") while the person recovers; it does not extend the working day.

The question: how best to handle the crises that would lead someone to press it, so that Sioul takes the right measures and speaks well to the person meanwhile.

Grading, as in the capacity report: strong (meta-analysis, large or multi-site trial, guideline built on a systematic review), moderate (several consistent studies, or one large well-run study), weak (one small study, cross-sectional only, case series), opinion (consensus without data, essay, lived-experience writing, vendor documentation). Lived-experience and autistic-led sources are marked as such; they are the best source on what a state feels like from inside and what helps, and the weakest on how often or how much. Verification: "verified" = DOI resolved on Crossref, or PMID on PubMed, with matching authors, year and venue, on 2026-10-06. "not checked" = could not be confirmed; do not cite without checking. Platform facts come from official documentation, with the URL.

Answers so far¶

(Updated 2026-10-06, 15:55, final. Numbers point to the entries below; "P" to the design criteria P1–P31, and to the 17 open questions at the end.)

1. The states that lead to the button share three things and differ in three. In a meltdown or shutdown, a panic attack, the minutes after bad news, an ADHD surge, dissociation, a crash of energy or a suicidal crisis, thinking narrows (acute stress impairs working memory and flexibility, meta-analysis; 40–80% of medical information is forgotten at once, and less is kept the more is said), speech and choice may go (autistic shutdown, situational mutism), and less input helps. They differ in whether company or solitude helps, whether breathing exercises help, and how long recovery takes (minutes for panic, hours for overload, days for a crash). So everything the pause does is set beforehand by the person, and the press asks nothing (P1). A suicidal crisis must be handled safely even when the pause was set up for noise: autistic adults and adults with ADHD carry a much higher risk, a crisis can peak within ten minutes, safety plans cut suicidal acts (RR 0.57) but not thoughts, and risk cannot be predicted (1.7). The emergency number and crisis line stay on the screen; Sioul never guesses risk, never calls anyone on its own, never asks for a promise (P13, P20, P30). Numbers checked: France 3114, 15, 112 (and 114 for deaf, hard-of-hearing and aphasic people); EU 112; UK 116 123, 999 (and 111, Shout 85258); US 988, 911; Canada 988, 911 (1.7k).

2. What helps in the moment. Trials: slow breathing lowers stress a little when practised (g ≈ −0.35) and is mixed as an in-the-moment tool; distraction regulates emotion and focusing on the feeling worsens it (meta-analysis of 306 experiments); psychological first aid gives immediate relief without preventing PTSD (one RCT). Consensus: WHO's and NCTSN's first aid, Hobfoll's five elements (safety, calm, self-efficacy, connectedness, hope), less stimulation, grounding, and for autistic people no questions, no instructions, no new information, and time. Folklore: the "5-4-3-2-1" exercise as a tested technique (no trial found), any one breathing pattern as the cure for panic (raising or lowering CO2 worked equally well), "just breathe". Breath focus can worsen panic and trauma reactions for some (breathing retraining added nothing to panic treatment; lasting meditation side effects in 6–14%), so the breathing guide is optional and off by default (2, P16).

3. What harms. Single-session debriefing (Cochrane: no benefit, PTSD more likely at one year in one trial), pressure to talk (people who did not express a reaction after 9/11 did better), "calm down" (trying to relax under load raises arousal), invalidation, expected positivity and affirmations, false reassurance, choices, countdowns, streaks of calm, and being watched or tracked (3). The exit asks nothing; nothing is counted (P23–P26).

4. Talking through a screen. State facts about Sioul, never claims about the person: "Paused. Sioul is keeping everything on hold. Nothing new will show here until you come back." At most about 40 words, literal, no "we", no idiom (so the brief's "we are holding the fort" becomes literal), no question, no praise, no exclamation mark, nothing moving unless started; the trauma-informed principles become concrete checks (4.1–4.3). Candidate texts in English and French are in 4.5, the supporter message in 5.4.

5. Alerting a trusted person. Support buffers stress, best when practical and low-key (5.1). Crisis plans made when well, in the person's words, and known to supporters cut compulsory admissions by about a quarter, and fail when supporters do not use them (5.2). A good message says it was prepared in advance, the state in the person's words, what helps, what not to do, when they will be in touch; text rather than call by default (5.3–5.4). Risks: accidental sends (Android's SOS caused waves of silent emergency calls, and devices that decide by themselves raise more false alarms than true ones: a short silent undo window, and pausing never requires telling anyone, since people leave alarm buttons unpressed for fear of bothering someone), supporter burden (ask them first, give them a note), an abusive contact (a supporter link is dual-use: no status, history, location or remote control), and nobody answering (no automatic escalation, other options visible) (5.6). Existing apps (MYPLAN, Brain in Hand, BackUp, Virtual Hope Box, PTSD Coach) teach: own words, agreed supporters, no surveillance, a correct "help now" path, nothing to unlock or answer; no study has tested a stand-alone alert button (5.7).

6. After. Re-entry lightened at once (the bad-day level, about 60 %), held items back slowly, doses never silenced without the person's choice, no reflection prompt, an optional note later; professional help suggested only after repeated pauses, practically and once, never on the exit screen (6).

7. Platform. Android: a "Pause" do-not-disturb mode owned by Sioul (required style since Android 15), letting through starred contacts, repeat callers and alarms, with no restricted permission; SMS sent directly needs SEND_SMS, which Android 15 hides behind "restricted settings" for sideloaded APKs, or a pre-filled SMS with one tap; Matrix and mail can send without a tap (7.1–7.2). Desktops: Plasma lets an app inhibit notifications cleanly; GNOME only through the person's own switch; macOS through a shortcut the person makes; Windows not without Microsoft's approval. Sioul can always quiet itself and must say what it could not silence (7.5).

8. Legal. Not a medical device if Sioul only stores and sends the person's message and its words claim no detection, prevention or treatment (MDR, MDCG 2019-11, MHRA, FDA, ANSM agree). The message is health data, but sent by the person to people they chose it is household correspondence; with no server or telemetry, the developer is outside the GDPR (CNIL 2024). Consent is made explicit at setup and by the press, revocable on the same screen, and never stretched into sending without a press (8).

Main uncertainties. No study tested an in-the-moment pause or alert button; the wording, the undo window, the breathing default, the threshold for suggesting help and the choice of SMS mode need testing with the people concerned (open questions).

1. The states that lead there¶

The button can be pressed in states that differ a lot: an autistic meltdown or shutdown, a panic attack, the minutes after a bad letter, an ADHD surge of frustration, dissociation, a crash of energy, a suicidal crisis. What most of them share, and what a screen has to allow for:

• Thinking narrows. Acute stress impairs working memory and cognitive flexibility (meta-analysis, 1.3c); attention narrows on the threat, and what is said around it is lost (1.3d). Decisions become hard or impossible (1.1f, 1.3e).
• Speech and action may go. In shutdowns, speech and movement can stop (1.1b, 1.1f); speaking autistic adults can lose speech for a while (1.1g).
• Less input helps nearly all of them. Removing noise, light, people, questions and instructions is the one measure every state below tolerates.

What differs, and why the pause must be set in advance by the person rather than chosen in the moment:

• Company or solitude. Many autistic adults go somewhere alone to avoid harm during a meltdown (1.1a), while some autistic youths wanted an adult or a pet and felt imposed isolation as punishment (1.1c); in a suicidal crisis, contact with others is protective (1.7).
• Breathing exercises. Helpful for some, neutral or worse for others (1.2, 2.5).
• Time course. A panic attack lasts minutes (1.2a); an acute stress reaction, hours to days (1.3a); a shutdown or meltdown leaves a recovery time of hours (1.1e); a crash of energy, days (1.6a); a suicidal crisis can peak within minutes (1.7e).

1.1 Autistic meltdown, shutdown and sensory overload¶

1.1a Lewis LF, Stevens K. 2023. The lived experience of meltdowns for autistic adults. Autism 27(6):1817–1825. doi:10.1177/13623613221145783 — verified (Crossref, PubMed 36632658).

• 32 autistic adults described their meltdowns in writing. Most felt overwhelmed by information, senses, and social and emotional stress; they felt extreme emotions (anger, sadness, fear) and had trouble thinking and remembering during the meltdown; they tried to stay in control and often felt "not themselves". Many tried to keep away from triggers, or made sure they were alone when a meltdown was coming, to avoid harm to their bodies, their emotions and their relationships.
• Qualitative, lived experience. Weak to moderate (one study, self-selected sample), strong as a description from the inside.
• Takeaway: being alone, away from input, is a protective strategy the person chooses; the pause should allow it without asking for any explanation, and should not count on the person reading or remembering much while it lasts.

1.1b Paris K, Lodestone AZ, Houser M, Lewis LF. 2025 (issue 2026). "Shutdowns are like you're stuck on the blue screen of death": a metaphor analysis of autistic shutdowns. Autism in Adulthood 8(4):687–697. doi:10.1089/aut.2024.0193 — verified (Crossref, PubMed 42383265).

• Participatory secondary analysis of two studies (86 autistic adults). Six metaphors: being frozen, a computer crash, going inside myself, when I can't keep up, survival mode, playing a role. Shutdowns were described as being stuck, "often involving physical and/or vocal immobilization", and as a response to perceived threat.
• Qualitative, participatory. Weak to moderate.
• Takeaway: during a shutdown the person may not be able to type, speak or choose; anything the pause needs from them must have been decided before, and the screen must ask for nothing.

1.1c Phung J, Penner M, Pirlot C, Welch C. 2021. What I wish you knew: insights on burnout, inertia, meltdown, and shutdown from autistic youth. Frontiers in Psychology 12:741421. doi:10.3389/fpsyg.2021.741421 — verified (Crossref; full text read).

• Interviews with 8 autistic children and youths (8–18), one autistic co-author. Meltdowns came from accumulated exhaustion plus extra demands, unexpected changes and overstimulation, with tunnel vision and fuzzy memory. What helped differed: play or sport, talking with a trusted adult, a hug or a pet for some; being alone for others. Harsh words ("you're acting like a little kid") made it worse; isolation imposed by adults felt like punishment ("they're literally barricading me in"). Shutdown: "a heavy blanket weighing 500 pounds"; "I can't really talk, I'll stutter, I'll like literally freeze up".
• Qualitative, small, young people. Weak; consistent with 1.1a–b.
• Takeaway: whether contact or solitude helps differs between people; the pause must follow the person's own prior choice and never impose either. Wording that talks down to the person makes things worse.

1.1d MacLennan K, O'Brien S, Tavassoli T. 2022. In our own words: the complex sensory experiences of autistic adults. J Autism Dev Disord 52(7):3061–3075. doi:10.1007/s10803-021-05186-3 — verified (Crossref, PubMed 34255236).

• Mixed-methods survey co-produced with stakeholders, 49 autistic adults: sensory hyper- and hyporeactivity and seeking across all senses; themes "outcomes", "control", "tolerance and management" and "the role of other people".
• Mixed methods, small. Weak to moderate.
• Takeaway: control over one's sensory input is central; the pause should remove what Sioul itself adds (sound, vibration, light, motion, new items) and not add anything of its own.

1.1e National Autistic Society. 2020. Meltdowns: a guide for all audiences. https://www.autism.org.uk/advice-and-guidance/topics/behaviour/meltdowns/all-audiences (last reviewed 14 August 2020; read 2026-10-06).

• "Give them some time – it can take a while to recover from information or sensory overload"; ask if they are OK, "but bear in mind they'll need more time to respond than you might expect"; "make space", ask people "not to stare", turn off loud music, turn down bright lights, "whatever you can think of to reduce the information overload"; warning signs come first (the "rumble stage"); "have a plan beforehand"; communicate in "short, clear sentences", written information, instant messaging. Among calming activities it lists "taking deep breaths" (compare 1.2 and 2.5).
• Charity guidance, not a systematic review. Opinion (consensus).
• Takeaway: time, less input, no staring (no one watching, 3.7), a plan made beforehand, short written sentences.

1.1f Leicestershire Partnership NHS Trust, Autism Space. What are autistic meltdowns and shutdowns? https://leicspart.nhs.uk/autism-space/health-and-lifestyle/meltdowns-and-shutdowns (read 2026-10-06; made with a lived-experience partner; quotes the CAPTAP training material).

• A shutdown can bring "situational mutism (where one is unable to communicate or speak)", wanting "to hide away somewhere dark and alone", no energy, and "making decisions becomes impossible". During a meltdown or shutdown: "Avoid asking questions, giving information or instructions, or using restrictive methods. Any of these would likely make the situation worse." Needs are best discussed "when they're calm. That way you can agree a plan". Time, a quiet space, removing what overloads. It also offers a breathing pattern (in 3, hold 3, out 6) as something that "can also help".
• NHS patient information with lived-experience input. Opinion (consensus).
• Takeaway: the clearest statement of the "no questions, no information, no instructions" rule; the plan is made when calm.

1.1g Zisk AH, Dalton E. 2019. Augmentative and alternative communication for speaking autistic adults: overview and recommendations. Autism in Adulthood 1(2):93–100. doi:10.1089/aut.2018.0007 — verified (Crossref, PubMed 36601528). Autistic-authored.

• "Autistic people who use speech may experience intermittent, unreliable, and/or insufficient speech"; speech varies with environment and content; speaking autistic adults use text messaging, chat and handwriting as communication aids; "communication—rather than speech—should be prioritized".
• Review of community knowledge, autistic-led. Opinion to weak.
• Takeaway: a pause that relies on a phone call, from or to the person, fails exactly when speech goes; every channel must work in writing.

1.1h Howard PL, Sedgewick F. 2021. "Anything but the phone!": communication mode preferences in the autism community. Autism 25(8):2265–2278. doi:10.1177/13623613211014995 — verified (Crossref, PubMed 34169750).

• Survey of autistic adults across scenarios: with unknown people or organisations, email was preferred and phone calls "very unpopular"; with friends and family, face-to-face and written forms (email, text messages) were preferred. Services should not depend on the phone and should offer written options.
• Survey. Moderate for preferences.
• Takeaway: the message to a supporter should ask for a text reply rather than a call, unless the person said otherwise (5.4).

1.1i Kinnaird E, Stewart C, Tchanturia K. 2019. Investigating alexithymia in autism: a systematic review and meta-analysis. Eur Psychiatry 55:80–89. doi:10.1016/j.eurpsy.2018.09.004 — verified (Crossref, PubMed 30399531).

• 15 studies with the Toronto Alexithymia Scale: alexithymia (difficulty identifying and describing one's feelings) in 49.9% of autistic participants against 4.9% of non-autistic ones; common, not universal.
• Meta-analysis of self-report. Moderate.
• Takeaway: "how are you feeling?" is a hard question for about half of autistic people even when calm; the pause must not ask it, and any state sent to a supporter must be pre-written, not described on the spot.

1.1j Autistic burnout as background. Raymaker DM, Teo AR, Steckler NA, et al. 2020. "Having all of your internal resources exhausted beyond measure and being left with no clean-up crew": defining autistic burnout. Autism in Adulthood 2(2):132–143. doi:10.1089/aut.2019.0079 — verified (Crossref, PubMed 32851204). Higgins JM, Arnold SR, Weise J, Pellicano E, Trollor JN. 2021. Defining autistic burnout through experts by lived experience: grounded Delphi method investigating #AutisticBurnout. Autism 25(8):2356–2369. doi:10.1177/13623613211019858 — verified (Crossref, PubMed 34088219).

• Raymaker (community-based participatory research, 19 interviews and 19 internet sources): chronic exhaustion, loss of skills and "reduced tolerance to stimulus", from cumulative load and barriers to relief; recovery came with acceptance and social support, "time off/reduced expectations" and unmasking; burnout was linked to suicidal behaviour. Higgins (Delphi led by autistic experts by experience): exhaustion, withdrawal, problems with thinking, reduced daily living skills; "psychological treatments for depression potentially could make autistic burnout worse".
• Qualitative and Delphi, autistic-led or participatory. Moderate as definitions.
• Takeaway: a meltdown or shutdown often happens on top of burnout; lowering tolerance to stimulus is part of the state, so the pause's after-care (section 6) is about reduced expectations over days, not only the next hour.

1.2 Panic attacks¶

1.2a NHS. Panic disorder. https://www.nhs.uk/mental-health/conditions/panic-disorder/ (read 2026-10-06).

• "Most panic attacks last between 5 and 20 minutes. Some have been reported to last up to an hour." Symptoms include racing heart, chest pain, shortness of breath, faintness, numbness, and "feeling like you're not connected to your body". "Most of these symptoms can also be symptoms of other conditions or problems, so you may not always be experiencing a panic attack."
• Patient information from a national health service. Opinion (consensus), sound for the time course.
• Takeaway: the pause will usually outlast a panic attack; the screen must never tell the person what they are having ("this is a panic attack"), since chest pain or breathlessness may have another cause, and the emergency number must stay visible (1.7).

1.2b Breathing retraining does not add to panic treatment, and may hinder some. Schmidt NB, Woolaway-Bickel K, Trakowski J, et al. 2000. Dismantling cognitive-behavioral treatment for panic disorder: questioning the utility of breathing retraining. J Consult Clin Psychol 68(3):417–424. doi:10.1037/0022-006X.68.3.417 — verified (Crossref, PubMed 10883558). Meuret AE, Wilhelm FH, Ritz T, Roth WT. 2003. Breathing training for treating panic disorder: useful intervention or impediment? Behav Modif 27(5):731–754. doi:10.1177/0145445503256324 — verified (Crossref, PubMed 14531164). Kim S, Wollburg E, Roth WT. 2012. Opposing breathing therapies for panic disorder: a randomized controlled trial of lowering vs raising end-tidal PCO2. J Clin Psychiatry 73(7):931–939. doi:10.4088/JCP.11m07068 — verified (Crossref; abstract from the publisher's page).

• Schmidt (RCT, 77 patients): CBT with or without breathing retraining; "some data suggested that the addition of BR yielded a poorer outcome", the results being "generally more consistent with treatment equivalence"; the authors' hypothesis was that breathing retraining is "a more attractive (but less adaptive) option for some patients". Meuret (review of 9 studies): the evidence is not compelling either way, and techniques "must take account of respiration rate and tidal volume", since both set blood CO2. Kim (RCT, 74 patients): training to raise CO2 and training to lower it both reduced panic severity, with large effects at one month — the direction did not matter, which points to non-specific effects (attention, practice, expectation).
• RCTs and review. Moderate.
• Takeaway: a breathing exercise is not a specific remedy for panic. It may help as something to do; it should be one option, never the default response or an instruction, and never "take a deep breath" (larger breaths lower CO2, which is what over-breathing already does; Meuret's rationale).

1.2c Safety behaviour: escape can maintain fear, or help. Helbig-Lang S, Petermann F. 2010. Tolerate or eliminate? A systematic review on the effects of safety behavior across anxiety disorders. Clin Psychol Sci Pract 17(3):218–233. doi:10.1111/j.1468-2850.2010.01213.x — verified (Crossref; abstract not read). Rachman S, Radomsky AS, Shafran R. 2008. Safety behaviour: a reconsideration. Behav Res Ther 46(2):163–173. doi:10.1016/j.brat.2007.11.008 — verified (Crossref, PubMed 18199423).

• Safety behaviours (escape, avoidance, protective rituals) can interfere with exposure therapy, which is why they are often called anti-therapeutic; Rachman et al. review experiments in which escape and safety aids helped fear reduction, and argue for their "judicious use", especially early on.
• Reviews of experimental and clinical studies. Moderate; the field is divided.
• Related trial: Guth M, Wiebe A, Ekhlas-Schumann M, et al. 2025. Mobile app-guided exposure therapy for panic disorder with and without agoraphobia: randomized controlled trial. J Med Internet Res 27:e76389. doi:10.2196/76389 — verified (Crossref). 111 adults over 5 weeks: an exposure app beat the waitlist (d = 0.55); reliable improvement in 35% with exposure against 6% with a general mindfulness app, whose users improved at first and then worsened again on four of six measures by follow-up. No RCT of a calming or SOS button used during attacks was found.
• Takeaway: for someone with recurrent panic, a pause used at every surge could become avoidance; Sioul is not a treatment and must not present the pause as the way to handle panic. When it is used often for the same reason, Sioul may mention once, later and never on the exit screen, that panic responds well to treatment (6.4).

1.3 Acute stress after bad news¶

1.3a ICD-11, QE84 Acute stress reaction. World Health Organization, ICD-11 for Mortality and Morbidity Statistics (classified among "factors influencing health status", not as a disorder). Text as reproduced at https://findacode.com/icd-11/code-505909942.html; WHO's own browser not readable without scripts (not checked at the WHO page).

• "Transient emotional, somatic, cognitive, or behavioural symptoms" after an extremely threatening or horrific event, "considered to be normal given the severity of the stressor", usually beginning to subside "within a few days after the event or following removal from the threatening situation".
• Classification (consensus). Opinion level as evidence, authoritative as a definition.
• Takeaway: a strong reaction to bad news is a normal response, not an illness; the pause should not medicalise it, and removal from the source (the letter put away, the inbox closed) is part of what lets it subside. Most bad news in admin mail is below ICD-11's "extremely threatening" threshold; the same logic applies with a shorter course.

1.3b Dickerson SS, Kemeny ME. 2004. Acute stressors and cortisol responses: a theoretical integration and synthesis of laboratory research. Psychol Bull 130(3):355–391. doi:10.1037/0033-2909.130.3.355 — verified (Crossref, PubMed 15122924).

• Meta-analysis of 208 laboratory studies: tasks that were uncontrollable or carried social-evaluative threat (being judged) raised cortisol; tasks with both brought "the largest cortisol and adrenocorticotropin hormone changes and the longest times to recovery".
• Meta-analysis. Strong.
• Takeaway: what prolongs stress is lack of control and being judged. The pause gives control back (nothing new comes in, nothing is due) and removes evaluation (no one sees, nothing asked, nothing scored).

1.3c Shields GS, Sazma MA, Yonelinas AP. 2016. The effects of acute stress on core executive functions: a meta-analysis and comparison with cortisol. Neurosci Biobehav Rev 68:651–668. doi:10.1016/j.neubiorev.2016.06.038 — verified (Crossref, PubMed 27371161).

• Acute stress impaired working memory and cognitive flexibility; effects on inhibition were mixed.
• Meta-analysis. Strong.
• Takeaway: during and shortly after a crisis, the person can hold little in mind and switch poorly between options; one thing per screen, no choices, nothing to remember.

1.3d Kessels RPC. 2003. Patients' memory for medical information. J R Soc Med 96(5):219–222. doi:10.1177/014107680309600504 — verified (Crossref, PubMed 12724430; full text read on PMC).

• "40-80% of medical information provided by healthcare practitioners is forgotten immediately. The greater the amount of information presented, the lower the proportion correctly recalled; furthermore, almost half of the information that is remembered is incorrect." Under stress, "attentional narrowing": the central message takes all attention and peripheral information is not stored.
• Narrative review. Moderate.
• Takeaway: whatever the pause screen says must fit in a glance; anything important (how to come back, who was told) must stay visible rather than be said once.

1.3e Starcke K, Brand M. 2012. Decision making under stress: a selective review. Neurosci Biobehav Rev 36(4):1228–1248. doi:10.1016/j.neubiorev.2012.02.003 — verified (Crossref, PubMed 22342781).

• Stress affects decision making; whether for better or worse depends on the task and on moderating variables.
• Selective review. Moderate.
• Takeaway: no decision with lasting effects (replying, paying, deleting, cancelling appointments) should be asked for during the pause; what can wait is deferred, and what cannot is handled by the plan set beforehand.

1.4 Overwhelm and emotional dysregulation in ADHD¶

1.4a Beheshti A, Chavanon ML, Christiansen H. 2020. Emotion dysregulation in adults with attention deficit hyperactivity disorder: a meta-analysis. BMC Psychiatry 20:120. doi:10.1186/s12888-020-2442-7 — verified (Crossref, PubMed 32164655). Shaw P, Stringaris A, Nigg J, Leibenluft E. 2014. Emotion dysregulation in attention deficit hyperactivity disorder. Am J Psychiatry 171(3):276–293. doi:10.1176/appi.ajp.2013.13070966 — verified (Crossref, PubMed 24480998).

• Beheshti: 13 studies, N = 2535: adults with ADHD had much more emotion dysregulation than controls (Hedges' g = 1.17; emotional lability g = 1.20); dysregulation correlated with ADHD symptom severity (r = .54). Shaw: emotion dysregulation is "prevalent in ADHD throughout the lifespan and is a major contributor to impairment".
• Meta-analysis of case-control studies; review. Strong for the association.
• Takeaway: intense surges of frustration, anger or distress are part of ADHD for many adults, not a rare crisis; the pause should be cheap to use and to leave, without drama.

1.4b Skirrow C, Ebner-Priemer U, Reinhard I, et al. 2014. Everyday emotional experience of adults with attention deficit hyperactivity disorder: evidence for reactive and endogenous emotional lability. Psychol Med 44(16):3571–3583. doi:10.1017/S0033291714001032 — verified (Crossref, PubMed 25066432).

• Ambulatory assessment, 41 men with ADHD and 47 controls, eight reports a day for a working week: more frequent bad events, more intense and unstable irritability and frustration, more intense anger; bad events weighed more in the ADHD group, but did not explain all of the difference.
• Experience sampling, small. Moderate.
• Takeaway: in ADHD the surge often follows an event (a demand, a setback, a bad message) and is reactive; removing the incoming stream for a while is a direct answer. "Rejection sensitive dysphoria", a widely used clinical and community term, has no validated measure or trial behind it that this search found; it is not used here.

1.5 Dissociation¶

1.5a Ozer EJ, Best SR, Lipsey TL, Weiss DS. 2003. Predictors of posttraumatic stress disorder and symptoms in adults: a meta-analysis. Psychol Bull 129(1):52–73. doi:10.1037/0033-2909.129.1.52 — verified (Crossref, PubMed 12555794).

• 68 studies, seven predictors; peritraumatic dissociation (dissociation during or right after the event) was the strongest (weighted r = .35), prior characteristics the weakest (r = .17).
• Meta-analysis. Strong for the association.
• Takeaway: dissociating at the time of bad news is a marker that the event hit hard; Sioul can say plainly, once and later, where to find help if feeling unreal or numb goes on for days (6.4), without alarm.

1.5b International Society for the Study of Trauma and Dissociation. 2011. Guidelines for treating dissociative identity disorder in adults, third revision. J Trauma Dissociation 12(2):115–187, doi:10.1080/15299732.2011.537247 — verified (Crossref; summary version 12(2):188–212, doi:10.1080/15299732.2011.537248). Brand BL, Schielke HJ, Putnam K, et al. 2025. A randomized controlled trial assists individuals with complex trauma and dissociation in Finding Solid Ground. Psychol Trauma 17(8):1717–1727. doi:10.1037/tra0001871 — verified (Crossref, PubMed 40014495).

• The guidelines place grounding (orienting to the present through the senses and the surroundings) in the first, stabilisation phase of treatment, by expert consensus. Brand et al. (interim report of an RCT, 291 outpatients with trauma-related dissociation): an online psychoeducation programme, which teaches grounding among other skills, added to therapy improved emotion regulation, PTSD symptoms and self-compassion against a 6-month waitlist.
• Consensus guideline (opinion); one RCT of a whole programme (moderate for the programme, nothing for grounding alone in the moment).
• Takeaway: grounding is consensus practice with no trial of its own as an in-the-moment tool; offer it as a short concrete option (2.4), never as an exercise to complete.

1.6 Burnout, and a crash of energy (post-exertional malaise)¶

1.6a NICE NG206. 2021. Myalgic encephalomyelitis (or encephalopathy)/chronic fatigue syndrome: diagnosis and management. https://www.nice.org.uk/guidance/ng206 (recommendations read 2026-10-06).

• Post-exertional malaise: worsening "that can follow minimal cognitive, physical, emotional or social activity"; symptoms "can typically worsen 12 to 48 hours after activity and last for days or even weeks". For a flare-up: identify possible triggers, temporarily reduce activity, and do "not [return] to usual activity levels until the flare‑up has resolved" (1.14.5). People may have trouble with "screens, sound and light sensitivity", headaches affecting reading, or "brain fog affecting their concentration". The time to return to previous activity "varies from person to person" (1.11.16).
• National guideline (systematic reviews, mostly low-quality evidence, with committee consensus). Strong as guidance.
• Takeaway: a crash is measured in days; pressing pause because energy has run out calls for lighter days ahead, not only a quiet hour (section 6). The crisis itself is an emotional exertion that can bring a delayed crash. The pause screen should be dim, still and readable at a glance, or not need reading at all.

1.6b Autistic burnout: see 1.1j. Recovery came with "time off/reduced expectations" (Raymaker 2020).

1.7 Suicidal crisis¶

The button may be pressed in a suicidal crisis, by someone who set it up for overload or bad news. Autistic adults and adults with ADHD carry a much higher risk than others (1.7f), and autistic burnout has been linked to suicidal behaviour (1.1j). What follows is what the pause can and cannot do there.

1.7a The Safety Planning Intervention. Stanley B, Brown GK. 2012. Safety planning intervention: a brief intervention to mitigate suicide risk. Cogn Behav Pract 19(2):256–264. doi:10.1016/j.cbpra.2011.01.001 — verified (Crossref). Steps as listed in the trial record NCT04230434 (https://clinicaltrials.gov/study/NCT04230434) and in practice summaries.

• A written plan made with a clinician, in the person's own words, kept by the person, used in order: (1) recognise one's own warning signs; (2) internal coping strategies that take the mind off suicidal thoughts; (3) people and places that provide distraction; (4) people who can help resolve the crisis; (5) professionals and urgent services to contact; (6) making the environment safer (lethal means). It replaced the "no-suicide contract".
• Clinical method. Its trials follow.
• Takeaway: the order matters: what the person can do alone, then people, then professionals, then emergency services. A pause screen that offers "your plan" should show the person's own words in that order, one step at a time.

1.7b Nuij C, van Ballegooijen W, de Beurs D, et al. 2021. Safety planning-type interventions for suicide prevention: meta-analysis. Br J Psychiatry 219(2):419–426. doi:10.1192/bjp.2021.50 — verified (Crossref, PubMed 35048835).

• 6 studies, 3536 participants: relative risk of suicidal behaviour 0.57 (95% CI 0.41–0.80; number needed to treat 16) with a safety plan against control; no significant effect on suicidal ideation.
• Meta-analysis (few studies, mixed designs). Moderate to strong for suicidal behaviour.
• Takeaway: a safety plan reduces suicidal acts, not suicidal thoughts. The pause should not promise to make the feeling go; it can make the next safe step easy to reach.

1.7c Stanley B, Brown GK, Brenner LA, et al. 2018. Comparison of the Safety Planning Intervention with follow-up vs usual care of suicidal patients treated in the emergency department. JAMA Psychiatry 75(9):894–900. doi:10.1001/jamapsychiatry.2018.1776 — verified (Crossref, PubMed 29998307).

• Cohort comparison (not randomised), 9 US veterans' emergency departments, 1640 patients: safety plan plus at least two follow-up phone calls went with fewer suicidal behaviours over 6 months (3.03% against 5.29%; odds ratio 0.56) and twice the odds of attending outpatient care.
• Large non-randomised comparison. Moderate.
• Takeaway: the plan was tested together with human follow-up; a plan alone in an app is a different thing, with no trial of its own on suicidal acts (5.7).

1.7d Bryan CJ, Mintz J, Clemans TA, et al. 2017. Effect of crisis response planning vs. contracts for safety on suicide risk in U.S. Army soldiers: a randomized clinical trial. J Affect Disord 212:64–72. doi:10.1016/j.jad.2017.01.028 — verified (Crossref, PubMed 28142085). Rudd MD, Mandrusiak M, Joiner TE. 2006. The case against no-suicide contracts: the commitment to treatment statement as a practice alternative. J Clin Psychol 62(2):243–251. doi:10.1002/jclp.20227 — verified (Crossref).

• RCT, 97 soldiers at an emergency behavioural health appointment: a crisis response plan (a similar written plan) against a contract for safety: attempts over 6 months 5% against 19% (hazard ratio 0.24, a 76% reduction), faster decline of suicidal ideation, fewer inpatient days. Rudd et al.: no-suicide contracts have no evidence of effect and can mislead clinicians.
• One small RCT (moderate); review (opinion to moderate).
• Takeaway: the pause must never ask the person to promise anything ("promise you'll be safe"); it offers steps, not a commitment.

1.7e A suicidal crisis can be short. Deisenhammer EA, Ing C-M, Strauss R, et al. 2009. The duration of the suicidal process: how much time is left for intervention between consideration and accomplishment of a suicide attempt? J Clin Psychiatry 70(1):19–24. doi:10.4088/JCP.07m03904 — verified (Crossref, PubMed 19026258). Simon OR, Swann AC, Powell KE, et al. 2001. Characteristics of impulsive suicide attempts and attempters. Suicide Life Threat Behav 32(1 Suppl):49–59. doi:10.1521/suli.32.1.5.49.24212 — verified (Crossref, PubMed 11924695).

• Deisenhammer (82 people interviewed within 3 days of an attempt): 47.6% said 10 minutes or less passed between the first current thought of suicide and the attempt; 76.8% had had some interpersonal contact during that time, so "there is at least some readiness for interpersonal contact". Simon (153 nearly lethal attempts, ages 13–34): 24% had spent less than 5 minutes between decision and attempt.
• Interview and case-control studies, recall after the event. Moderate.
• Takeaway: whatever the pause offers for a suicidal crisis must be reachable in seconds and without reading: the crisis line and the emergency number visible on the pause screen itself, the person's chosen contact one tap away. Contact with someone close is plausible even then.

1.7f Who is at risk. Cassidy S, Bradley P, Robinson J, et al. 2014. Suicidal ideation and suicide plans or attempts in adults with Asperger's syndrome attending a specialist diagnostic clinic: a clinical cohort study. Lancet Psychiatry 1(2):142–147. doi:10.1016/S2215-0366(14)70248-2 — verified (Crossref, PubMed 26360578). Cassidy S, Bradley L, Shaw R, Baron-Cohen S. 2018. Risk markers for suicidality in autistic adults. Mol Autism 9:42. doi:10.1186/s13229-018-0226-4 — verified (Crossref, PubMed 30083306). Septier M, Stordeur C, Zhang J, et al. 2019. Association between suicidal spectrum behaviors and attention-deficit/hyperactivity disorder: a systematic review and meta-analysis. Neurosci Biobehav Rev 103:109–118. doi:10.1016/j.neubiorev.2019.05.022 — verified (Crossref, PubMed 31129238). Camm-Crosbie L, Bradley L, Shaw R, Baron-Cohen S, Cassidy S. 2019. "People like me don't get support": autistic adults' experiences of support and treatment for mental health difficulties, self-injury and suicidality. Autism 23(6):1431–1441. doi:10.1177/1362361318816053 — verified (Crossref, PubMed 30497279).

• Cassidy 2014 (374 adults newly diagnosed): 66% reported lifetime suicidal ideation (odds ratio 9.6 against a general UK sample), 35% plans or attempts. Cassidy 2018 (survey designed with autistic adults; 164 autistic and 169 other adults): 72% of autistic adults above the cut-off for suicide risk against 33%; camouflaging and the number of unmet support needs predicted suicidality. Septier 2019 (57 studies): ADHD went with suicide attempts (OR 2.37), ideation (3.53), plans (4.54) and death by suicide (6.69). Camm-Crosbie 2019 (200 autistic adults): difficulties accessing support, lack of understanding of autistic people by services.
• Cohort, surveys, meta-analysis of observational studies. Strong for raised risk.
• Takeaway: for the people Sioul is made for, a suicidal crisis is not an edge case; the pause must be safe in one even when it was set up for something else. Services built around the phone fail many of them (1.1h), so written channels (Shout 85258, 988 by text, the person's own contact by text) matter.

1.7g Risk cannot be predicted, and must not be guessed by software. Franklin JC, Ribeiro JD, Fox KR, et al. 2017. Risk factors for suicidal thoughts and behaviors: a meta-analysis of 50 years of research. Psychol Bull 143(2):187–232. doi:10.1037/bul0000084 — verified (Crossref, PubMed 27841450). NICE NG225. 2022. Self-harm: assessment, management and preventing recurrence. https://www.nice.org.uk/guidance/ng225 (read 2026-10-06). Barnett I, Torous J. 2019. Ethics, transparency, and public health at the intersection of innovation and Facebook's suicide prevention efforts. Ann Intern Med 170(8):565–566. doi:10.7326/M19-0366 — verified (Crossref).

• Franklin (365 studies): prediction from risk factors was "only slightly better than chance for all outcomes" and has not improved in 50 years. NICE: "Do not use risk assessment tools and scales to predict future suicide or repetition of self-harm" (1.6.1); a safety plan should be "held by the person", "shared … as decided by the person" and "accessible … at times of crisis" (1.11.8). Barnett & Torous: Facebook's algorithmic detection of suicide risk, which could lead to police wellness checks, ran without transparency, consent or evaluation.
• Meta-analysis (strong); national guideline (strong); commentary (opinion).
• Takeaway: Sioul must not infer risk from what the person does or writes, must not escalate on its own (no automatic call to emergency services, no message the person did not set), and must keep any plan in the person's hands. A press of the button is not a risk signal to act on; it is the person's own act.

1.7h Crisis lines, by voice and by text. Gould MS, Kalafat J, Harris-Munfakh JL, Kleinman M. 2007. An evaluation of crisis hotline outcomes. Part 2: suicidal callers. Suicide Life Threat Behav 37(3):338–352. doi:10.1521/suli.2007.37.3.338 — verified (Crossref, PubMed 17579545). Hoffberg AS, Stearns-Yoder KA, Brenner LA. 2020. The effectiveness of crisis line services: a systematic review. Front Public Health 7:399. doi:10.3389/fpubh.2019.00399 — verified (Crossref, PubMed 32010655). Gould MS, Pisani A, Gallo C, et al. 2022. Crisis text-line interventions: evaluation of texters' perceptions of effectiveness. Suicide Life Threat Behav 52(3):583–595. doi:10.1111/sltb.12873 — verified (Crossref, PubMed 35599358).

• Gould 2007 (1085 suicidal callers to eight US centres): suicidality decreased significantly during the call, hopelessness and psychological pain kept decreasing over the following weeks. Hoffberg (33 studies): "high quality evidence demonstrating crisis line effectiveness is lacking"; 80% of studies at high risk of bias; most measured only immediate outcomes. Gould 2022 (85,877 texters to Crisis Text Line): by the end of the conversation nearly 90% of suicidal texters found it helpful and nearly half felt less suicidal (some authors sit on the service's advisory boards, one is employed by it).
• Observational evaluations; systematic review. Weak to moderate for effect; consistent for immediate relief.
• Takeaway: crisis lines are the standard, reasonable signpost, and text lines suit people who cannot speak; their evidence is thin, so Sioul signposts them without overselling ("they answer day and night", not "they will save you").

1.7i Mentioning suicide does not plant it. Blades CA, Stritzke WGK, Page AC, Brown JD. 2018. The benefits and risks of asking research participants about suicide: a meta-analysis of the impact of exposure to suicide-related content. Clin Psychol Rev 64:1–12. doi:10.1016/j.cpr.2018.07.001 — verified (Crossref, PubMed 30014862). Dazzi T, Gribble R, Wessely S, Fear NT. 2014. Does asking about suicide and related behaviours induce suicidal ideation? What is the evidence? Psychol Med 44(16):3361–3363. doi:10.1017/S0033291714001299 — verified (Crossref).

• 18 studies: exposure to suicide-related questions or content went with small reductions in suicidal ideation (g = −0.13) and fewer suicidal acts (OR 0.71); no increase in risk.
• Meta-analysis of research settings (moderate); review (moderate).
• Takeaway: a quiet line naming the crisis line on the pause screen does not harm; it can stay visible for everyone, in calm wording, without asking whether the person is suicidal.

1.7j Interpersonal theory: belonging and burden. Van Orden KA, Witte TK, Cukrowicz KC, et al. 2010. The interpersonal theory of suicide. Psychol Rev 117(2):575–600. doi:10.1037/a0018697 — verified (Crossref, PubMed 20438238).

• Theory: the most dangerous suicidal desire comes from thwarted belongingness together with perceived burdensomeness (and hopelessness about both).
• Theory with partial empirical support. Moderate as theory.
• Takeaway: the message to a supporter, and the words about it on screen, must never frame reaching out as a burden ("sorry to bother you"); and a contact who does not answer must not read as rejection (5.6).

1.7k Numbers, checked at official sources on 2026-10-06.

Sources: 3114.fr ("accessible 24h/24 et 7j/7, gratuitement, en France entière"); service-public.gouv.fr, actualité A17758, 16 October 2024, updated 18 October 2024 (the twelve free emergency numbers; the 0 800 112 112 callback); info.urgence114.fr; digital-strategy.ec.europa.eu/en/policies/112; samaritans.org (contact page: "call 116 123", free from landlines and mobiles); nhs.uk, "Where to get urgent help for mental health" (111 mental health option; 999 "if someone's life is at risk"); giveusashout.org/get-help; relayuk.bt.com (emergencySMS); 988lifeline.org (call, text, chat); 988.ca (call or text, "24/7/365"); textwith911.ca; text-to-911 availability from US county pages (FCC page not reachable; not checked at the FCC). Numbers change: they must live in one data file with the date checked, be checked again before each release, and be shown for the person's country, not guessed from the language (an English speaker in France needs 3114 and 15).

1.7l What the pause must and must not do in a suicidal crisis (inference from 1.7a–k and 1.3).

• Must: show the emergency number and the crisis line of the person's country on the pause screen itself, in calm, plain words, without a question first (1.7e, 1.7i); offer the person's own safety plan, if they wrote one, in its order, one step at a time (1.7a, NICE 1.11.8); let the person reach their chosen contact in one tap, by text (1.7e, 1.1h); keep the numbers correct and dated (1.7k).
• Must not: guess risk, score it, or act on a guess (1.7g); call emergency services or anyone the person did not choose (1.7g); ask the person to promise safety (1.7d); say it is monitoring or that help is coming when it is not (3.3, 8.4); tell the person what they feel (1.2a, 4.2); show anything long to read (1.3c–d).
• Sioul is not a crisis service and must say so where the pause is set up, in one sentence, with the numbers (section 8).

2. Immediate help that has evidence¶

Summary of strength, for what an app can do in the minutes after the press:

2.1 Psychological first aid. World Health Organization, War Trauma Foundation, World Vision International. 2011. Psychological first aid: guide for field workers. WHO, Geneva. ISBN 978-92-4-154820-5 (read in full, 64 pages, from the MHPSS Hub copy of the WHO PDF). National Child Traumatic Stress Network and National Center for PTSD. 2006. Psychological First Aid: Field Operations Guide, 2nd ed. (core actions as listed at https://www.nctsn.org, read 2026-10-06). Evidence: Fox JH, Burkle FM, Bass J, et al. 2012. Disaster Med Public Health Prep 6(3):247–252. doi:10.1001/dmp.2012.39 — verified. Dieltjens T, Moonens I, Van Praet K, De Buck E, Vandekerckhove P. 2014. PLoS ONE 9(12):e114714. doi:10.1371/journal.pone.0114714 — verified. Hermosilla S, Forthal S, Sadowska K, et al. 2023. We need to build the evidence: a systematic review of psychological first aid on mental health and well-being. J Trauma Stress 36(1):5–16. doi:10.1002/jts.22888 — verified. Figueroa RA, Cortés PF, Marín H, et al. 2022. The ABCDE psychological first aid intervention decreases early PTSD symptoms but does not prevent it: results of a randomized-controlled trial. Eur J Psychotraumatol 13(1):2031829. doi:10.1080/20008198.2022.2031829 — verified. Figueroa RA, Cortés PF, Miller C, et al. 2024. Eur J Psychotraumatol 15(1):2364443. doi:10.1080/20008066.2024.2364443 — verified (all PubMed abstracts read).

• What WHO's guide says, quoted: "Don't pressure people to tell you their story"; "Don't force help on people, and don't be intrusive or pushy"; "Make it clear to people that even if they refuse help now, they can still access help in the future"; "Don't give false promises or false reassurances"; "Don't think and act as if you must solve all the person's problems for them"; "Don't take away the person's strength and sense of being able to care for themselves"; "Don't say: 'You shouldn't feel that way'"; "Give information in a way the person can understand – keep it simple"; "Allow for silence"; "Remind them that they are safe, if it is true." To calm: a calm, soft voice; for someone who "feels unreal or disconnected", place and feel the feet on the floor, tap fingers on the lap, notice non-distressing things, "focus on their breathing, and … breathe slowly". Its three actions: look, listen, link. NCTSN's eight core actions: contact and engagement (non-intrusive), safety and comfort, stabilisation if needed, information gathering on current needs, practical assistance, connection with social supports, information on coping, linkage with services.
• Evidence: Fox 2012 and Dieltjens 2014 found no controlled study and "no evidence about the effectiveness of PFA interventions"; Hermosilla 2023 found 12 studies (1437 people), one RCT, generally positive results at high risk of bias, and "an imbalance between popular support for PFA and scant evidence". Figueroa 2022 (RCT, 221 recent trauma survivors in Chilean emergency services): PFA gave more immediate relief than psychoeducation (d = 0.30) and fewer PTSD symptoms at one month (d = 0.42), but did not prevent PTSD and showed no difference at six months; Figueroa 2024 (RCT, 166, half lost to follow-up): no advantage at three months, a dose–response effect, and no more adverse effects than control.
• Consensus guidance (strong as consensus), thin trial evidence (weak to moderate), which so far shows short-term relief and no harm.
• Takeaway: PFA's prohibitions are the most directly usable text for a screen: no pressure to talk, no false reassurance, no taking over, no judging the feeling, simple information, silence allowed. "Remind them that they are safe, if it is true": software cannot know whether the person is safe, so the pause never says "you are safe"; it can say what is true about Sioul ("nothing will come in until you open it again").

2.2 Hobfoll SE, Watson P, Bell CC, et al. 2007. Five essential elements of immediate and mid-term mass trauma intervention: empirical evidence. Psychiatry 70(4):283–315. doi:10.1521/psyc.2007.70.4.283 — verified (Crossref, PubMed 18181708).

• A worldwide panel extrapolated from related research, since trials for every disaster are unlikely, and agreed on five principles: promote (1) a sense of safety, (2) calming, (3) a sense of self- and community efficacy, (4) connectedness, (5) hope. WHO's PFA guide uses the same list.
• Expert consensus built on indirect evidence. Moderate as principles; not tested as a package.
• Takeaway (inference), the five elements as checks for the pause:
  • safety: what Sioul controls is quiet and stays quiet; the emergency number is visible (1.7l);
  • calming: less input, nothing moving that the person did not choose (2.3, 4.4);
  • self-efficacy: the person started the pause and ends it; nothing is done to them that they did not set; nothing taken over that they did not hand over;
  • connectedness: the chosen contact is told, if the person set it, and can be reached in one tap (section 5);
  • hope: the screen says the pause ends when they choose, and that what waits will still be there, smaller (section 6), without promising that things will be fine.

2.3 Reducing stimulation. Novak T, Scanlan J, McCaul D, MacDonald N, Clarke T. 2012. Pilot study of a sensory room in an acute inpatient psychiatric unit. Australas Psychiatry 20(5):401–406. doi:10.1177/1039856212459585 — verified (Crossref, PubMed 23014117). Haig S, Hallett N. 2023. Use of sensory rooms in adult psychiatric inpatient settings: a systematic review and narrative synthesis. Int J Ment Health Nurs 32(1):54–75. doi:10.1111/inm.13065 — verified (Crossref, PubMed 36082841). Scanlan JN, Novak T. 2015. Sensory approaches in mental health: a scoping review. Aust Occup Ther J 62(5):277–285. doi:10.1111/1440-1630.12224 — verified (Crossref).

• Novak: using the sensory room went with significant reductions in distress, more so with a weighted blanket; no change in seclusion or aggression rates. Haig & Hallett (18 studies): no evidence on seclusion, restraint or violence, but sensory rooms "are, however, likely to support a reduction in patient distress", and patients and staff report better emotional regulation and self-management.
• Uncontrolled before–after studies and narrative review. Weak for effect; consistent with the autism (1.1e–f), ME/CFS (1.6a) and PFA (2.1: "a quiet place … minimize outside distractions") consensus.
• Takeaway: less input is the measure with the broadest agreement and no known harm. For Sioul, it means: nothing new on screen, no sound, no vibration, no badge, the window dimmed and still; on the phone, other apps' interruptions held too if the person allowed it (section 7). What the person uses to calm down (a weighted blanket, headphones, a dark room) is theirs; the pause can remind them of their own list, written beforehand, if they made one.

2.4 Grounding. WHO PFA guide (2.1); ISSTD 2011 guidelines and the Finding Solid Ground RCT (1.5b).

• Grounding means turning attention to the present through the senses: feet on the floor, hands on something, naming a few neutral things one sees or hears. It is in every first-aid and trauma-stabilisation guide, and part of programmes with trial support (1.5b), but no trial of grounding alone, as an in-the-moment technique, was found. The popular "5-4-3-2-1" counting form (five things you see, four you hear…) has no trial behind it that this search found.
• Consensus. Opinion for grounding alone.
• Takeaway: if offered, one short line, chosen beforehand by the person ("Feet on the floor. Something cold in your hand."), not a counted exercise with steps to complete, which adds a task and a way to fail.

2.5 Slow paced breathing. Fincham GW, Strauss C, Montero-Marin J, Cavanagh K. 2023. Effect of breathwork on stress and mental health: a meta-analysis of randomised-controlled trials. Sci Rep 13:432. doi:10.1038/s41598-022-27247-y — verified (Crossref, PubMed 36624160). Laborde S, Allen MS, Borges U, et al. 2022. Effects of voluntary slow breathing on heart rate and heart rate variability: a systematic review and a meta-analysis. Neurosci Biobehav Rev 138:104711. doi:10.1016/j.neubiorev.2022.104711 — verified (Crossref, PubMed 35623448). Balban MY, Neri E, Kogon MM, et al. 2023. Brief structured respiration practices enhance mood and reduce physiological arousal. Cell Rep Med 4(1):100895. doi:10.1016/j.xcrm.2022.100895 — verified (Crossref, PubMed 36630953). Lachowicz AM, Houben M, Ottaviani C, et al. 2026. Efficacy of slow-paced breathing as a just-in-time adaptive intervention for anxiety — a randomized controlled study. Appl Psychophysiol Biofeedback 51(3):579–595. doi:10.1007/s10484-025-09760-8 — verified (Crossref, PubMed 41498894; abstract read from the preprint, doi:10.31219/osf.io/5dg7z). Cautions: Schmidt 2000, Meuret 2003, Kim 2012 (1.2b); Farias M, Maraldi E, Wallenkampf KC, Lucchetti G. 2020. Adverse events in meditation practices and meditation-based therapies: a systematic review. Acta Psychiatr Scand 142(5):374–393. doi:10.1111/acps.13225 — verified. Britton WB, Lindahl JR, Cooper DJ, Canby NK, Palitsky R. 2021. Defining and measuring meditation-related adverse effects in mindfulness-based programs. Clin Psychol Sci 9(6):1185–1204. doi:10.1177/2167702621996340 — verified. Harrison OK, Köchli L, Marino S, et al. 2021. Interoception of breathing and its relationship with anxiety. Neuron 109(24):4080–4093. doi:10.1016/j.neuron.2021.09.045 — verified.

• For: Fincham (12 RCTs for stress): breathwork lowered self-reported stress (g = −0.35), anxiety (g = −0.32, 20 trials) and depressive symptoms (g = −0.40), with a warning against "a miscalibration between hype and evidence". Laborde (223 studies): slow breathing raises vagally mediated heart-rate variability during practice, right after one session and after training, "with few adverse effects expected". Balban (remote RCT, one month, 5 minutes a day): exhale-focused "cyclic sighing" improved mood and lowered breathing rate more than mindfulness meditation (one author advises a wearables company). Lachowicz (80 adults with subclinical anxiety, prompted to breathe slowly when anxiety or rumination rose): anxiety and rumination fell from just before to just after each exercise, but daily-life anxiety did not differ from control: "mixed evidence" for in-the-moment use.
• Against, or for caution: in panic disorder, breathing retraining added nothing and may hinder some (1.2b). Meditation-related adverse effects are "not uncommon" (Farias: 8.3% overall across 83 studies, anxiety the most common); in an 8-week programme, lasting bad effects in 6–14%, "associated with signs of dysregulated arousal (hyperarousal and dissociation)" (Britton). Anxiety relates to how breathing sensations are perceived and predicted at every level (Harrison), which is why attending to one's breath is not neutral for anxious people. Trauma-sensitive practice advises offering breath focus as a choice, with other anchors available (Treleaven 2018, Trauma-Sensitive Mindfulness, Norton; book, opinion, not re-read).
• Meta-analyses (moderate to strong for small benefits of regular practice); one in-the-moment RCT (weak to moderate, mixed); adverse-effect reviews (moderate).
• Takeaway: a breathing guide is a reasonable option for people who already like it, chosen in advance; it should be off by default, never framed as the cure for panic, never "deep" breaths, with a slow exhale and an adjustable pace (people differ), stoppable at once, and replaced by another anchor (sound, touch, a picture) for those for whom breath focus is a trigger. How to show it on screen: 4.4.

2.6 Distraction against attention to the feeling. Webb TL, Miles E, Sheeran P. 2012. Dealing with feeling: a meta-analysis of the effectiveness of strategies derived from the process model of emotion regulation. Psychol Bull 138(4):775–808. doi:10.1037/a0027600 — verified (Crossref, PubMed 22582737). Sheppes G, Scheibe S, Suri G, Gross JJ. 2011. Emotion-regulation choice. Psychol Sci 22(11):1391–1396. doi:10.1177/0956797611418350 — verified (Crossref, PubMed 21960251). Lieberman MD, Eisenberger NI, Crockett MJ, et al. 2007. Putting feelings into words: affect labeling disrupts amygdala activity in response to affective stimuli. Psychol Sci 18(5):421–428. doi:10.1111/j.1467-9280.2007.01916.x — verified. Torre JB, Lieberman MD. 2018. Putting feelings into words: affect labeling as implicit emotion regulation. Emotion Review 10(2):116–124. doi:10.1177/1754073917742706 — verified. Ford BQ, Lam P, John OP, Mauss IB. 2018. The psychological health benefits of accepting negative emotions and thoughts: laboratory, diary, and longitudinal evidence. J Pers Soc Psychol 115(6):1075–1092. doi:10.1037/pspp0000157 — verified (Crossref, PubMed 28703602).

• Webb (306 experimental comparisons): distraction regulated emotion (d+ = 0.27); concentrating on the emotion did not, and made it slightly worse (d+ = −0.26); suppressing the experience of an emotion did not work (d+ = −0.04). Sheppes (three experiments): at high intensity people chose distraction, which "blocks emotional processing at an early stage", and at low intensity reappraisal. Lieberman and Torre & Lieberman: naming a feeling in words dampens limbic responses in the lab, a small, implicit effect. Ford (N = 1003, 156, 222): people who habitually accept rather than judge their emotions react less to stressors and have better psychological health months later.
• Meta-analysis of lab experiments (strong for short-term effects in the lab); experiments; lab and longitudinal studies (moderate).
• Takeaway: in the acute moment, the pause should not prompt the person to look into the feeling ("notice what you feel", "what is going on for you?"); something absorbing and neutral, chosen beforehand, fits better: their music, a known film, a game, their special interest (2.7). The wording should accept the state rather than fight it ("It is all right to stop", not "let's calm down"). Naming the feeling is offered nowhere on the pause screen; the optional note afterwards may allow it (section 6).

2.7 Autism-specific: fewer demands, no questions, time. Sources in 1.1 (NAS 2020; Leicestershire Partnership NHS Trust; Zisk & Dalton 2019; Howard & Sedgewick 2021), and: Nicolaidis C, Raymaker DM, Ashkenazy E, et al. 2015. "Respect the way I need to communicate with you": healthcare experiences of adults on the autism spectrum. Autism 19(7):824–831. doi:10.1177/1362361315576221 — verified (Crossref, PubMed 25882392). Grove R, Hoekstra RA, Wierda M, Begeer S. 2018. Special interests and subjective wellbeing in autistic adults. Autism Research 11(5):766–775. doi:10.1002/aur.1931 — verified (Crossref, PubMed 29427546).

• Nicolaidis (community-based participatory research, autistic co-researchers; 39 autistic adults and 16 supporters): factors in healthcare interactions included slow processing speed, sensory sensitivities, challenges with body awareness, and verbal communication; what helped was providers' "willingness to allow written communication", "use of accessible language" and other accommodations. Grove (survey of autistic adults): special interests went with higher subjective well-being and satisfaction with leisure and social contact, though very high intensity of engagement went with lower well-being.
• Qualitative participatory study (moderate); survey (moderate); the "no questions, no instructions, give time" rule itself is consensus from autistic-informed guidance (1.1e–f).
• Takeaway: the pause asks nothing, instructs nothing, informs of nothing new; it waits. Text over speech everywhere. The person's own interests are a legitimate, evidence-consistent comfort to offer from their own list.

3. What harms¶

3.1 Forced debriefing. Rose S, Bisson J, Churchill R, Wessely S. 2002. Psychological debriefing for preventing post traumatic stress disorder (PTSD). Cochrane Database Syst Rev (2):CD000560. doi:10.1002/14651858.CD000560 — verified (Crossref, PubMed 12076399). van Emmerik AAP, Kamphuis JH, Hulsbosch AM, Emmelkamp PMG. 2002. Single session debriefing after psychological trauma: a meta-analysis. Lancet 360(9335):766–771. doi:10.1016/S0140-6736(02)09897-5 — verified. Bisson JI, Jenkins PL, Alexander J, Bannister C. 1997. Randomised controlled trial of psychological debriefing for victims of acute burn trauma. Br J Psychiatry 171:78–81. doi:10.1192/bjp.171.1.78 — verified. Mayou RA, Ehlers A, Hobbs M. 2000. Psychological debriefing for road traffic accident victims: three-year follow-up of a randomised controlled trial. Br J Psychiatry 176:589–593. doi:10.1192/bjp.176.6.589 — verified. Lilienfeld SO. 2007. Psychological treatments that cause harm. Perspect Psychol Sci 2(1):53–70. doi:10.1111/j.1745-6916.2007.00029.x — verified. NICE NG116. 2018. Post-traumatic stress disorder (read 2026-10-06). WHO 2013 guidelines on conditions specifically related to stress (2.1).

• Single-session debriefing asks the person, soon after the event, to go over it and their feelings in detail. Cochrane (11 trials): it "did not reduce psychological distress nor prevent the onset of" PTSD; one trial found a higher risk of PTSD at one year (odds ratio 2.88); "compulsory debriefing of victims of trauma should cease". van Emmerik (7 studies): no-intervention controls improved (effect 0.47) while critical incident stress debriefing did not (0.13). Bisson (133 burn victims): PTSD at 13 months in 26% of the debriefed against 9% of controls (groups not fully balanced at baseline). Mayou (road accidents, 3 years): the debriefed group did worse, and those with high initial intrusion and avoidance "remained symptomatic if they had received the intervention, but recovered if they did not". Lilienfeld lists it among potentially harmful treatments. NICE: "Do not offer psychologically-focused debriefing for the prevention or treatment of PTSD"; WHO recommends psychological first aid instead.
• Cochrane review, meta-analysis, RCTs, national guidelines. Strong.
• Takeaway: Sioul must not ask, during or right after the pause, what happened, how it felt, or what caused it; no "reflection" screen at the exit; no prompt to journal the event. An optional, later, unprompted note is a different thing (section 6).

3.2 Pressure to talk. Seery MD, Silver RC, Holman EA, Ence WA, Chu TQ. 2008. Expressing thoughts and feelings following a collective trauma: immediate responses to 9/11 predict negative outcomes in a national sample. J Consult Clin Psychol 76(4):657–667. doi:10.1037/0022-006X.76.4.657 — verified (Crossref, PubMed 18665693). WHO PFA guide (2.1).

• 2138 members of a national panel could write their reactions to 9/11 on the day and after; over two years, "participants who chose not to express any initial reaction reported better outcomes over time than did those who expressed an initial reaction", and longer responses predicted worse outcomes. PFA: "Don't pressure people to tell you their story"; "Allow for silence".
• Large longitudinal survey (moderate; choosing not to express may reflect resilience rather than cause it); consensus.
• Takeaway: not saying anything is a healthy option, not a symptom to fix. The pause has no text field, no "tell us what happened", and the exit asks nothing.

3.3 "Calm down" orders, invalidation, false reassurance, forced positivity. Wegner DM, Broome A, Blumberg SJ. 1997. Ironic effects of trying to relax under stress. Behav Res Ther 35(1):11–21. doi:10.1016/S0005-7967(96)00078-2 — verified (Crossref, PubMed 9009039). Shenk CE, Fruzzetti AE. 2011. The impact of validating and invalidating responses on emotional reactivity. J Soc Clin Psychol 30(2):163–183. doi:10.1521/jscp.2011.30.2.163 — verified (Crossref; abstract from OpenAlex). Bastian B, Kuppens P, Hornsey MJ, et al. 2012. Feeling bad about being sad: the role of social expectancies in amplifying negative mood. Emotion 12(1):69–80. doi:10.1037/a0024755 — verified (Crossref, PubMed 21787076). Wood JV, Perunovic WQE, Lee JW. 2009. Positive self-statements: power for some, peril for others. Psychol Sci 20(7):860–866. doi:10.1111/j.1467-9280.2009.02370.x — verified (Crossref, PubMed 19493324). Kirby, Day & Sagar 2019 (cited in Sioul's research.md, rule 23).

• Wegner (two experiments): people told to relax while under mental load or stress showed higher skin conductance than those not told to relax. Shenk & Fruzzetti (experiment): people given invalidating responses during a stressor had higher negative affect, heart rate and skin conductance over time than those given validating responses. Bastian (four studies, two experimental): when people think others expect them not to feel sad, they feel sad more often and more intensely, through negative self-evaluation. Wood (two experiments): repeating "I'm a lovable person" made people with low self-esteem feel worse. Warm affirmations can threaten the most self-critical (Kirby 2019, in Sioul's research.md). PFA: no "you shouldn't feel that way", no false reassurance (2.1).
• Experiments (moderate each; consistent together).
• Takeaway: no imperative to relax or calm down, no "everything will be fine", no "you've got this", no cheerful affirmations, no "don't worry". The screen states facts about what Sioul does and accepts the state as it is.

3.4 Too many choices. Chernev A, Böckenholt U, Goodman J. 2015. Choice overload: a conceptual review and meta-analysis. J Consum Psychol 25(2):333–358. doi:10.1016/j.jcps.2014.08.002 — verified (Crossref; also Sioul's research.md, rule 14). Shields et al. 2016 (1.3c); Starcke & Brand 2012 (1.3e); Leicestershire Partnership NHS Trust (1.1f: "making decisions becomes impossible").

• Choice overload appears when options are hard to compare, preferences unclear and effort to be saved; stress impairs working memory and flexibility; in shutdown, deciding can become impossible.
• Meta-analysis (strong), meta-analysis (strong), consensus.
• Takeaway: the press itself makes no choice: everything the pause does was decided beforehand. The pause screen offers at most one action besides waiting, and the exit at most one question with a default that needs no answer (6.1).

3.5 Timers and countdowns. Maule AJ, Hockey GRJ, Bdzola L. 2000. Effects of time-pressure on decision-making under uncertainty: changes in affective state and information processing strategy. Acta Psychol 104(3):283–301. doi:10.1016/S0001-6918(00)00033-0 — verified (Crossref, PubMed 10900697). Wennberg et al. 2018 and Sioul's rule 18 (research.md). NAS 2020 (1.1e).

• Participants given a deadline were "more anxious and energetic" than those with unlimited time. Sioul's own research found visual timers help time management in children with ADHD, while a ticking countdown can raise anxiety. The NAS suggests timers counting down to transitions for autistic people, because they make change predictable.
• One experiment (moderate for the direction); trial in children (moderate); consensus.
• Takeaway: no countdown on the pause screen, no "back to work in 12:00", no time elapsed shown. The pause lasts until the person ends it, or until a time they set beforehand, shown as a time of day ("until 15:30, or until you come back"), not as a ticking number. Conflict: an undo window before an alert is sent is itself a countdown (5.6); it should be short, silent, and not shown as seconds ticking away.

3.6 Gamified "streaks of calm". Silverman J, Barasch A. 2022 (issue dated 2023). On or off track: how (broken) streaks affect consumer decisions. J Consum Res 49(6):1095–1117. doi:10.1093/jcr/ucac029 — verified (Crossref). Etkin J. 2016. The hidden cost of personal quantification. J Consum Res 42(6):967–984. doi:10.1093/jcr/ucv095 — verified (Crossref). Cheng VWS, Davenport T, Johnson D, Vella K, Hickie IB. 2019. Gamification in apps and technologies for improving mental health and well-being: systematic review. JMIR Ment Health 6(6):e13717. doi:10.2196/13717 — verified (Crossref, PubMed 31244479).

• A broken streak lowers later engagement, more so when people blame themselves; measuring an enjoyable activity makes it feel like work; in mental-health apps, gamification (levels, points, rewards) is common and "not driven by health behavior change theory".
• Experiments (moderate); systematic review of designs, not outcomes (weak for effects).
• Takeaway: no count of pauses, no "calm days" streak, no badge for recovering, no history shown back to the person unless they open it themselves; a pause is never a score, good or bad.

3.7 Being watched or tracked. Dickerson & Kemeny 2004 (1.3b). Siegel R, König CJ, Lazar V. 2022. The impact of electronic monitoring on employees' job satisfaction, stress, performance, and counterproductive work behavior: a meta-analysis. Comput Hum Behav Rep 8:100227. doi:10.1016/j.chbr.2022.100227 — verified (Crossref; abstract from the preprint, doi:10.31234/osf.io/fk5ez). Huckvale K, Torous J, Larsen ME. 2019. Assessment of the data sharing and privacy practices of smartphone apps for depression and smoking cessation. JAMA Netw Open 2(4):e192542. doi:10.1001/jamanetworkopen.2019.2542 — verified (Crossref, PubMed 31002321). Barnett & Torous 2019 (1.7g). Sioul's research.md rules 11 and 19 (Rudberg Selin et al. 2026; Chen, Meng & Nie 2026).

• Social-evaluative threat, being observed and possibly judged, brings the largest and longest cortisol responses (1.3b). Electronic monitoring of employees (59 samples) went with more stress (r = .12) and less satisfaction (r = −.09). Of 36 top depression and smoking apps, 33 transmitted data to third parties, mostly Google and Facebook, often undisclosed. Adults with ADHD disliked tools that watch them; self-tracking puts interpretive and emotional demands on neurodivergent people (Sioul's research.md). The NAS asks bystanders "not to stare" (1.1e).
• Meta-analyses (strong, small effects); cross-sectional audit (moderate); qualitative (moderate).
• Takeaway: the pause keeps only a plain record the person can see and delete (6.3), sends nothing to anyone but the contact they chose, and never to the developer; it does not detect, infer or log states; the supporter is told what the person wrote, nothing more (no location unless the person chose it, 5.5); and the pause screen shows nothing about the person that someone glancing at the phone could read.

4. Talking to a person in crisis, through a screen¶

4.1 Trauma-informed frameworks. SAMHSA. 2014. SAMHSA's concept of trauma and guidance for a trauma-informed approach. HHS Publication No. (SMA) 14-4884. Read in full on 8 October 2026 from SAMHSA's official PDF as archived on 4 September 2025, https://web.archive.org/web/20250904005813id_/https://library.samhsa.gov/sites/default/files/sma14-4884.pdf : SAMHSA's own library no longer serves it (its address answers "Page Not Found" since late 2025, for no reason found). The four Rs are on its p. 9, the six principles on pp. 10–11, the "universal precautions approach" on p. 10. It is guidance for service organisations, not for software, and never speaks of disclosure; the argument against making people disclose trauma is Chen et al.'s (2022, p. 7). Chen JX, McDonald A, Zou Y, et al. 2022. Trauma-informed computing: towards safer technology experiences for all. CHI 2022. doi:10.1145/3491102.3517475 — verified (Crossref; abstract from OpenAlex). Scott CF, Marcu G, Anderson RE, Newman MW, Schoenebeck S. 2023. Trauma-informed social media: towards solutions for reducing and healing online harm. CHI 2023. doi:10.1145/3544548.3581512 — verified. Zheng W, Walquist E, Datey I, et al. 2024. "It's not what we were trying to get at, but I think maybe it should be": learning how to do trauma-informed design with a data donation platform for online dating sexual violence. CHI 2024. doi:10.1145/3613904.3642045 — verified. Kollig F, Voelker K, Ryan E, Pfafman R, Nova FF. 2026. Current landscape of mental health conversational agents from a trauma-informed care lens: scoping review. JMIR Ment Health 13:e77876. doi:10.2196/77876 — verified (Crossref; abstract from OpenAlex). Lee J-y, Koh Y. 2026. Designing a trauma-informed safety by design framework for nonconsensual intimate image removal platforms: a 2-phase sequential exploratory study. JMIR Hum Factors 13:e94438. doi:10.2196/94438 — verified (Crossref; abstract from OpenAlex).

• SAMHSA's six principles: safety; trustworthiness and transparency; peer support; collaboration and mutuality; empowerment, voice and choice; cultural, historical and gender issues. Chen et al. adapt them to computing as safety, trust, peer support, collaboration, enablement and intersectionality, with examples for UX design, security and privacy, and AI. Scott et al. apply them to social media. Zheng et al. describe a design process: adapt the principles' definitions to the domain, map design activities to them, then derive choices. Kollig et al. (38 publications, 28 mental-health chatbots): none used the trauma-informed framework explicitly; peer support and cultural, historical and gender issues were "largely absent". Lee et al. derive six interface guidelines for trauma-affected users of a removal platform (emotional scaffolding, process transparency, self-pacing, trust signals, context-aware support, grounding tools), judged plausible by eight evaluators who were not survivors: preliminary.
• Frameworks and design case studies; one scoping review. Opinion to weak as evidence of effect; strong as shared vocabulary.
• Takeaway, the principles applied to the pause (inference):
  • safety: no surprise, nothing new arrives, the way out is always visible, emergency numbers present;
  • trust and transparency: the screen says exactly what Sioul did (held mail, moved tasks, sent a message to whom, when), and nothing it did not do; Sioul is named as software ("Sioul is holding…"), never "we", which suggests people are watching or present;
  • peer support: the chosen person, not Sioul, is the support; Sioul only carries the message;
  • collaboration: every measure was set by the person beforehand, and can be changed later;
  • empowerment, voice, choice: the person starts and ends the pause; one tap undoes the alert during its short window; nothing is decided for them in the moment;
  • cultural and personal fit: texts are editable, in the person's language, with a numbers list for their country.

4.2 Wording. Sources: Shenk & Fruzzetti 2011 and the other studies in 3.3; WHO PFA guide (2.1). Dillard JP, Shen L. 2005. On the nature of reactance and its role in persuasive health communication. Commun Monogr 72(2):144–168. doi:10.1080/03637750500111815 — verified. Miller CH, Lane LT, Deatrick LM, Young AM, Potts KA. 2007. Psychological reactance and promotional health messages: the effects of controlling language, lexical concreteness, and the restoration of freedom. Hum Commun Res 33(2):219–240. doi:10.1111/j.1468-2958.2007.00297.x — verified. Williams KN, Herman R, Gajewski B, Wilson K. 2009 (online 2008). Elderspeak communication: impact on dementia care. Am J Alzheimers Dis Other Demen 24(1):11–20. doi:10.1177/1533317508318472 — verified (Crossref; abstract from OpenAlex). W3C. 2021. Making content usable for people with cognitive and learning disabilities (W3C Working Group Note), https://www.w3.org/TR/coga-usable/ (read 2026-10-06). Inclusion Europe. 2009. Information for all: European standards for making information easy to read and understand (French: L'information pour tous : règles européennes pour une information facile à lire et à comprendre), https://www.inclusion-europe.eu/easy-to-read-standards-guidelines/ (not re-read). Sutherland RJ, Isherwood T. 2016. The evidence for easy-read for people with intellectual disabilities: a systematic literature review. J Policy Pract Intellect Disabil 13(4):297–310. doi:10.1111/jppi.12201 — verified. Chinn D, Homeyard C. 2017. Easy read and accessible information for people with intellectual disabilities: is it worth it? A meta-narrative literature review. Health Expect 20(6):1189–1200. doi:10.1111/hex.12520 — verified. CDC, Crisis and Emergency Risk Communication, chapter "Psychology of a crisis" (CDC pages refused automated access; content as indexed — not checked at the source).

• Validate the act, not a guessed feeling. Validating responses lowered negative affect, heart rate and skin conductance under stress, invalidating ones raised them (Shenk & Fruzzetti). PFA acknowledges feelings the person has told the helper; software has been told nothing, and naming a feeling or a state ("you are having a panic attack", "you seem overwhelmed") is a guess that may be wrong (1.2a) and is hard to check for people with alexithymia (1.1i). What Sioul knows is that the person pressed pause: "You stopped. That is allowed." validates without diagnosing.
• No demands. Controlling language ("you must", "you should", "do this now") provokes reactance, anger and rejection of the message; concrete language gets more attention (Dillard & Shen 2005, two studies; Miller et al. 2007). Autistic-informed guidance: no instructions during a meltdown or shutdown (1.1f). Offers are phrased as what is available ("Your list of what helps is here"), never as commands ("Breathe", "Relax", "Drink water").
• No praise, no positivity, no reassurance Sioul cannot back. 3.3: "calm down" raises arousal; expected positivity amplifies sadness; affirmations backfire for low self-esteem; PFA forbids false reassurance and "you shouldn't feel that way". No "Well done for taking a break", "You've got this", "Everything will be fine", no exclamation marks, no emoji.
• No infantilising. Infantilising talk ("elderspeak") more than doubled resistance to care in videotaped nursing-home episodes (probability .55 against .26 with normal talk; 80 episodes, 20 residents); a youth in 1.1c named "you're acting like a little kid" as what made it worse; autistic adults ask to be addressed with respect and accessible language (2.7, Nicolaidis 2015). No diminutives, no "sweetie", no cartoon mascots, no "little" anything; the same adult register as the rest of Sioul.
• Plain, literal, short. COGA: use clear words, a simple tense and voice, "literal and concrete language" ("Do not use metaphors and similes unless you include an explanation", because "an autistic person may not understand jokes and similes"), "short sentences. Have only one point per sentence", paragraphs under 50 words; limit interruptions, with "an easy to use pause option"; its "cognitive stress" user story asks for "simple, consistent content" and to "know I am safe and secure". Easy Read helps by format and text difficulty, but its evidence is thin and mixed (Sutherland & Isherwood: 11 evaluations; Chinn & Homeyard: "limited evidence for the impact", individually tailored information more likely to work). Crisis communication guidance (CERC, not checked at source) says people in a crisis simplify messages, hold on to current beliefs, look for more information and opinions, and believe the first message; and people forget most of what they are told under stress (1.3d).
• Evidence: experiments (moderate); observational care study (moderate); guidelines and standards (consensus); reviews of Easy Read (weak evidence of effect).
• Takeaway: one point per sentence, five words to twelve, present tense, literal; facts about Sioul rather than claims about the person; the same words every time (predictable); editable by the person, whose own words beat any default (Chinn & Homeyard; WRAP and safety plans use the person's words, 5.3). The brief's "we are holding the fort" is an idiom and says "we": both go against Sioul's language rule (i18n.md: "no idioms") and COGA's literal-language pattern; a literal version keeps the meaning ("Sioul is keeping everything on hold").

4.3 What to show, and how much. Inference from 1.3c–d, 2.3, 3.4–3.7, 4.2 and COGA (4.2).

• On the pause screen, in this order, nothing else: (1) one word or two saying the state ("Paused."); (2) one sentence saying what Sioul is doing, true for this pause; (3) if an alert was set: who was told and when, as a fact, with a short undo while it has not left (5.6); (4) the person's own short list of what helps, folded, if they wrote one; (5) one quiet line with the crisis line and the emergency number of their country (1.7k–l); (6) one button to come back. At most about 40 words visible at once, the same layout every time.
• Not shown: counts (of mail, tasks, pauses), times running, the reason for the pause, any content from mail or tasks, any question, anything moving that the person did not start, any image of a face, any brand or decoration. Nothing that someone glancing at the phone could read about the person beyond "Paused" (3.7).
• Visual calm: Sioul's existing rules apply and suffice (muted palette with even lightness steps, no red, no pure white or black, no sounds or badges, reduced motion respected, layout never reflowing; design.md "Keyboard and senses"); the pause screen uses the darker end of the palette, large type, wide margins. Whether a dim screen helps more than a familiar one is untested; the person chooses (light or dark follows their theme).
• The "lock" locks Sioul's content, never the person: it hides mail, tasks and the plan behind the pause screen; it must not block the phone, other apps, calls or emergency dialling, and the way out is always one visible button (COGA "let users go back"; self-efficacy, 2.2). An optional second step before coming back (hold the button for a second) protects against coming back by accident; no PIN, no puzzle (shutdown, 1.1b).

4.4 Should a breathing animation be shown? Chittaro L, Sioni R. 2014. Evaluating mobile apps for breathing training: the effectiveness of visualization. Comput Hum Behav 40:56–63. doi:10.1016/j.chb.2014.07.049 — verified (Crossref; findings from the publisher's summary). Miri P, Jusuf E, Uusberg A, et al. 2020. Evaluating a personalizable, inconspicuous vibrotactile (PIV) breathing pacer for in-the-moment affect regulation. CHI 2020. doi:10.1145/3313831.3376757 — verified (Crossref; abstract from OpenAlex). W3C WCAG 2.2, Success Criterion 2.2.2 Pause, Stop, Hide (level A) and 2.3.3 Animation from Interactions (level AAA), https://www.w3.org/WAI/WCAG22/Understanding/ (read 2026-10-06). Cautions: 1.2b, 2.5.

• Chittaro & Sioni compared three app designs: a visualisation of the breathing pattern worked better than audio-only instructions, on physiology and self-report. Miri et al.: a personalised vibration pacer reduced anxiety during a cognitive stressor, with individual differences in who benefited and how engaged they were. WCAG 2.2.2: any moving content that starts automatically, lasts more than five seconds and sits beside other content needs a way to pause, stop or hide it, because "some people with cognitive disabilities and attention deficits are distracted by continuous movement"; 2.3.3: motion can cause "distraction or nausea" and vestibular reactions, so it must be possible to turn it off.
• Small experiments (weak to moderate); accessibility standard (consensus, normative for the web).
• Takeaway: not by default. When the person turned it on beforehand: it starts only when they tap it, never by itself; it shows the pattern as a slow, low-contrast shape (no text counting "in 4… hold 4…", no seconds), with the exhale longer than the inhale and the pace set in settings (about five to six breaths a minute as a start, adjustable); a tap stops it; with "reduce motion" on, it is replaced by a still picture or, on the phone, by a gentle vibration pacer, also optional; never offered by default to someone who said breath focus does not suit them (panic, trauma, asthma). Never captioned "Breathe" as an order; the button says what it shows ("Breathing guide").

4.5 Candidate texts. In the house style: short, plain, literal, no therapy talk, no exclamation marks, no praise; French with "vous", typographic apostrophes, and a narrow no-break space before ":" as in Sioul's catalogue. Each line names its sources. These are candidates to test with the people concerned, not findings.

Pause screen

Exit

Supporter message (default, editable; text, not a call): see 5.4.

Not to use, with the reason: "Calm down" / "Calmez-vous" (3.3); "Breathe" as an order (2.5, 4.2); "You are safe" (2.1: cannot know); "We are here for you" (implies people present, 4.1); "Great job taking a break!" (praise, exclamation, 3.3); "How are you feeling?" (question, alexithymia, 1.1i); "It looks like you're having a panic attack" (diagnosis, 1.2a); "Don't worry, everything will be OK" (false reassurance, 2.1); "Take all the time you need" followed by a countdown (contradiction, 3.5); "Sorry to bother you" in the supporter message (burden framing, 1.7j).

5. Alerting a trusted person¶

5.1 Social support in an acute crisis. Brewin CR, Andrews B, Valentine JD. 2000. Meta-analysis of risk factors for posttraumatic stress disorder in trauma-exposed adults. J Consult Clin Psychol 68(5):748–766. doi:10.1037/0022-006X.68.5.748 — verified (Crossref, PubMed 11068961). Ozer et al. 2003 (1.5a). Heinrichs M, Baumgartner T, Kirschbaum C, Ehlert U. 2003. Social support and oxytocin interact to suppress cortisol and subjective responses to psychosocial stress. Biol Psychiatry 54(12):1389–1398. doi:10.1016/S0006-3223(03)00465-7 — verified. Hostinar CE, Sullivan RM, Gunnar MR. 2014. Psychobiological mechanisms underlying the social buffering of the hypothalamic–pituitary–adrenocortical axis. Psychol Bull 140(1):256–282. doi:10.1037/a0032671 — verified. Seltzer LJ, Prososki AR, Ziegler TE, Pollak SD. 2012. Instant messages vs. speech: hormones and why we still need to hear each other. Evol Hum Behav 33(1):42–45. doi:10.1016/j.evolhumbehav.2011.05.004 — verified. Bolger N, Zuckerman A, Kessler RC. 2000. Invisible support and adjustment to stress. J Pers Soc Psychol 79(6):953–961. doi:10.1037/0022-3514.79.6.953 — verified. Bolger N, Amarel D. 2007. Effects of social support visibility on adjustment to stress: experimental evidence. J Pers Soc Psychol 92(3):458–475. doi:10.1037/0022-3514.92.3.458 — verified. Deisenhammer et al. 2009 (1.7e).

• After trauma, factors acting during or after the event, "such as trauma severity, lack of social support, and additional life stress", predicted PTSD more than prior factors (Brewin, 14 risk factors; Ozer, 68 studies, with social support among the significant predictors). Support from a best friend during preparation for a stress test suppressed the cortisol response (Heinrichs, RCT, 37 men); the "social buffering" of the stress axis is found across species and ages (Hostinar, review). The form matters: girls who instant-messaged their mother after a stressor did not show the hormonal relief seen with a phone call or a visit (Seltzer, small experiment); and in couples and in three experiments, support the recipient noticed as support could raise distress, while "invisible" support (practical or subtle) reduced it, because visible support can communicate "a sense of inefficacy" (Bolger 2000, 2007; N = 257 in the experiments). 76.8% of people after a suicide attempt had had some interpersonal contact during the short suicidal process (1.7e).
• Meta-analyses (strong for association); experiments (moderate).
• Takeaway: telling a trusted person is worth offering. What helps is presence and practical, low-key support that does not make the person feel incapable; the message to the supporter should ask for that kind of help. Text may soothe less than a voice for some (Seltzer, children), while for many autistic adults a call is the barrier (1.1g–h): the channel is the person's choice, set beforehand, with text as the default.

5.2 Crisis plans and advance statements: their trials. Henderson C, Flood C, Leese M, et al. 2004. Effect of joint crisis plans on use of compulsory treatment in psychiatry: single blind randomised controlled trial. BMJ 329(7458):136. doi:10.1136/bmj.38155.585046.63 — verified (Crossref, PubMed 15240438). Thornicroft G, Farrelly S, Szmukler G, et al. 2013. Clinical outcomes of joint crisis plans to reduce compulsory treatment for people with psychosis: a randomised controlled trial. Lancet 381(9878):1634–1641. doi:10.1016/S0140-6736(13)60105-1 — verified (PubMed 23537606). de Jong MH, Kamperman AM, Oorschot M, et al. 2016. Interventions to reduce compulsory psychiatric admissions: a systematic review and meta-analysis. JAMA Psychiatry 73(7):657–664. doi:10.1001/jamapsychiatry.2016.0501 — verified. Molyneaux E, Turner A, Candy B, et al. 2019. Crisis-planning interventions for people with psychotic illness or bipolar disorder: systematic review and meta-analyses. BJPsych Open 5(4):e53. doi:10.1192/bjo.2019.28 — verified. Campbell LA, Kisely SR. 2009. Advance treatment directives for people with severe mental illness. Cochrane Database Syst Rev (1):CD005963. doi:10.1002/14651858.CD005963.pub2 — verified. Swanson JW, Swartz MS, Elbogen EB, et al. 2006. Facilitated psychiatric advance directives: a randomized trial of an intervention to foster advance treatment planning among persons with severe mental illness. Am J Psychiatry 163(11):1943–1951. doi:10.1176/ajp.2006.163.11.1943 — verified (abstract not read). Tinland A, Loubière S, Mougeot F, et al. 2022. Effect of psychiatric advance directives facilitated by peer workers on compulsory admission among people with mental illness: a randomized clinical trial. JAMA Psychiatry 79(8):752–759. doi:10.1001/jamapsychiatry.2022.1627 — verified (PubMed 35662314). Cook JA, Copeland ME, Jonikas JA, et al. 2012. Results of a randomized controlled trial of mental illness self-management using Wellness Recovery Action Planning. Schizophr Bull 38(4):881–891. doi:10.1093/schbul/sbr012 — verified (PubMed 21402724).

• Henderson (RCT, 160 people with psychosis or bipolar disorder): a joint crisis plan (contacts, illnesses, treatments, relapse signs, preferences for care) halved compulsory admission or treatment over 15 months (13% against 27%, risk ratio 0.48). Thornicroft (CRIMSON, RCT, 569 people, 64 teams): no significant effect (18% against 20%), plans "not fully implemented in all study sites" and not acted on in routine reviews; only the therapeutic relationship improved. Meta-analyses: advance statements cut compulsory admissions by 23% (de Jong: 13 RCTs overall, advance statements RR 0.77, n = 1102); crisis-planning interventions by 25% (Molyneaux: 5 RCTs, RR 0.75), with no effect on voluntary admissions. Tinland (France, RCT, 394 people with schizophrenia or bipolar I disorder): advance directives written with a peer worker and shared with relatives and psychiatrists lowered compulsory admissions (27.0% against 39.9%) and symptoms. Cook (RCT, 519 adults with serious mental illness): the peer-led Wellness Recovery Action Plan (WRAP) course reduced symptoms and raised hopefulness and quality of life, at 8 weeks and 6 months.
• RCTs and meta-analyses. Moderate to strong that plans made in advance with others change what happens in a crisis; the null CRIMSON result shows a plan only works if the people around it know and use it.
• Takeaway: the pre-set alert is a very small crisis plan. What the trials share: written when well, in the person's words, shared with the people who will act on it, and agreed by them. A plan the supporter never heard of until the alert arrives is the CRIMSON failure mode; Sioul should help the person tell their supporter in advance (5.6).

5.3 What a good message to a supporter contains. Wellness Recovery Action Plan, crisis plan section: abbreviated version adapted from Copeland ME. 2002. Wellness Recovery Action Plan. Peach Press (adaptation read at https://www.getselfhelp.co.uk/docs/WRAP.pdf; WRAP itself is lived-experience-led; the RCT is in 5.2). Stanley & Brown 2012, step 4 (1.7a). Chalmers KJ, Rossetto A, Reavley NJ, et al. 2022. Redevelopment of mental health first aid guidelines for supporting someone experiencing a panic attack: a Delphi study. BMC Psychol 10:136. doi:10.1186/s40359-022-00843-3 — verified (Crossref; full text read). Chalmers KJ, Jorm AF, Kelly CM, et al. 2020. Offering mental health first aid to a person after a potentially traumatic event: a Delphi study to redevelop the 2008 guidelines. BMC Psychol 8:105. doi:10.1186/s40359-020-00473-7 — verified (PubMed 33023667). Ross AM, Kelly CM, Jorm AF. 2014. Re-development of mental health first aid guidelines for suicidal ideation and behaviour: a Delphi study. BMC Psychiatry 14:241. doi:10.1186/s12888-014-0241-8 — verified (PubMed 25213799). Marshall P, Sansom K, Jagfeld G, Jones S, Lobban F. 2023. Caring for a friend or family member who has experienced suicidal behaviour: a systematic review and qualitative synthesis. Psychol Psychother 96(2):426–447. doi:10.1111/papt.12449 — verified (PubMed 36786386). Lascelles KM, Davey Z, Jackson D, Aveyard H. 2024. Experiences and needs of adult informal carers of adults at risk of suicide: a systematic review with mixed methods analysis. J Adv Nurs 80(5):1686–1718. doi:10.1111/jan.15940 — verified (PubMed 38010822). Doherty M, Neilson S, O'Sullivan J, et al. 2022. Barriers to healthcare and self-reported adverse outcomes for autistic adults: a cross-sectional study. BMJ Open 12(2):e056904. doi:10.1136/bmjopen-2021-056904 — verified (PubMed 35193921).

• WRAP's crisis plan, written "when you are well", "in collaboration with those you are asking to support you so that they fully understand and agree to their role", asks for: signs of a crisis; the first person to contact and others; for each supporter, "details of support needed" (some for practical tasks such as children or bills, others for staying with the person); people "I do not want … involved in any way"; what helps; what to avoid; and when supporters can stop using the plan ("positive changes [that] indicate to my supporters that I am in control of things again"). Supporters sign and keep a copy.
• Mental Health First Aid Delphi guidelines (professionals and people with lived experience): for panic attacks (2022 redevelopment), the helper should be "calm, patient, respectful and reassuring, without making judgements or assumptions", ask what the person feels would help and "let the person experiencing the panic attack lead if they can"; the lived-experience group rejected telling the person they are having a panic attack and distracting them, as possibly "patronising, unnecessary or harmful" for people with experience; general encouragement of slow, even breathing was accepted but not specific breathing strategies "(e.g. counting breaths)"; blanket advice to see a doctor after an attack was rejected (the lived-experience group endorsed it at 42.1%), and only 37.8% of them endorsed a check-in later that day or the next. For trauma (2020: 103 of 183 statements endorsed), the guidelines add how to talk about what happened, short-term help and when to seek professional help; for suicidal thoughts (2014: 164 of 436 statements endorsed by both panels), both panels agreed on asking promptly and directly about suicide, an understanding, empathic and calm manner, and connecting the person with professional help, which is the supporter's role, not software's.
• Carers' side: after suicidal behaviour, carers react with "profound anxiety and intensive monitoring", struggle to know how to respond, and lack support (Marshall: 19 studies); repeated crises bring hypervigilance, burden and burnout, and the carer's hypervigilance reduces the person's autonomy (Lascelles: 31 studies). Autistic adults' barriers include phone calls (62% had difficulty booking by telephone; Doherty, 507 autistic adults).
• Lived-experience-led plan with one RCT (moderate); Delphi consensus (opinion, with lived-experience panels); qualitative and mixed-methods reviews (moderate).
• Takeaway: a good message is short, written by the person beforehand, and tells the supporter: (1) that this was set up in advance and sent by the person's own press; (2) the state, in the person's words; (3) what would help now (often: a text, no call; or come over; or nothing yet); (4) what not to do (no calls, no questions now, do not come unannounced, do not contact others); (5) how and when the person will be in touch, if they know; and nothing else: no diagnosis, no location unless chosen, no instructions to monitor. The supporter needs, once, before any alert, a short guide of their own (5.6).

5.4 Default supporter message, editable (candidates). Sources as in 5.3, 1.1h, 1.7j; house style as in 4.5. Placeholders in brackets are filled by the person when setting up the pause.

Notes: no "sorry to bother you" (1.7j); no diagnosis ("panic attack", "meltdown") unless the person writes it; inclusive French forms are a choice for the person, who edits the text in their own voice, tu or vous; the "automatic message" opening tells the supporter it was prepared in advance, which lowers alarm and avoids the reading that something new and extreme was typed in the moment (inference). Whether the default should say "overloaded" or something vaguer ("I have paused everything") is open: vaguer protects privacy, specific helps the supporter respond.

5.5 Consent and privacy. Legal detail in 8.3 and 8.5. UN Committee on the Rights of Persons with Disabilities. 2014. General comment No. 1, Article 12: Equal recognition before the law, CRPD/C/GC/1, para. 17. Gremmen I, Widdershoven G, Beekman A, Zuijderhoudt R, Sevenhuijsen S. 2008. Ulysses arrangements in psychiatry: a matter of good care? J Med Ethics 34(2):77–80. doi:10.1136/jme.2006.019240 — verified. Gergel T, Das P, Owen G, et al. 2021. Reasons for endorsing or rejecting self-binding directives in bipolar disorder: a qualitative study of survey responses from UK service users. Lancet Psychiatry 8(7):599–609. doi:10.1016/S2215-0366(21)00115-2 — verified. Saleh AZ, Appelbaum PS, Liu X, Stroup TS, Wall M. 2018. Deaths of people with mental illness during interactions with law enforcement. Int J Law Psychiatry 58:110–116. doi:10.1016/j.ijlp.2018.03.003 — verified.

• Health data: the message, and the press of a button known to mean "not coping", reveal mental state, so they are treated as special-category data in the design (8.3). Who decides: the UN committee on the disability convention calls advance planning "an important form of support" by which people state "their will and preferences which should be followed at a time when they may not be in a position to communicate their wishes", and says the moment a directive takes effect "should be decided by the person … it should not be based on an assessment that the person lacks mental capacity". Self-binding ("Ulysses") arrangements are debated: 82% of 565 UK service users with bipolar disorder endorsed them (Gergel 2021), while Gremmen et al. ask, from an ethic of care, for attention to "vulnerability, responsibility and mutuality", not only to non-interference. Escalation has risks of its own: in US data, 23% of people killed by police in 2015 showed signs of mental illness (Saleh 2018; not transferable as such to Europe).
• Authoritative interpretation (opinion level as evidence); peer-reviewed ethics and survey; observational data.
• Takeaway: the pause fits the convention's model exactly when its trigger is the person's own press, set up in advance, never a detection or someone else's judgement; it binds nothing (the person can cancel the send during the undo window, end the pause, change or remove contacts at any time). What is shared is the minimum: the message the person wrote, from their own account, to the people they chose; no location unless switched on for that contact; nothing about Sioul's content (no mail, tasks or letters). The supporter's own consent is not legally required (household exemption), but practice and the crisis-plan trials call for it (5.2, 5.6).

5.6 Risks.

• Burden on the supporter. Carers of people in suicidal crises live with fear, uncertainty and hypervigilance, and need information and support of their own (5.3, Marshall 2023, Lascelles 2024); parents of young people who self-harm described stress, guilt, sleep broken by night-time vigilance, and dread of "what I would find when I came home" (Ferrey AE, Hughes ND, Simkin S, et al. 2016. The impact of self-harm by young people on parents and families: a qualitative study. BMJ Open 6(1):e009631. doi:10.1136/bmjopen-2015-009631 — verified; 37 parents, weak to moderate); an alert left open with no follow-up is a known source of worry, which an "I am OK for now" message closes. Inference: (a) the person should be prompted, when setting the alert up, to ask the supporter first and share the message text with them (the WRAP and joint crisis plan trials rest on supporters who agreed, 5.2); (b) Sioul can give the person a short text to send to the supporter at setup: what the alert means, what it does not mean ("it is not an emergency call; Sioul does not watch me or know where I am"), what helps, what to do if they cannot answer ("nothing; I have other options"), and where to call if they fear for safety; (c) the alert never repeats on its own, and is never sent to several people at once unless the person chose that; (d) a count of alerts is not shown to the supporter.
• When nobody answers. A suicidal crisis can be short and contact with others is common (1.7e); a message left unanswered may be read as rejection by someone who already feels a burden (1.7j). Inference: the pause screen never says "no reply yet" or shows read receipts; it says what was sent, and keeps the other options visible (the person's own list, the crisis line, the emergency number). Sioul never escalates by itself to a second person or to emergency services; a second contact is only messaged if the person set two and presses again, or chose "both at once" beforehand.
• False alarms and accidental presses. Google's Emergency SOS (five presses of the power button) brought record numbers of silent emergency calls in the UK after its rollout between October 2022 and February 2023 (police statement as reported by Police Professional, 23 June 2023; 169 silent calls in 19 hours in one county, about 20 minutes each to check, as reported by TechSpot; the European Emergency Number Association reported "a surge in automatic false calls" from Android devices, sometimes "without the handset owner realizing"); Google's fix, as its help page now reads, is to "touch and hold inside the red circle for 3 seconds" or to choose an automatic start after a 5-second countdown (https://support.google.com/android/answer/9319337). Apple's crash detection placed 155 false calls in one Minnesota county in a winter (KTVU, February 2023), from phones tucked away on ski slopes. Press reports and vendor pages (opinion level as evidence, consistent with each other). When the device decides, false alarms outnumber true ones: of 49 automatic crash-detection calls to one Japanese city's fire bureau in six months, 35 were false (Hongo T, Yamamoto S, Nojima T, et al. 2023. Automatic emergency calls from smartphone/smartwatch applications in trauma. Acute Med Surg 10(1):e875. doi:10.1002/ams2.875 — verified; weak). Personal alarm pendants for older adults are "highly accepted but rarely used": people did not press even in emergencies because they were "afraid to bother", wanted to "see if it passes", or did not want to be "dragged off to hospital"; one UK register counted 40% of calls as false alarms, and some wearers feared "triggering it by accident" (Stokke R. 2016. The personal emergency response system as a technology innovation in primary health care services: an integrative review. J Med Internet Res 18(7):e187. doi:10.2196/jmir.5727 — verified; 33 studies, moderate). Takeaway: one-gesture triggers fire by accident at scale, and a button that costs someone else something goes unpressed; so pausing must never require telling anyone, and sending is a separate choice made in setup. The pause itself (holding Sioul, do-not-disturb, the calm screen) can start at once, because it is harmless and undone in one tap; the message to a person cannot be taken back, so it leaves only after a short, silent undo window shown in words ("Your message to Sam leaves in a moment. Cancel"), or after a deliberate hold if the person prefers; no countdown sound, which would defeat the purpose in sensory overload; and a one-tap "False alarm, I am fine" follow-up the person can send.
• An abusive or controlling contact. Freed D, Palmer J, Minchala D, Levy K, Ristenpart T, Dell N. 2018. "A stalker's paradise": how intimate partner abusers exploit technology. CHI 2018. doi:10.1145/3173574.3174241 — verified. Chatterjee R, Doerfler P, Orgad H, et al. 2018. The spyware used in intimate partner violence. IEEE S&P 2018:441–458. doi:10.1109/SP.2018.00061 — verified. Matthews T, O'Leary K, Turner A, et al. 2017. Stories from survivors: privacy & security practices when coping with intimate partner abuse. CHI 2017:2189–2201. doi:10.1145/3025453.3025875 — verified. Leitão R. 2019. Anticipating smart home security and privacy threats with survivors of intimate partner abuse. DIS 2019:527–539. doi:10.1145/3322276.3322366 — verified. Levy K, Schneier B. 2020. Privacy threats in intimate relationships. J Cybersecur 6(1):tyaa006. doi:10.1093/cybsec/tyaa006 — verified. Findings: abusers are often "UI-bound" adversaries with legitimate access to the victim's device or accounts; 47 accounts of tracking through location and "safety-based services"; most spyware in partner abuse is "dual-use" (child safety, anti-theft); survivors' needs differ between "physical control, escape, and life apart". Qualitative and technical studies (moderate). Takeaway: a supporter link is a dual-use feature. The supporter receives only what the person sends at a press: no status, no "pause active" indicator, no history, no read receipts, no way to trigger, query or configure anything; nobody but the person can change the setup; contacts can be removed at once with no message to them; location off by default, per contact, read once at the press, coarse unless chosen; Sioul says plainly that a sent SMS or mail stays in the phone's message store and Sent folder, where someone holding the phone can see it; one sentence at setup: "Choose someone who makes you feel safe. You can change this at any time, and they will not be told." WRAP's crisis plan has the same idea: a list of people "I do not want … involved in any way" (5.3).
• The supporter is not asked. Not a legal requirement (8.3), but a supporter surprised by a crisis message may escalate (call emergency services, come over unasked) in ways the person did not want; in France, a supporter who learns of an imminent danger may be expected to call for help (Code pénal 223-6, 8.4). Takeaway: the setup asks "Have you talked to Sam about this?", offers the editable explanatory note of the burden point above, and the message itself says what the person wants ("please text, do not call", "please do not call emergency services unless I ask" if the person writes it).

5.7 Existing apps with crisis or SOS features, and their evaluations. All DOIs below verified on Crossref on 2026-10-06; descriptions of app features from the papers, and from vendor pages where marked (opinion level).

Reviews of suicide-prevention and safety-planning apps.

• Larsen ME, Nicholas J, Christensen H. 2016. PLoS ONE 11(4):e0152285. doi:10.1371/journal.pone.0152285. Of 49 interactive apps, 27 let users reach friends or family (14 inside a safety plan), 13 gave access to crisis services, most had only one interactive component, and two listed "means of instant death". Moderate (market review). A bare "contact someone" button is the common single-component pattern; the authors favour complete safety plans.
• Martinengo L, Van Galen L, Lum E, et al. 2019. BMC Med 17:231. doi:10.1186/s12916-019-1461-z. 69 apps: 5 had all six guideline strategies; 94% emergency contacts; 38% a safety-plan template, 4 could share it with the support network; 6 (9%) gave a wrong or dead crisis number, corrected in two after the authors told the developers. Moderate.
• Parrish EM, Filip TF, Torous J, et al. 2022. Are mental health apps adequately equipped to handle users in crisis? Crisis 43(4):289–298. doi:10.1027/0227-5910/a000785. 31 popular apps: 35.5% had any crisis resource, some four taps deep; wellbeing apps (2/17) and productivity apps (0/5) least likely to have any crisis language; 81% of terms of service disclaimed liability; an Australian number shown to US users. Moderate. Sioul is closer to a productivity app; the authors recommend an international vetted directory and saying plainly what the app will and will not do.
• Bennett-Poynter L, Groves S, Kemp J, et al. 2025. BMJ Open 15(1):e087468. doi:10.1136/bmjopen-2024-087468. 52 apps in Canada and the UK: 38.5% malfunctioned during review (frozen screens, dead links and helplines), 14 more vanished from the stores; 73% also carried content for supporters; some safety plans listed methods. Moderate. Dead links persist years after Martinengo: an automated check and a visible "last checked" date.
• Gryglewicz K, Orr VL, McNeil MJ, et al. 2024. Translating suicide safety planning components into the design of mHealth app features: systematic review. JMIR Ment Health 11:e52763. doi:10.2196/52763. Seven safety-planning apps: all store trusted people and let the user contact them; 3 use GPS only to find services, none sends the user's location; emergency numbers on every page in 4; good usability but low real use; MYPLAN alone lets users pre-write messages to supporters, and users feared misunderstanding while relatives feared such messages could understate the danger. Weak for outcomes, moderate as an inventory.
• Outcome meta-analyses: Witt K, Spittal MJ, Carter G, et al. 2017 (BMC Psychiatry 17:297, doi:10.1186/s12888-017-1458-0); Torok M, Han J, Baker S, et al. 2020 (Lancet Digit Health 2(1):e25–e36, doi:10.1016/S2589-7500(19)30199-2: g = −0.18 on suicidal ideation for self-guided programmes); Melia R, Francis K, Hickey E, et al. 2020 (JMIR Mhealth Uhealth 8(1):e12516, doi:10.2196/12516: phone apps alone, no significant effect on ideation, better coping or less distress); Büscher R, Torok M, Terhorst Y, Sander L. 2020 (JAMA Netw Open 3(4):e203933, doi:10.1001/jamanetworkopen.2020.3933: web CBT courses); Burr SK, et al. 2025 (Crisis 46(3):176–186, doi:10.1027/0227-5910/a000996: g = 0.11) and Sim K, Bae SM. 2025 (Behav Sci 15(10):1345, doi:10.3390/bs15101345: g = 0.39 with non-randomised studies; the authors say people at imminent risk need referral). Moderate to strong, small effects, for programmes used over weeks. No study tested an in-the-moment pause or alert button on its own.

Apps one by one.

• MYPLAN (Denmark): Buus N, Juel A, Haskelberg H, et al. 2019. JMIR Ment Health 6(4):e11965. doi:10.2196/11965; Buus N, et al. 2020. Arch Suicide Res 24(1):48–63. doi:10.1080/13811118.2018.1489319. Designed with users: a yellow "Help" button (up to 10 people, "I need to talk") and a red "Alarm" button (one person or the emergency number), each at most two taps from the home screen; "quick messages" written by users in advance; location sent only by the user's own act. Relatives, shown a vignette where a phone alerts parents and shares location, rejected surveillance ("It has to be optional"; the person "would have to control it herself") and doubted whether they could cope with receiving alerts. The RCT (NCT02877316) has no published results. Weak. The closest existing design to the pause.
• BeyondNow (Australia): Melvin GA, Gresham D, Beaton S, et al. 2019. Suicide Life Threat Behav 49(3):846–858. doi:10.1111/sltb.12490. Uncontrolled pilot, 36 patients: 30.6% pressed the emergency button (whether a call followed was unknown); 5.6% used sharing; people listed few contacts (1.75 friends or family on average) against six warning signs and six coping strategies. Weak. The person-to-person part is the hardest to fill when unwell: set up on a good day, with one or two people.
• Virtual Hope Box (US): Bush NE, Smolenski DJ, Denneson LM, et al. 2017. Psychiatr Serv 68(4):330–336. doi:10.1176/appi.ps.201600283. RCT, 118 veterans with recent suicidal ideation: better coping with unpleasant emotions at 3 and 12 weeks than printed materials; no effect on ideation. Content is the person's own photos, music, recorded messages from loved ones, distraction games, relaxation. Moderate. The best trial evidence in this field is for a calm toolbox the person fills in advance, not for alerts.
• BackUp (Belgium): Pauwels K, Aerts S, Muijzers E, et al. 2017. PLoS ONE 12(6):e0178144. doi:10.1371/journal.pone.0178144. No account, no server, no GPS, no score or anything one can "lose", red avoided as anxiety-provoking, crisis line on every screen, a linked supporter who gets guidance ("Being BackUp"); users found it reassuring to have even when unused. Weak. Its choices match Sioul's.
• Others: MY3 (US; one-tap calls to three chosen people; no texting; in one small registry-only trial, the bundle including it did no better than usual care; its staff voicemail told callers what to do in an acute crisis when nobody answered); ASAP/BRITE (Kennard BD, Goldstein T, Foxwell AA, et al. 2018. Am J Psychiatry 175(9):864–872. doi:10.1176/appi.ajp.2018.17101151: graded response, skills at low distress, people at high distress; pilot RCT, 66 adolescents, attempts 16% against 31%, not significant); Jaspr (Dimeff LA, et al. 2021, JMIR Ment Health 8(3):e23022, doi:10.2196/23022: emergency-department tablet; staff mostly chose "comfort and skills" content; developers co-own the company); Stay Alive (commissioned evaluation, grey literature: "Find Help Now" was the most viewed page); SafePlan (no automatic sharing, by clinical decision); Be Safe (prepared "scripts" of what to say when reaching out; vendor, no evaluation); Calm Harm (calming tools open without the passcode that protects the private journal; vendor). Weak to opinion.

Autism support tools.

• Brain in Hand (UK, commercial): Tromans S, Henley W, Summers I, et al. 2023. BJPsych Open 9(3):e96. doi:10.1192/bjo.2023.57. Uncontrolled cohort, 99 autistic adults (people at any suicide risk excluded), partly company-funded with an autistic user panel: anxiety fell (HADS −2.23), functioning improved, a third dropped out. Pressing the "red" button helped by itself as a self-signal ("just the act of doing that kind of pulled me out of that cycle"); but "many participants reported anxiety when receiving a call from the response service", and asked for more choices. Vendor help pages: the default response is a phone call from a company responder, switchable to text or email; "On-demand support can take up to 30 minutes to respond"; for immediate help it points to Samaritans; a "personal responder" the person knows can receive notifications only, or also see the person's plans (opinion level). Weak. Lessons: the press itself can help; a call in response can harm; channel and level of access are the person's to set, with agreement.
• Molehill Mountain: Oakley B, et al. 2026. Autism Res 19(1):e70153. doi:10.1002/aur.70153. Feasibility study, 100 autistic people: anxiety fell (d = 0.45, no control); valued support "without the requirement for social contact"; wanted reminders only with "greater control over timing and delivery"; one withdrew because the app raised anxiety. Weak.
• Autism-adapted safety plans: Rodgers J, Cassidy S, Pelton M, et al. 2024. eClinicalMedicine 73:102662. doi:10.1016/j.eclinm.2024.102662. Pilot RCT, 49 autistic adults: plans made with a trained helper, by phone or video, over several sittings, in adapted wording ("what is important to me"); 68% satisfied, 41% rated usable on a scale that rewards quick solo completion; fewer serious adverse events with the plan (3) than in usual care (9), none related to the study. Weak. Setup must be unhurried, resumable, and allow help from a trusted person.
• Stress Autism Mate: Spaargaren KL, Roke Y, Begeer SM, et al. 2025. Autism 29(10):2588–2603. doi:10.1177/13623613251346885. RCT, 214 autistic adults: lower stress (d = 0.27), better well-being (d = 0.35), but 42.9% said "I got stressed from using the app" (prompts, response deadlines, PINs and weekly logins). Moderate. A co-designed tool can still stress half its users through prompts and deadlines: the pause adds none, and needs no unlock at the moment of use.
• 988 and autistic youths: Rothman EF, et al. 2026. Autism in Adulthood (online). doi:10.1177/25739581251374833 (read through a summary). Some feared emergency dispatch; users reported long waits and scripted counsellors. Weak. Another reason Sioul never contacts emergency services itself.

Panic, anxiety and trauma apps.

• PTSD Coach: Kuhn E, Kanuri N, Hoffman JE, et al. 2017. J Consult Clin Psychol 85(3):267–273. doi:10.1037/ccp0000163 (RCT, 120 adults: more improvement than waitlist); Owen JE, Jaworski BK, Kuhn E, et al. 2015. JMIR Ment Health 2(1):e7. doi:10.2196/mental.3935 (153,834 downloads: median session about 47 seconds; distress fell 1.6–2.0 points on 0–10 among the 21% who rated, used a tool and rated again; when distress stays high the app suggests a support person or a crisis line). Moderate. In-the-moment use lasts under a minute; people come second, after a tool, not first.
• Market reviews of panic and anxiety apps: Van Singer M, Chatton A, Khazaal Y. 2015. Front Psychiatry 6:96. doi:10.3389/fpsyt.2015.00096; Sucala M, Cuijpers P, Muench F, et al. 2017. Depress Anxiety 34(6):518–525. doi:10.1002/da.22654; Alyami M, Giri B, Alyami H, Sundram F. 2017. Evid Based Ment Health 20(3):65–70. doi:10.1136/eb-2017-102664: poor quality, rarely tested (3.8% rigorously tested in Sucala), often with advertising. Commercial panic apps (Rootd, Dare) send unsolicited motivational notifications and use guilt wording when the user leaves (walkthrough study: Nalepa A, Kopecka-Piech K. 2024, issue 2025. Annales UMCS sectio J 37(4):123–145. doi:10.17951/j.2024.37.4.123-145). Wysa (vendor data): after pressing SOS, 49.2% chose the safety plan and 46.6% grounding, 2.4% "chose to call helplines when repeatedly encouraged". Opinion to moderate. People in crisis mostly choose self-help over calls; repeated encouragement is pressure.

Personal-safety SOS apps (vendor documentation, opinion level). Noonlight, Life360 and bSafe (US and Norway) share a pattern: two levels, a "heads up" (hold, for feeling unsafe) and "help now" (tap or release); a PIN or a call to confirm or cancel; and, in paid plans, dispatch to emergency services when the user does not answer ("even if you can't answer", Life360). notOK (US, 2018) sends a fixed text with a GPS link to up to five contacts ("Hey I'm notOK. Please call me, text me, or come check up on me.", wording from its MIT Solve page; no evaluation, no consent step for contacts documented). Eisenhut K, Sauerborn E, García-Moreno C, Wild V. 2020. Mobile applications addressing violence against women: a systematic review. BMJ Glob Health 5(4):e001954. doi:10.1136/bmjgh-2019-001954 — verified: of 171 apps, 46.8% were emergency apps alerting chosen contacts or police, the most downloaded kind, almost none evaluated; the authors warn of privacy breaches and of "loss of personal contact with supporters". Takeaway: the two-level pattern is worth considering (a pause that tells no one, and a pause that also tells the chosen person); dispatch when the person does not answer is not, for Sioul's users; a fixed message written by the app is the opposite of the person's own words; an alert should support a relationship agreed in advance, not replace it.

What users disliked. Kang RM, Reynolds TL. 2024. "This app said I had severe depression, and now I don't know what to do": the unintentional harms of mental health applications. CHI 2024. doi:10.1145/3613904.3642178. Bowie-DaBreo D, Sas C, Iles-Smith H, Sünram-Lea S. 2022. CHI 2022. doi:10.1145/3491102.3517498. Store-review analyses: automated judgements with no next step, pressure (prompts, deadlines, streaks), large menus at a bad moment, technical failures, commercial models. No peer-reviewed study naming "infantilising" tone in crisis features was found; the closest evidence is the preference for neutral, plain wording (Stress Autism Mate), users rewriting clinical texts (MYPLAN), and "A good thing to say would be nothing" (an autistic youth in Phung 2021).

Conflicts. MYPLAN users asked for a red "Alarm" button to signal urgency; BackUp avoided red as anxiety-provoking (neither tested). Calm Harm closes with "Did it help?"; the debriefing and expression evidence (3.1–3.2) argues for no question at the exit; a one-tap, optional rating of a tool is a different act from asking about the event.

Takeaway. The pause is closest to MYPLAN's "Help" button and Brain in Hand's personal responder, with BackUp's privacy choices and Virtual Hope Box's own-content toolbox. What none of them has, and Sioul would add: holding the person's admin (mail, tasks, plan) and protecting the rest of the day. The evaluated lessons: messages written by the person; supporters agree beforehand; text, not a call, unless chosen; no surveillance or automatic location; the help path ("Find help now") visible and correct; nothing to unlock, rate or answer at the moment of use.

6. After the crisis¶

6.1 Gentle re-entry. Sources: capacity report criterion 36 ("when the person says today is bad, re-plan today to a lighter level they choose (default ~60% of budget, or rest), move the non-urgent rest", built as the day's weather); NICE NG206 (1.6a: do not return to usual activity until a flare has resolved; post-exertional malaise 12–48 hours later); Raymaker 2020 (1.1j: time off, reduced expectations); Dickerson & Kemeny 2004 (1.3b: uncontrollable and judged stress takes longest to recover from); Shields 2016 (1.3c); Sioul's health.md and areas.md (doses come through sleep unless the person chose otherwise).

• No study tested how software should bring someone back after a crisis. What the sources agree on: recovery takes longer than the trigger (hours for overload and acute stress, days for a crash or burnout); demands should come back reduced and one at a time; the person, not the planner, judges when they are ready; and an emotional crisis is itself an exertion that can bring a delayed crash.
• Takeaway (inference): (a) ending the pause opens a calm screen, not the Porch: what was held stays held until the next admin window or until the person opens it, except what Sioul already lets through in quiet time (codes the person just asked for); (b) the rest of the day is re-planned at the bad-day level (criterion 36: about 60 %, or rest), without asking, if the person allowed it in setup, and the paused hours are not pushed into the evening (the brief: the pause does not extend the working day); (c) one optional offer, default "no": lighten tomorrow too (for burnout and post-exertional malaise, 1.6); (d) the pause and the day after never count as a bad outcome for learning the long-term budget unless the person says so in the evening review (capacity criterion 36 records "today is bad" as an outcome without lowering the long-term budget unless the previous days were heavy); (e) dose reminders and the alarms the person set keep coming during the pause, as they do during sleep, unless the person chose otherwise for the pause in setup: a missed dose can matter more than an interruption (Sioul's own rule in health.md, "Sioul never silences a dose without that choice").

6.2 No forced reflection; an optional note later. Sources: 3.1 (debriefing), 3.2 (Seery 2008). Frattaroli J. 2006. Experimental disclosure and its moderators: a meta-analysis. Psychol Bull 132(6):823–865. doi:10.1037/0033-2909.132.6.823 — verified (Crossref, PubMed 17073523).

• Writing about upsetting experiences has a small positive average effect (146 randomised studies, r = .075), with moderators; the studies are of chosen, scheduled writing sessions, not prompts right after an event. Early, structured re-telling (debriefing) did not help and harmed some (3.1); choosing not to express a reaction went with better outcomes (3.2).
• Meta-analyses and trials. Moderate to strong for "no early prompted re-telling"; weak for any benefit of a note right after.
• Takeaway: the exit asks no question about what happened. The person's ordinary notes stay available, and an optional line ("A note about today, if you want one, is in Notes.") can appear later in the day, once, without a prompt to fill it; nothing about the pause appears in the evening review unless the person writes it there. Any note stays private, in their own files, like the rest of Sioul's notes.

6.3 What Sioul keeps. Sources: 3.6–3.7.

• Takeaway: a pause leaves a plain local record the person can see if they look (date, time, what was held, whether a message was sent, to whom), and can delete; no counts, no charts, no history shown unprompted, nothing synced to anyone but the person's own devices. Whether to keep even that is a setting, defaulting to keep the record only for undo and for the person's own reference (open question 6).

6.4 When and how to suggest professional help, without pushing. Chalmers et al. 2022 (5.3); NICE NG116 (3.1: "consider active monitoring" for subthreshold PTSD symptoms within a month, with follow-up within a month); WHO 2013 (2.1); ICD-11 QE84 (1.3a); Doherty et al. 2022 (5.3); Camm-Crosbie et al. 2019 (1.7f); Dillard & Shen 2005 and Miller et al. 2007 (4.2); Ng JYY, Ntoumanis N, Thøgersen-Ntoumani C, et al. 2012. Self-determination theory applied to health contexts: a meta-analysis. Perspect Psychol Sci 7(4):325–340. doi:10.1177/1745691612447309 — verified (PubMed 26168470). Motto JA, Bostrom AG. 2001. A randomized controlled trial of postcrisis suicide prevention. Psychiatr Serv 52(6):828–833. doi:10.1176/appi.ps.52.6.828 — verified (PubMed 11376235). Milner AJ, Carter G, Pirkis J, Robinson J, Spittal MJ. 2015. Letters, green cards, telephone calls and postcards: systematic and meta-analytic review of brief contact interventions for reducing self-harm, suicide attempts and suicide. Br J Psychiatry 206(3):184–190. doi:10.1192/bjp.bp.114.147819 — verified (PubMed 25733570).

• After one panic attack, a blanket suggestion to see a doctor was rejected by people with lived experience, who warned against assuming a disorder; the guideline keeps it for recurring attacks or when the person has changed their life to avoid them (Chalmers 2022). Acute stress reactions usually subside within days (ICD-11); NICE advises watchful follow-up within a month for symptoms after trauma, and treatment when PTSD symptoms persist. Autistic adults' main barriers to care are deciding whether symptoms warrant a visit (72%) and booking by phone (62%); they prefer online or text booking and sending the reason in advance (Doherty). Support for autonomy goes with autonomous motivation and better health outcomes across 184 data sets (Ng); controlling language provokes rejection, and a sentence restoring the reader's freedom softens it (Miller). After a suicidal crisis, regular caring letters from professionals lowered suicide for two years in one large trial (Motto & Bostrom: 843 people who had refused care), but brief contact interventions overall did not significantly reduce repeat self-harm or suicide (Milner: 12 trials in meta-analysis).
• Delphi consensus; guidelines; survey (moderate); meta-analysis (strong for autonomy support in general); one large RCT and a meta-analysis with mixed results for follow-up contact.
• Takeaway: (a) never on the exit screen, never after a single pause; (b) offered once, in plain words, when the person used the pause several times for the same kind of reason in a few weeks (threshold to be chosen with users; open question 8), or when they mark "this keeps happening"; (c) the offer is practical and removes the barrier: "Some people talk to a doctor about this. Sioul can draft a message or an appointment request for you to check." with "Not now" and "Do not suggest this again" (a postscript that restores freedom, Miller 2007); (d) for suicidal crises the crisis line stays visible every time (1.7l) — that is signposting, not a suggestion to be rationed; (e) Sioul sends no "checking on you" messages of its own: software imitating caring contacts would be a false presence (4.1); a person who wants follow-up can ask their supporter for it, or schedule a note to themselves.

7. Platform facts¶

From official documentation and the Android source (AOSP), read on 2026-10-06; the Android 15 do-not-disturb change and the SMS permission's place on Android 15's restricted list were re-checked at the source. "Not checked" marks what could not be confirmed at an official page.

7.1 Do-not-disturb on Android, with exceptions.

• Access: ACCESS_NOTIFICATION_POLICY is only a "marker permission"; the real gate is "Do Not Disturb access", a switch the person turns on in system settings, opened with Settings.ACTION_NOTIFICATION_POLICY_ACCESS_SETTINGS ("In some cases, a matching Activity may not exist, so ensure you safeguard against this"); isNotificationPolicyAccessGranted() tells whether it is on. https://developer.android.com/reference/android/app/NotificationManager
• Android 15 changed the model: "Apps that target Android 15 (API level 35) and higher can no longer change the global state or policy of Do Not Disturb (DND) on a device … Instead, apps must contribute an AutomaticZenRule, which the system combines into a global policy with the existing most-restrictive-policy-wins scheme." The old calls (setInterruptionFilter, setNotificationPolicy) now create an implicit rule, shown as "Do Not Disturb (Sioul)". https://developer.android.com/about/versions/15/behavior-changes-15
• The cleaner way: an app-owned AutomaticZenRule created at setup (type TYPE_OTHER, name "Pause", a trigger description such as "When you press Pause in Sioul"), switched with setAutomaticZenRuleState(id, Condition(…, STATE_TRUE / STATE_FALSE, SOURCE_USER_ACTION)); on Android 15 and later the person sees it in Settings ▸ Modes and can edit it, and Sioul should respect their edits (areAutomaticZenRulesUserManaged()). ZenPolicy.Builder: allowCalls and allowMessages with PEOPLE_TYPE_STARRED (or contacts, anyone, none), allowRepeatCallers(true), allowAlarms(true), allowPriorityChannels(true) (API 35). Turning the pause off turns off only Sioul's rule, never a mode the person started. https://developer.android.com/reference/android/app/AutomaticZenRule ; https://developer.android.com/reference/android/service/notification/ZenPolicy.Builder
• Who gets through: Android has no per-person allow list. The options are starred contacts, all contacts, anyone or no one (calls and messages separately), "priority" conversations (marked by the person in the messaging app; an app cannot set them), and repeat callers ("if the same person calls twice in 15 minutes", Google help https://support.google.com/android/answer/9069335). Starring is the contacts provider's STARRED column, writable only with WRITE_CONTACTS, and it changes the person's favourites everywhere: Sioul should not write stars, but open the Contacts app's starred list for the person to edit.
• Doses and alarms during the pause: a notification channel created with setBypassDnd(true) (allowed for apps with do-not-disturb access, "only if the channel hasn't been updated by the user since its creation") plus allowPriorityChannels and allowAlarms in the rule. https://developer.android.com/reference/android/app/NotificationChannel#setBypassDnd(boolean)
• /e/OS is built on LineageOS and AOSP; these are framework APIs that need no Google services (no /e/OS-specific change found; not checked at an /e/OS page). Test on the build in use.
• Takeaway: do-not-disturb with exceptions is feasible on Android without any restricted permission, as a "Pause" mode the person can see and adjust; "approved people" means starred contacts (plus repeat callers), which the person curates in their Contacts app.

7.2 Sending the message from the phone.

• Direct SMS: SmsManager.sendTextMessage needs SEND_SMS ("Protection level: dangerous. This is a hard restricted permission", allowlisted by default for any installer except where an app store narrows it); its sentIntent reports success or the failure reason (no service, radio off…), and deliveryIntent a delivery report only if the network returns one; on dual-SIM phones use createForSubscriptionId (no SIM dialog). For an app that is not the default SMS app, "the system automatically writes messages sent using this method to the SMS Provider": the message appears in the person's Messages app. https://developer.android.com/reference/android/telephony/SmsManager
• Android 15's "restricted settings": for apps installed from a local or downloaded APK file, Android 15 puts SEND_SMS (and the other SMS permissions, accessibility, notification listener, display over other apps) behind Enhanced Confirmation Mode: asking for it shows a "Restricted setting" dialog with no "allow" button; the person must find Settings ▸ Apps ▸ Sioul ▸ ⋮ ▸ "Allow restricted settings", then grant it (AOSP EnhancedConfirmationService, android15-release, PROTECTED_SETTINGS.add(Manifest.permission.SEND_SMS), and PACKAGE_SOURCE_LOCAL_FILE / PACKAGE_SOURCE_DOWNLOADED_FILE treated as dangerous sources; Google help https://support.google.com/android/answer/12623953). Do-not-disturb access and full-screen intents are not on that list. Apps installed through a store client may avoid the wall depending on the device's trusted-installer list. Vendor builds (Samsung, /e/OS): not checked.
• Google Play's SMS policy restricts SEND_SMS to default SMS apps or declared exceptions, one being "Physical safety/emergency alerts" (https://support.google.com/googleplay/android-developer/answer/10208820); it concerns Play distribution only, and declaring an "emergency alert" use would contradict "not an emergency service" (8.1).
• Google's developer verification for apps installed outside Play: from 30 September 2026 for installs from participating stores in Brazil, Indonesia, Singapore and Thailand on certified devices, globally in 2027; direct sideloading is not yet affected; an "advanced flow" (developer mode, a coaching check, a restart, a one-day wait) or ADB will allow unregistered apps; a free limited account covers 20 devices. https://developer.android.com/developer-verification . /e/OS ships microG and is not a certified device; an /e/ community moderator wrote that it "does not impact /e/OS" (forum post, February 2026; not an official /e/ page).
• Without any permission: ACTION_SENDTO with an smsto:<number> URI and an sms_body extra opens the person's SMS app with the text written; the person presses send there (one tap). https://developer.android.com/guide/components/intents-common#SendMessage . Third-party apps cannot send RCS (press reports; not checked at an official source). WhatsApp and Signal links only pre-fill (not checked at their official pages); Telegram bots cannot start conversations (https://core.telegram.org/bots); Matrix's client-server API lets Sioul send from the person's own account without a tap (PUT /_matrix/client/v3/rooms/{roomId}/send/{eventType}/{txnId}, https://spec.matrix.org/latest/client-server-api/), with end-to-end encryption needing a client library.
• Takeaway: three modes, chosen at setup: (a) direct SMS, after granting SEND_SMS on a calm day (and, on Android 15 with a sideloaded build, crossing the restricted-settings screen, which Sioul should explain without urging); (b) a pre-filled SMS the person sends with one tap; (c) mail or Matrix from the person's own account. Each reports honestly what happened (P19).

7.3 The calm screen on Android. Activity.setShowWhenLocked(true) keeps Sioul's screen visible above the lock screen if the phone locks during the pause (it must then show nothing private). Screen pinning (startLockTask() without a device owner) shows a system dialog and blocks reaching the dialler while pinned: too strong. Kiosk lock-task needs a device-owner app: out of scope. Full-screen intents are not needed (the person is already in Sioul), and since Android 14 Play limits them to calling and alarm apps. https://developer.android.com/reference/android/app/Activity#setShowWhenLocked(boolean) ; https://developer.android.com/reference/android/app/Activity#startLockTask() . Takeaway: "lock" means Sioul's own windows show the pause screen until the person ends it; the rest of the phone stays free (P14).

7.4 The phones' own emergency features. Android Emergency SOS (Android 12 and later; power button pressed five times; "touch and hold inside the red circle for 3 seconds" or an automatic 5-second countdown; calls the emergency number, can share location with emergency contacts), Google's Safety Check (a timer; "If you don't choose one of the options in 60 seconds, emergency sharing will begin"), Samsung's Emergency SOS (https://www.samsung.com/ae/support/mobile-devices/setting-up-safety-and-emergency-on-your-galaxy-device/) and Apple's Emergency SOS (countdown, then a text with location to emergency contacts after the call). https://support.google.com/android/answer/9319337 ; https://support.apple.com/guide/personal-safety/emergency-call-text-iphone-apple-watch-ips4f0cd709b/web . No public API lets an app read the system's emergency contacts (AOSP keeps them in the Emergency Info app's private preferences). Accidental calls after these features spread: 5.6. Takeaway: life-threatening situations are what the phone's own SOS is built for, with location and the operators' infrastructure; Sioul's setup can point to it in one line, and never duplicates it or calls emergency services itself.

7.5 Do-not-disturb on desktops. Re-checked at the sources (GNOME and KDE source code, Microsoft's API page).

• Freedesktop: the Desktop Notifications Specification has no do-not-disturb or inhibition method, and the XDG desktop portals have none either (the Inhibit portal covers logout, user switch, suspend and idle only). https://specifications.freedesktop.org/notification-spec/latest/ ; https://flatpak.github.io/xdg-desktop-portal/docs/api-reference.html . Notifications sent with Critical urgency pass GNOME's and Plasma's do-not-disturb by default.
• GNOME: the "Do Not Disturb" switch is the setting org.gnome.desktop.notifications show-banners, inverted (gnome-shell js/ui/status/doNotDisturb.js); with it on, banners are hidden except critical ones, notifications still go to the list; no allow-list of people. Any unsandboxed process of the session can write it (inferred from how dconf works; not stated on an official page); a Flatpak app needs extra dconf permissions (Flatpak documentation). It is the person's global switch: Sioul must remember its value before the pause, restore it after, never turn off a do-not-disturb the person had set, and after a crash offer to restore it.
• KDE Plasma: Plasma's notification server adds Inhibit(desktop_entry, reason, hints) → cookie, UnInhibit(cookie) and an Inhibited property to org.freedesktop.Notifications (plasma-workspace org.freedesktop.Notifications.xml); the inhibition ends when Sioul calls UnInhibit or when its D-Bus connection closes (a crash cannot leave the desktop silenced), and the applet shows it as "While Sioul is active (Pause)". Plasma also goes quiet by itself when a full-screen window has focus (default on).
• Windows 11: no supported way for an ordinary app to turn do-not-disturb on: Windows.UI.Shell.FocusSessionManager.TryStartFocusSession is "part of a Limited Access Feature" needing an unlock token from Microsoft, on a page marked prerelease; ToastNotificationManagerForUser.NotificationMode reads the current mode (PriorityOnly, AlarmsOnly). https://learn.microsoft.com/en-us/uwp/api/windows.ui.shell.focussessionmanager . Windows' own settings can turn do-not-disturb on "when using an app in full-screen mode" (Microsoft Support). Undocumented WNF toggles exist in community tools: unsupported.
• macOS: no public API turns a Focus on. INFocusStatusCenter only reads the status, with the person's permission; Focus filters (SetFocusFilterIntent) let an app adapt when a Focus changes. An app can run a shortcut the person made (shortcuts run "Sioul pause"), which can switch a Focus (the "Set Focus" action was not confirmed at an official Apple page). https://developer.apple.com/documentation/intents/infocusstatuscenter ; https://support.apple.com/guide/shortcuts-mac/run-shortcuts-from-the-command-line-apd455c82f02/mac
• Takeaway: on every desktop Sioul can quiet itself; desktop-wide do-not-disturb is possible on Plasma (cleanly), on GNOME (by changing the person's own switch, restored afterwards), on macOS through a shortcut the person makes at setup, and not on Windows without Microsoft's approval, where Sioul's full-screen pause window and the person's own setting are the fallback. The pause screen says truthfully what was silenced (P10).

7.6 Mail from the person's own account. Sioul already sends mail through the person's SMTP account, normally from the Outbox with a deliberate key (design.md). For the pause, the setup is the review: the message was written, addressed and confirmed in advance (P4), so the press sends it at once (after the undo window), and the status line says whether the server accepted it. Takeaway: no new infrastructure; one exception to the Outbox rule, explicit in setup.

8. Legal and ethical¶

Not legal advice. Primary texts were read at EUR-Lex copies on the Publications Office, the European Commission, gov.uk, fda.gov, ansm.sante.fr, has-sante.fr, cnil.fr and edpb.europa.eu on 2026-10-06; the MDCG 2019-11 rev.1 and FDA 2022 quotes below were re-checked against the PDFs.

8.1 Not a medical device: intended purpose. Regulation (EU) 2017/745 (MDR), https://eur-lex.europa.eu/eli/reg/2017/745/oj. MDCG 2019-11 rev.1 (June 2025), Guidance on qualification and classification of software, https://health.ec.europa.eu (PDF mdcg_2019_11_en.pdf). Manual on borderline and classification, version 5 (April 2026). CJEU C-219/11 Brain Products (2012) and C-329/16 Snitem and Philips France (2017).

• A medical device is software "intended by the manufacturer" for a medical purpose (Art. 2(1)), and the intended purpose is what the maker says "on the label, in the instructions for use or in promotional or sales materials or statements" (Art. 2(12)); software "intended for life-style and well-being purposes is not a medical device" (Recital 19); claims must not ascribe functions the device does not have (Art. 7). MDCG 2019-11: software whose action is limited to "storage, archival, communication, simple search" is not medical device software; "e-mailing, web or voice messaging" are named as non-medical; "Communication systems are normally based on software for general purposes", but "a software module generating alarms based on the monitoring and analysis of patient specific physiological parameters" is a device. Device examples close to the pause: a smartwatch app that alerts when it recognises an irregular heartbeat; apps to "assess, monitor, and manage depression". The borderline manual judged an app for preventing sexually transmitted infections, which, "where the functionality has previously been set up by the user", sends automatic notifications to chosen partners, to be communication, not a medical device. The CJEU: use in a medical context is not enough, the maker's purpose must be "specifically medical" (Snitem, para. 24).
• Binding law and official (non-binding) guidance.
• Takeaway: the pause as described (hold the app, send a message the person wrote to a person they chose, set do-not-disturb, show fixed text) is communication and settings, not a medical device, as long as Sioul's own words give it no medical purpose and it does not detect anything. Words to avoid everywhere Sioul speaks (app, README, website, store listing, talks): "detects a crisis / panic attack / meltdown", "monitors your mental state", "prevents suicide", "keeps you safe", "treats / relieves / calms anxiety or panic", "manages autism / ADHD", "therapeutic", "clinically proven", "alerts your care team", "notifies emergency services", "SOS". Wording that describes what the software does: "Pause puts Sioul on hold"; "If you set it up, Pause sends a message you wrote in advance to a person you chose. It sends nothing else."; "Sioul does not watch you and cannot tell how you are. It acts only when you press the button."; "Sioul is not a medical tool and not an emergency service." Calling the feature "Pause" rather than "Emergency" avoids suggesting an emergency service. Automatic crisis detection from any signal (typing, heart rate, sleep, missed doses, words in mail) would likely make it a device, possibly in a high class under Rule 11 (Annex VIII).

8.2 Elsewhere, for comparison. MHRA, Digital mental health technology: device characterisation, regulatory qualification and classification, v1.3 (2026), https://www.gov.uk/government/publications/digital-mental-health-technology-qualification-and-classification: even with a medical purpose, "low functionality" (stores data "without change", "communicates data / information without change or prioritisation", shows fixed content) is not software as a medical device; an app that "alarms when digital biomarker crosses a threshold" indicating "crisis" is. FDA, Policy for Device Software Functions and Mobile Medical Applications (28 September 2022): functions under enforcement discretion include "Software functions that enable a patient or caregiver to create and send an alert or general emergency notification to first responders" and coping-skill aids for people with diagnosed psychiatric conditions; "lists of emergency hotlines" are not devices. FDA General Wellness guidance (6 January 2026): "relaxation or stress management" are wellness claims; "helps treat an anxiety disorder" is not. ANSM (France, updated 2026): software whose only purpose is "la communication de données sans fonction d'alertes auprès d'un médecin" is not a device; remote monitoring with alerts to a doctor is. HAS's mobile-health référentiel (2021, updated 2024) is voluntary but a good checklist: state the purpose and limits plainly (C12–C13), involve users in development (C11).

• Official guidance. Takeaway: the same line everywhere: sending the person's own message unchanged is communication; analysing data to raise an alert, or alerting a clinician, is the regulated side.

8.3 Data protection. GDPR (Regulation (EU) 2016/679) Art. 4(15), 9, 2(2)(c), Recital 18, 35; CJEU Lindqvist C-101/01, Ryneš C-212/13, OT C-184/20; EDPB Guidelines 07/2020 (controller and processor) and 05/2020 (consent); CNIL, Recommandation relative aux applications mobiles (24 September 2024), https://www.cnil.fr/fr/recommandations-applications-mobiles; CNIL, "Qu'est-ce ce qu'une donnée de santé ?" (2018); ePrivacy Directive 2002/58/EC Art. 9.

• The message, and even the fact of pressing a button known to mean "I am not coping", is very likely health data (Art. 4(15), read widely by the CJEU, including data capable of revealing health "by means of an intellectual operation involving comparison or deduction", OT, paras. 120–128). But a person sending their own message to a friend from their own device is correspondence, inside the household exemption (Art. 2(2)(c); Recital 18 names "correspondence and the holding of addresses"; Ryneš para. 32). The CNIL's 2024 recommendation settles the developer's side in France: when processing is "initié à la discrétion de la personne, opéré sous son contrôle et pour son seul compte" and done in a closed environment where the publisher "ne peut plus agir en aval sur les données", the publisher "ne fait que fournir un logiciel au service de l'utilisateur. Le RGPD n'est pas applicable au logiciel fourni"; it calls such apps good practice, and recommends processing location on the device, at the lowest precision needed, never in the background. Location sharing by the person's own act is outside ePrivacy Art. 9 (which governs network and service providers).
• Binding law; official guidance; CJEU case law. Unsettled at the edges (local-first open-source apps).
• Takeaway: keep the developer out of the path: no relay server, telemetry, crash report or analytics that could carry a pause event or its content, ever. Then the GDPR does not bind the developer and the person's use is household. Apply its standards anyway, as design (8.5). Location is off by default, per contact, read only at the press, never continuous.

8.4 Duty of care and liability. Directive (EU) 2024/2853 (new Product Liability Directive), Art. 2(2), Recital 14; French Code pénal art. 223-6 (from a secondary source, not checked on Légifrance). Martinengo L, Van Galen L, Lum E, et al. 2019. Suicide prevention and depression apps' suicide risk assessment and management: a systematic assessment of adherence to clinical guidelines. BMC Med 17:231. doi:10.1186/s12916-019-1461-z — verified. Larsen ME, Nicholas J, Christensen H. 2016. A systematic assessment of smartphone tools for suicide prevention. PLoS ONE 11(4):e0152285. doi:10.1371/journal.pone.0152285 — verified. Larsen ME, Huckvale K, Nicholas J, et al. 2019. Using science to sell apps: evaluation of mental health app store quality claims. NPJ Digit Med 2:18. doi:10.1038/s41746-019-0093-1 — verified. Martinez-Martin N, Kreitmair K. 2018. Ethical issues for direct-to-consumer digital psychotherapy apps: addressing accountability, data protection, and consent. JMIR Ment Health 5(2):e32. doi:10.2196/mental.9423 — verified. Miner AS, Milstein A, Schueller S, et al. 2016. Smartphone-based conversational agents and responses to questions about mental health, interpersonal violence, and physical health. JAMA Intern Med 176(5):619–625. doi:10.1001/jamainternmed.2016.0400 — verified. Parker L, Halter V, Karliychuk T, Grundy Q. 2019. How private is your mental health app data? Int J Law Psychiatry 64:198–204. doi:10.1016/j.ijlp.2019.04.002 — verified. Wykes T, Schueller S. 2019. Why reviewing apps is not enough: Transparency for Trust (T4T) principles. J Med Internet Res 21(5):e12390. doi:10.2196/12390 — verified. Samaritans. 2020. Managing self-harm and suicide content online: guidelines for sites and platforms. NICE. Evidence standards framework for digital health technologies (ECD7, updated 2022). WHO. 2023. Preventing suicide: a resource for media professionals, update 2023. Robinson J, et al. 2018 and 2023, #chatsafe Delphi studies, PLoS ONE 13(11):e0206584 and 18(8):e0289494 — both verified.

• Liability: the new Product Liability Directive "does not apply to free and open-source software that is developed or supplied outside the course of a commercial activity" (Art. 2(2)); fault-based national tort law remains. In France, failing to assist a person in danger (Code pénal 223-6) concerns people who know of the danger, which is the supporter, not the developer: a reason to tell supporters in advance what they might receive (5.6).
• What the literature found about apps: 6 of 69 suicide-prevention and depression apps (9%) gave a crisis number that was wrong, dead or led to a non-evidence-based organisation, two of them downloaded over a million times each; 94% gave emergency contact information (Martinengo). Among 49 interactive suicide-prevention apps, contacting friends and family was the most common feature (27) and access to crisis support the best evidenced; some contained harmful content such as lists of lethal means (Larsen 2016). 64% of mental-health apps in stores claimed effectiveness, one cited literature (Larsen 2019). In a direct-to-consumer app "there are no clear lines of accountability or associated ethical obligations to protect the user" (Martinez-Martin & Kreitmair). Voice assistants answered crisis statements "inconsistently and incompletely" (Miner). 41% of 61 mental-health apps had no privacy policy (Parker); 92% of 36 shared data with third parties (3.7).
• Guidance: signpost emergency and 24-hour services and trusted people; "Be concise and use clear and accessible language, because it can be hard to process lots of information when experiencing distress"; be "non-judgemental", avoid "making assumptions about the user" (Samaritans 2020). A tool for "communicating about health and care" with chosen people sits in NICE's tier B; one that monitors a condition or triggers "immediate or near-term action" is tier C; standard 9 asks for safeguarding of vulnerable users. Speak of suicide factually and point to help; never describe methods (WHO 2023).
• Law, peer-reviewed audits and ethics papers, charity and agency guidance. Takeaway: the duty that matters is ethical and practical, and it is concrete: correct numbers, honest limits, no data leaving the device, no promise of a response, a plain statement that Sioul is not an emergency service, and words tested with the people concerned.

8.5 Consent: explicit and revocable, even where no law requires it. GDPR Art. 4(11), 7, 9(2)(a), Recital 32; EDPB Guidelines 05/2020 on consent (paras. 64, 93, 98, 111, 113–114); the psychiatric advance-directive and joint-crisis-plan trials (5.2); WRAP (5.3).

• Sioul's rule is that nothing is sent without the person. The pre-set alert makes the press the act of sending, so the consent is given twice: when setting it up, and when pressing. The design standard borrowed from the GDPR and the EDPB: consent "freely given, specific, informed and unambiguous", by "a clear affirmative action" (no pre-ticked switch); the person sees exactly what will be sent, to whom, by which channel; consent can be withdrawn "at any time" and "as easy to withdraw as to give", "via the same electronic interface" and "without detriment"; consent is "refreshed at appropriate intervals"; a two-step confirmation can make explicit consent verifiable. The crisis-plan literature adds: written when well, with the people concerned, in the person's words, reviewed after use (WRAP: "when it is safe to stop this plan").
• Binding law applied as a design standard (official guidance; consensus from crisis-plan practice).
• Takeaway (inference): (a) setup shows the full message, the recipient and the channel, and is switched on by one deliberate action, with a short test to the person's own address or number if they wish; (b) the same screen switches it off, at once, at any time, and removing a contact sends that contact nothing; (c) every few months, and after each use, Sioul shows the setup once ("Still right?"), without blocking anything; (d) "vital interests" (Art. 9(2)(c)) is not a basis Sioul uses: nothing is ever sent without a press, so a rule such as "if I do not answer in 10 minutes, send" is out of scope, or at most a separate, explicit, revocable choice the person makes knowingly (open question 4).

Design criteria for Sioul¶

Numbered rules for a "Pause" feature, each with the entries it rests on. "Inference" marks a rule that combines sources or extends them to software; values marked "default" are starting points to test with the people concerned, not findings.

Setting up, on a calm day P1. Everything the pause does is decided beforehand, in a setup made on a calm day, unhurried and resumable, alone or with someone the person trusts: what is held, whether to message someone and what, do-not-disturb and who gets through, the person's own list of what helps, the breathing guide (off by default), and what happens when they come back. The press itself asks nothing. A try-out on a calm day, which sends nothing, lets the screen be familiar before it is needed (inference from 1.1e, "have a plan beforehand", and Sioul's rule that uncertainty fuels anxiety, research.md 5). — 1.1b, 1.1f, 1.3c, 3.4, 5.2 (plans made when well), 5.7 (Rodgers 2024: several sittings, with help), NICE NG225 1.11.8; inference. P2. The person's own words come first: their message, their list of what helps, their lines for the screen; Sioul's texts are editable drafts. — 4.2 (Chinn & Homeyard: tailored information), 5.3 (WRAP), 1.7a (safety plan in the person's words), 5.7 (MYPLAN users rewrote clinical texts). P3. Setup asks whether the supporter knows ("Have you talked to Sam about this?") and offers a short, editable note for the supporter: what a pause message means and does not mean, what helps, that Sioul does not watch the person or know where they are, that they need not answer at once, and where to call if they fear for the person's safety. — 5.2 (CRIMSON: plans unused by others do not work), 5.3 (carers' needs), 5.6, 8.4 (Code pénal 223-6); inference. P4. Consent is explicit, specific and revocable: setup shows the exact text, the recipient and the channel as one sentence ("When I press Pause, Sioul sends this text to Sam by SMS"); one deliberate action switches it on; an optional test send proves the number works; the same screen switches it off or changes it at once, at any time, during a pause too; a removed contact is never told; after each use and every few months Sioul shows the setup once ("Still right?"), never as a pop-up. — 8.5 (GDPR Art. 7, EDPB 05/2020 as design standard), 5.5 (CRPD general comment 1), 5.6 (abuse). P5. Setup says once, plainly: "Sioul is not an emergency service. It does not watch you and cannot tell how you are. It acts only when you press Pause." with the emergency number and the crisis line of the person's country. — 1.7g, 8.1, 8.4 (Martinez-Martin & Kreitmair; Samaritans; Parrish 2022).

Pressing it P6. One action, reachable without reading: a fixed button in Sioul's window and on the phone (a home-screen shortcut and a quick-settings tile, inference), the same place every time. Pausing never requires telling anyone; a person who also set an alert may prefer two distinct actions, "Pause" and "Pause and tell Sam" (the two levels of MYPLAN and of personal-safety apps; open question 17). The pause starts at once; only an outgoing message waits for a short, silent undo window (default 10 seconds, shown in words, "Your message to Sam leaves in a moment. Cancel", never as ticking seconds or with a sound), or, if the person chose it, a press-and-hold. — 5.6 (accidental SOS calls; Google's hold-to-confirm fix), 3.5, 1.1b. P7. The pause holds everything Sioul controls: no notification, no new mail, letter or task on screen, no reminder, no pause-to-move offer, no meal notice; the work planned in the paused hours goes to later days, never into the evening. Two exceptions, as during sleep: dose reminders and the alarms the person set keep coming, unless the person chose otherwise for the pause in setup. — the brief (the working day is not extended); 2.3; 6.1; health.md ("Sioul never silences a dose without that choice"). P8. A pause on one device is a pause on all of the person's devices running Sioul. — the brief ("puts everything on hold"); inference. P9. On the phone, do-not-disturb through an app-owned mode named "Pause", created at setup once access is granted: calls and messages from starred contacts, repeat callers allowed, alarms allowed, Sioul's dose channel allowed; Sioul never edits the person's stars but opens the Contacts app's starred list for them; ending the pause turns off only Sioul's mode, never one the person started. Emergency services call back from an unknown number (in France, 0 800 112 112 since October 2024), and no exception for such callbacks was found in Android's call filtering (ZenModeFiltering, Telecom's Ringer: not exhaustive): repeat callers let a second call through within 15 minutes, and setup can suggest starring the national callback number. — 7.1, 1.7k; inference. P10. On a computer, Sioul holds its own notifications always, and sets the desktop's do-not-disturb only where the platform lets an application do so, saying plainly what it could and could not silence. — 7.5.

The pause screen P11. The screen states facts about Sioul, never claims about the person: "Paused. Sioul is keeping everything on hold. Nothing new will show here until you come back." No "you are safe", no diagnosis, no "we", no idiom. — 2.1 (PFA: "if it is true"), 4.1, 4.2, 1.2a, i18n.md. P12. At most about 40 words visible, one point per sentence, literal, present tense; no question, no instruction, no praise, no exclamation mark, no countdown, no time elapsed, no count; nothing moves unless the person started it; the same layout every time. — 4.2 (COGA), 3.3, 3.5, 3.6, 4.3, 4.4 (WCAG 2.2.2). P13. The emergency number and the crisis line of the person's country are always on the screen, in one calm line, one tap to call, with written options where they exist (3114 by phone; Shout 85258 and 988 by text; 114 for those it serves); the numbers live in one data file with the date and source of each check, are checked before every release by a test, and the person can add their own. — 1.7e, 1.7i, 1.7k, 8.4 (Martinengo: 9% wrong numbers), 5.7 (Bennett-Poynter: broken links). P14. The lock covers Sioul's content only: the phone, other apps, calls and the system's emergency dialling stay free; one visible button ends the pause; no PIN, no puzzle, no unlock step; an optional hold-to-leave against accidental exits; on a locked phone the screen may stay visible but shows nothing private. — 4.3, 7.3, 2.2 (self-efficacy), 5.7 (Stress Autism Mate: logins stressed users). P15. The person's own list of what helps can be unfolded (or their safety plan, if they wrote one, one step at a time in its order), and its items can open what the person chose (a playlist, a film, an app, a folder of photos); Sioul adds no generic advice and no library to browse. — 2.6, 2.7, 1.1e, 1.7a, 1.7l, 5.7 (Virtual Hope Box; "guidance shortfall" in large libraries). P16. The breathing guide is off by default; turned on, it starts only when tapped, shows a slow, low-contrast shape with a longer exhale, at a pace set in settings (about five to six breaths a minute to start), with no counting text and no "deep breaths"; a tap stops it; with reduced motion it becomes a still image or, on the phone, an optional gentle vibration; it is never suggested to someone who marked breath focus as unhelpful. — 2.5, 4.4, 1.2b, 5.3 (Chalmers 2022: no specific breathing strategies imposed). P17. Grounding, if wanted, is one short line written or chosen by the person, never a counted exercise. — 2.4.

The message P18. Text by default: SMS on the phone, mail from the person's own account on a computer, or Matrix if they use it; a call request only if the person chose it. The message opens by saying it was prepared in advance, gives the state in the person's words, what would help, what not to do, and when they will be in touch; it carries no location unless switched on for that contact (read once, coarse unless chosen), no diagnosis, no "sorry to bother you", nothing from Sioul's content. — 5.3, 5.4, 1.1g–h, 1.7j, 5.6 (abuse), 8.3 (CNIL: location on the device, minimal). P19. Sending is reported honestly and stays on screen: "sent at 14:32", "handed to your SMS app", or "not sent: no network", with a retry and the way to send it by hand; never "delivered" without a delivery report; never "help is coming". — 4.1, 7.2, 8.4. P20. No escalation by Sioul: no second contact, no emergency service, no repeat, no "if no reply" rule, no read receipts shown; anything more needs a new press by the person, and a person who wants two contacts at once sets that beforehand. — 1.7g, 5.6, 8.5 (no vital-interests sending). P21. Nothing about a pause ever reaches the developer: no server, relay, analytics, crash report or log carries a pause event or its content. Sioul says what does leave the device: the message, to the chosen person, through the person's operator or mail provider, and that a copy stays in the phone's Messages app or the Sent folder. — 3.7, 8.3 (CNIL 2024), 5.6.

Coming back P22. The pause ends when the person chooses, or at a clock time they chose beforehand; it never ends with a sound. — 3.5, 6.1. P23. The exit asks nothing about what happened. It says: "You are back. Nothing was lost."; who was told and when; that the rest of today is lighter (the bad-day level, about 60 %, or rest, if allowed in setup); one optional offer, default no, to lighten tomorrow; an optional "I am OK for now" message to the supporter, never sent automatically. — 3.1, 3.2, 6.1, capacity criterion 36, 5.4. P24. What was held comes back slowly: the Porch stays closed until the next admin window unless the person opens it; Now shows one small next step or "nothing is needed today"; the pause and the day after never lower the long-term budget unless the person's evening review says the day was too much. — 6.1, 1.3c, capacity criteria 20 and 36. P25. No reflection prompt, no journaling prompt, no rating of the pause; the evening review does not mention it; a note in the person's own Notes remains possible, unprompted. — 3.1, 3.2, 6.2. P26. No count, streak, chart or badge of pauses; a plain local record (date, what was held, what was sent to whom) that the person can see and delete, kept for undo and their own reference. — 3.6, 3.7, 6.3. P27. Professional help is suggested only after several pauses for the same kind of reason within a few weeks (threshold to set with users) or when the person asks, once, never on the exit screen, as a practical offer ("Sioul can draft a message or an appointment request for you to check") with "Not now" and "Do not suggest this again"; the crisis line itself is never rationed. Sioul sends no "checking on you" messages of its own. — 6.4, 5.3 (Chalmers 2022), Doherty 2022, Miller 2007, 4.1.

Words and claims P28. Call it "Pause" ("Pause", "En pause"), not "Emergency" or "SOS", which suggest an emergency service. — 8.1. P29. Describe only what the software does, everywhere Sioul speaks (app, README, website, store listings): no detecting, monitoring, preventing, treating, calming, managing or keeping safe; no effect sizes borrowed from safety-plan or crisis-plan trials. — 8.1, 8.2, 8.4 (Larsen 2019), 5.7. P30. The button is the only trigger: Sioul never infers a crisis from typing, sleep, heart rate, missed doses, words in mail or anything else. — 1.7g, 3.7, 8.1 (MDCG: detection plus alarm is a device), 8.4 (Miner 2016). P31. Before release, the texts and the setup are tried with autistic, ADHD, anxious and burnt-out adults, in English and French written natively, and changed by what they say. — 4.1 (trauma-informed collaboration), 8.2 (HAS C11), 8.4 (Samaritans: lived experience); inference.

Open questions and conflicts between sources¶

1. Text or voice for support. Hearing a familiar voice relieved stress hormones in children where instant messages did not (Seltzer 2012); many autistic adults find calls the barrier and lose speech under stress (1.1g–h), and a response call raised anxiety in Brain in Hand's users (5.7). Text is the default here; the person can choose a call request. Untested for adults in overload.
2. Company or solitude. Autistic adults often go somewhere alone to avoid harm (Lewis & Stevens 2023), some autistic youths wanted company and felt imposed isolation as punishment (Phung 2021), and contact with others is protective in suicidal crises (1.7e). Solved by the person's prior choice, not by a default; which default to ship is open.
3. Breathing. WHO's first-aid guide, the NAS and an NHS trust page include slow breathing; breathing retraining added nothing to panic treatment and may hinder some (Schmidt 2000), the 2022 first-aid guidelines rejected specific breathing strategies, and meditation-related adverse effects run at 6–14% for lasting ones (Britton 2021). Off by default here; whether to offer it at all in the pause, or only as one of the person's own items, is open.
4. An automatic second step. Some apps send without a press (Google's Safety Check after 60 seconds without answer). Sioul's rule is that nothing is sent without the person; "if I do not answer in N minutes, send" would help the rare person who presses and then cannot act, and would make Sioul a monitoring tool (8.1, 8.5). Out of scope unless people ask for it, and then only as a separate, explicit choice.
5. The undo window. Long enough to catch accidental presses (5.6), short enough not to delay someone in a short suicidal crisis (1.7e). 10 seconds is a guess; whether a press-and-hold is better for autistic users with motor or sensory differences is untested. A silent undo window is still a countdown (3.5).
6. What to keep. A local record helps undo and the person's own understanding (and the suggestion of help in P27 needs a count somewhere), but any record is health data and visible to someone holding the device (3.7, 8.3). Keep by default, delete on request, or keep nothing: open.
7. Specific or vague message. "I am overloaded" helps the supporter respond; "I have paused everything" protects privacy (5.4). The person chooses; the default wording is open.
8. When to suggest professional help. The first-aid panel's lived-experience members rejected advice after a single panic attack and doubted later check-ins (37.8% endorsed); caring letters after suicidal crises lowered suicide in one large trial (Motto & Bostrom 2001) but not across trials (Milner 2015). The threshold in P27 needs the people concerned.
9. Crisis line for everyone. Showing it does not raise risk (Blades 2018), but someone who paused because of noise may find it alarming or patronising. Shown here in one quiet line; to test.
10. Safety behaviour. For panic disorder, escape can maintain fear (Helbig-Lang & Petermann 2010) while judicious use can help (Rachman 2008); for sensory overload, withdrawal is the protective act (Lewis & Stevens 2023). The pause sits on both sides; Sioul frames it as protection from overload and tasks, not as a treatment.
11. Distraction. Self-chosen distraction helps in the moment (Webb 2012), while a helper distracting someone was rejected by lived-experience panellists as possibly patronising (Chalmers 2022). Hence the person's own list, never Sioul's suggestions.
12. Red for urgency. MYPLAN's users asked for a red alarm button; BackUp avoided red; Sioul avoids red for lateness. Not tested; the pause button stays in Sioul's palette.
13. A safety plan inside Sioul. Safety plans work as part of clinical care with follow-up (1.7a–c); app versions show coping benefits, and no trial has yet shown an effect on suicidal acts (5.7). Whether Sioul should offer a full safety plan, link to dedicated apps (BeyondNow, Stay Alive) or only hold the person's own list is open; P15 keeps the minimum.
14. SMS on Android. Sending directly needs SEND_SMS, which Android 15 hides behind "restricted settings" for sideloaded apps (7.2); the pre-filled SMS needs one more tap in another app at the worst moment. Which to make the default, and whether Matrix or mail can replace SMS for most people, is open.
15. Dim or familiar. Whether the pause screen should be darker than Sioul's usual theme (less light) or identical (predictable) is untested (4.3).
16. The phrase "holding the fort". The brief's warm idiom breaks Sioul's literal-language rule (i18n.md) and COGA's pattern; the literal version may feel colder. To test both with users.
17. One button or two. Fear of bothering someone keeps alarm buttons unpressed (Stokke 2016), which argues for a pause that tells no one by default and a separate "Pause and tell Sam"; but two buttons are a choice at the worst moment (3.4), and one pre-set behaviour asks nothing. MYPLAN's users chose two levels; untested with autistic or ADHD adults.

Sources checked (verification log)¶

• Method (2026-10-06, 13:08–15:30): every DOI in this note was resolved on Crossref (title, first author, year, venue) and, where indexed, its abstract read on PubMed or OpenAlex; "verified" in an entry means that check matched. Full texts read: Phung et al. 2021, Kessels 2003, Chalmers et al. 2022 (MHFA panic guidelines), the WHO psychological first aid guide (2011), the abbreviated WRAP workbook, and, for the app and legal entries, the papers and documents marked "full text read". Platform facts come from developer.android.com, AOSP source (android.googlesource.com), GNOME and KDE source, learn.microsoft.com and developer.apple.com, with URLs in section 7; legal texts from the Publications Office copies of EUR-Lex texts, the European Commission, gov.uk, fda.gov, ansm.sante.fr, has-sante.fr, cnil.fr and edpb.europa.eu.
• Numbers (1.7k) were checked at the services' own pages or national government pages on the same day: 3114.fr, service-public.gouv.fr, info.urgence114.fr, digital-strategy.ec.europa.eu, samaritans.org, nhs.uk, giveusashout.org, relayuk.bt.com, 988lifeline.org, 988.ca, textwith911.ca. The FCC's text-to-911 page refused automated access (not checked there).
• Not checked at the primary source, marked in the entries: CDC's "Psychology of a crisis" chapter; ICD-11 QE84 wording (from a mirror of the WHO text); the NAS meltdown guide's sources; Treleaven 2018 and Inclusion Europe 2009 (not re-read); French Code pénal 223-6 (secondary source); the consolidated MDR text (original 2017 text read); the original UK police statement on Android SOS calls (press reports); /e/OS statements (community forum); RCS restrictions (press). Abstracts not read: Helbig-Lang & Petermann 2010, Swanson et al. 2006.
• Vendor and press material is marked "opinion" or "vendor": Brain in Hand help pages, Rootd, Wysa, Headspace, Stay Alive's commissioned evaluation, Be Safe, Calm Harm, Google's, Samsung's and Apple's help pages, and the press reports on accidental emergency calls.
• Left out, for lack of a checkable figure: a suicide-specific mortality ratio in autism (Hirvikoski et al. 2016, Br J Psychiatry 208(3):232–238, doi:10.1192/bjp.bp.114.160192, verified, reports raised mortality in autism in its abstract but no suicide figure there); "rejection sensitive dysphoria" (no validated measure found).
• Final pass (2026-10-06, 15:50): all 172 DOIs cited in this note resolved on Crossref.
• Platform claims not checked at an official page, marked in section 7: RCS sending, WhatsApp and Signal pre-fill, GNOME's lack of a permission check on its setting (inferred), the Shortcuts "Set Focus" action, /e/OS statements (forum), vendor builds' handling of Android 15's restricted settings, and the absence of a do-not-disturb exception for emergency callbacks (AOSP search not exhaustive).
• Read as abstract or summary only, as stated in the entries: among others Lewis & Stevens 2023, Howard & Sedgewick 2021, Rachman et al. 2008, Torok et al. 2020, Kuhn et al. 2017, Bush et al. 2017, Kang & Reynolds 2024, Bowie-DaBreo et al. 2022, Rothman et al. 2026 (summary); MYPLAN's focus-group finding on prewritten messages comes through the Gryglewicz et al. 2024 review.
• No reference in this note was invented. Where a claim rests on a page that could not be read, it says so.