Prism Break is built on an idea… and all ideas can become a revolution, if you want them to. The idea is this: a person who feels broken is, more often than they know, refracted… not in pieces, but carrying every colour at once, waiting for the right glass. Prism Break (prismbreak.art) is a small, free web app built on that idea, for people living with the aftermath of trauma, including the long form clinicians call complex PTSD. It is a companion, not a treatment. It helps a person ground, settle, and do small kind things; it does not diagnose, does not process traumatic memory, and does not replace therapy. There are no accounts, nothing is tracked, and it costs nothing.
The name carries the design idea. Traumatic distress often arrives undifferentiated: everything at once, a flood with no handles. Newton showed that a prism adds nothing to light; it only separates what was already there. The app makes the same deliberately small claim. It fixes nothing. It separates an overwhelming state into seven specific, manageable things a person can actually do: breathe, come back to the senses, eat something kind, notice one good thing, make something, watch the sky, write one moment down. Prism Break began as one instrument of a larger digital library built by the same maker; this paper describes the app alone, and the app stands alone.
The app opens on a dial of seven stained glass panes rather than a menu. Each pane opens one section, which the app calls a window. Three principles run through everything behind them.
Agency. Trauma is, at its root, the experience of having no say. The interface is built as a small rehearsal of the opposite: nothing on the dial moves until the user moves it, and nothing opens until they press. Agency rehearsed in small, safe doses is a reasonable first principle for a nervous system relearning that it has hands.
Weather, not numbers. The mood system is seven weathers (storm, rain, fog, cloud, breeze, sun, rainbow) rather than a rating out of ten. A self-rating invites comparison, tracking, and the quiet arithmetic of failure; weather arrives, does its work, and passes, which is an accurate and useful model of affect. The distinction the app draws is precise: numbers that count the world (five things you can see, a twelve minute brew) are used freely, because they are handholds. Numbers that rate the person (scores, ratings, streaks) appear nowhere, because they are verdicts.
No ornamental dishonesty. The dial’s star ring shows the seven classical planets at their true current positions, and the dream window’s moon shows tonight’s real phase. For users whose trust has been broken, even decorative falsehood is a cost, so the app’s poetry is kept riding on real physics throughout.
One more thing underpins the whole instrument, and a careful reader will notice it anyway, so it should be named. The maker’s working hypothesis is that the sciences and the older, symbolic traditions of meaning are not rivals but different registers of the same human work. People have always steadied themselves with story, symbol, sky and ritual, and the app deliberately keeps those old instruments in the room alongside the evidence: a dial carrying the seven classical planets, archetypal costumes in the art window, clouds read for shapes the way skies have always been read, a dream handed back as a haiku. None of it is in your face. It is woven through quietly, as atmosphere rather than argument.
The rule that keeps the weave honest never bends: the symbolic is used as material for meaning, never as a claim about the world. The planets stand at their true astronomical positions and predict nothing. The archetypes are projective costumes, never diagnoses. The clouds are pareidolia, employed knowingly. Wherever an old form appears, the science described in this paper carries the weight beside it. And the weave runs all the way to the present day: the oldest human technologies of meaning, story, symbol and nightly ritual, are here run on the newest technology there is, a large language model, which this paper names in full below. The ancient carried by the most modern is not an accident of the build. It is the signature.
Much of what follows is built as play: a wordless game of gold seams, a cup of tea made precisely to taste, clouds to find shapes in, a crystal to turn under a finger. This is a deliberate clinical position rather than decoration, and it stands on several distinct lines of evidence.
Play is a primary process, and it requires safety. In mammals, play is not a luxury behaviour but one of the brain’s basic emotional systems, and it has a defining precondition: animals play only when they feel safe enough to, and stop the moment they do not (Panksepp, 1998). That makes play a double instrument for this population. An activity that requires a felt sense of safety also practises it, and an interface that invites play is quietly signalling, below the level of argument, that the environment is safe.
Games restore what trauma takes. The motivational research on games finds that their pull comes chiefly from satisfying the needs for autonomy and competence (Ryan, Rigby & Przybylski, 2006). A game is a small world in which actions reliably produce effects, effort visibly accumulates, and failure is survivable. For a person whose larger world taught the opposite lessons, that is not escapism; it is rehearsal, and it is the same agency principle that shapes the dial, expressed as something you can actually do with your hands.
Absorption is regulation. Games reliably produce flow, the state of full absorption in a manageable challenge (Csikszentmihalyi, 1990). Attention absorbed in a task is attention unavailable to rumination and intrusion; the Tetris findings described under the red window below are a sharp clinical instance of this general mechanism.
Play is how hard material becomes approachable. Winnicott (1971) placed play at the centre of therapeutic change: it is in playing that a person can approach difficult material sideways, at a depth they choose, without being overwhelmed by it. Several of the app’s activities are built directly on that grammar, meeting each user only as deep as they are ready to go.
And game-framed self-help has direct trial evidence. A randomised controlled trial of SuperBetter, a game-framed self-help app, found significantly greater reductions in depressive symptoms than a waitlist control (Roepke et al., 2015). The design lineage of using games deliberately as a scaffold for recovery and post-traumatic growth runs through McGonigal’s work (McGonigal, 2015), and Prism Break sits gratefully in that family: smaller in scope, and holding to the same conviction that a game can be a serious instrument wearing a light disguise.
Red is the anchor window and is deliberately nearly empty, because a person in overwhelm cannot choose between nine options; choice itself is a burden. It holds three things.
A breathing guide runs a longer exhale than inhale. The rationale is physiological: heart rate slows slightly on each exhalation (respiratory sinus arrhythmia; Grossman & Taylor, 2007), because the exhale is when vagal influence on the heart is strongest. A frightened person cannot will their pulse down, but they can lengthen an out-breath, and the pulse follows. It is the one part of the body’s alarm system reliably within voluntary reach.
A five senses exercise walks the standard 5-4-3-2-1 grounding sequence used widely in trauma care: five things seen, four heard, three touched, two smelled, one tasted, ending in the room the person is actually standing in.
Goldwork is a slow, wordless visuospatial game: seating shaped pieces, seam by seam, into broken forms, in the visual language of kintsugi. It is built squarely in the family of the Tetris studies. Holmes and colleagues showed that playing Tetris after viewing distressing film reduced intrusive memories in the following week (Holmes et al., 2009), that a verbal task produced the opposite effect, isolating the visuospatial channel as the active ingredient (Holmes et al., 2010, which coined the phrase “cognitive vaccine”), and that the effect survived contact with the real world when Tetris was given in an emergency department within six hours of trauma (Iyadurai et al., 2018). The boundary is drawn honestly: those trials dosed the intervention in the hours after the wound, under supervision. Goldwork simply keeps the same wordless channel open for whoever reaches for it, whenever an intrusive image comes, and claims nothing more than that.
Red is also where the app’s deepest safety rule was enforced during development. Trauma care draws a bright line between stabilisation, which is safe to practise alone, and processing, the deliberate reopening of memory, which requires a trained clinician (the phase-based consensus following Herman, 1992, and the ISTSS complex PTSD guidelines). An early bilateral stimulation exercise was removed from this window for exactly that reason: it edged toward processing. The app grounds, regulates, and lets a person express. It never processes.
Trauma and disordered eating are strongly comorbid (Brewerton, 2007), which makes conventional food tracking machinery, calories, goals, weigh-ins, a loaded instrument for this population. The nourishment window therefore contains no measuring numbers at all. It asks only what the day feels like and offers food suggestions matched to that weather: something warm, something kind, enough water.
It also contains a deliberately playful tea ritual: vessel, brew, strength, milk, temperature, all chosen by the user, with nothing depending on any of it. This is agency in miniature, the same first principle at the lowest possible stakes: choices that are real, in a frame where nothing can be failed. (The international standard for brewing tea, BS 6008, sits behind a link as a smile, alongside the only sentence that matters: the formula that counts is your own.)
The brain weights negative information more heavily than positive: threats are registered faster, weighted heavier, and remembered longer than kindnesses of equal size (Baumeister, Bratslavsky, Finkenauer & Vohs, 2001, under the accurate title “Bad is stronger than good”). In a nervous system already tuned to alarm, this bias compounds. The yellow window trains the deliberate noticing of small safety cues, borrowing the word “glimmer” from Dana (2018). The wider polyvagal framework the word comes from is scientifically contested, and the app borrows it as vocabulary only; the practice itself is simply attentional bookkeeping correction, not positive thinking.
The window also holds the app’s only door to the news, kept deliberately behind a gentle threshold because the app is opened on people’s worst days. Behind it, an AI summarist gathers the morning’s real reporting and retells it once, calmly, as a single story rather than a feed, told from a little higher up, with a quiet thread of the older register of meaning described earlier woven through the telling. Its rules are strict: it may select and reframe but never invent; every story originates in real human journalism and links back to it; and on days when the wires hold nothing kind, it falls back to evergreen factual material instead. One story, then an ending, because a feed pours and a story ends.
Traumatic memory is stored heavily in image and body rather than in language. In an early and much cited neuroimaging study, Broca’s area, the brain’s speech production region, showed reduced activation while participants relived traumatic memories (Rauch, van der Kolk et al., 1996); it is a single early study, and the paper leans on it only lightly, but it named something clinicians see daily: a person can know their story perfectly and still be run by it, because the part that knows and the part that suffers are not on speaking terms. Art making enters through the sensory channel where the material actually lives. It is also calming at the simplest physiological level: forty-five minutes of art making measurably lowered cortisol in roughly three quarters of participants, with prior skill making no difference (Kaimal, Ray & Muniz, 2016). The window’s motto, “the wonkier the better,” is that finding worn as an invitation.
The activities are built for graded depth. The keystone exercise asks the user to crumple a page, smooth it out, and trace every crease in colour. A person who needs it as metaphor will find kintsugi in their own hands; a person who just wants to make creased art makes creased art. Nobody is forced below their own waterline, which is how a sub-verbal layer is approached safely in an unsupervised setting.
The window offers twelve classical archetypes (the Gardener, the Warrior, and so on) as perspectives to make from. Their scientific standing is stated honestly: as biology or prediction, archetypes have been rightly demoted in the academy. Their one legitimate working use is the one this window makes of them: projective material in art practice, a costume lent for a session, never an identity assigned. For the same reason, astrology as prediction appears nowhere in the app: handing a person’s locus of control to the sky runs opposite to trauma recovery, so the app keeps the real astronomy and leaves the fortune telling out.
The blue window practises optional cognitive reappraisal. The user writes the heavy thing as a grey cloud, and is offered, never required, the chance to turn it and write a silver lining. Declining is a first-class outcome: the cloud stays grey, and the app says that is allowed. A reappraisal demanded of someone functions as denial; a reappraisal offered and declined preserves both honesty and the option to return.
The window also holds cloud gazing (soft, machine-dreamed clouds, drifting one at a time, with the question: what do you see in it?), employing pareidolia as gentle projective play. Beneath it sits one line of graded encouragement toward the real sky: pop your head out the door if you can, and if not, this sky came to you. The principle is graded exposure with the person setting every rung: the door is always pointed at, never pushed.
The dream catcher is a dream diary with one addition. The user records a fragment of a dream, and the app returns a three line haiku that condenses it. The design follows three constraints: the haiku condenses but never interprets, since dream interpretation has no reliable evidence base and meaning is left with the user; it never alters or softens the dream’s content; and it adds no commentary. If an entry appears to describe present distress rather than a dream, the haiku is still returned, with crisis line information displayed directly beneath it.
The rationale is twofold. Dream recall decays rapidly after waking, which makes prompt written capture the basic tool of dream research. And condensing a difficult experience into a few plain words draws on affect labelling: putting feelings into words measurably reduces the brain’s threat response (Lieberman et al., 2007). A haiku is exactly that: a few plain words, with no room in them for a verdict.
Dreams get a window at all partly because of the overnight therapy hypothesis: REM sleep appears to soften the emotional charge of experience while keeping its content (van der Helm, Walker et al., 2011). The app holds this lightly, as young science. The catcher’s own job is smaller: making sure that when the night does its work, somebody wrote down what it said.
The deepest window rests on a finding nearly a century old. Zeigarnik (1927) observed that interrupted, unfinished tasks were remembered roughly twice as well as completed ones, a result the following century has sometimes replicated and sometimes not, as delicate findings are. The mind keeps its unfinished business in a kind of standing tension, re-presenting it unbidden; a person with an unresolved story knows the phenomenon as three in the morning. The hopeful modern coda is that the mind does not require the task to be finished to release its grip; it requires a plan. Writing down specifically how an unfinished thing will be handled measurably quiets its intrusions (Masicampo & Baumeister, 2011). There is also a long evidence base behind structured writing about hard experience itself: the expressive writing paradigm (Pennebaker, from 1986; meta-analysis in Frattaroli, 2006) shows small but persistent effects on health outcomes.
The mechanic built on this: the user writes one hard moment plainly, chooses a single interpretive light, and receives back a short retelling that finds a kinder meaning in the same facts. The eight lights are labelled with intent. Three (Radical Kindness, Radical Acceptance, Radical Forgiveness) nod to dialectical behaviour therapy, where radical acceptance is a core distress tolerance skill (Linehan, 1993). Others wear the paper’s older register (Karmic, Perfect Balance, The Long View) or open mischief (Cosmic Giggles), in the same spirit in which DBT deliberately deploys irreverence: a lens a person can smile at is easier to pick up on a bad day, and humour is itself a documented reappraisal strategy (Samson & Gross, 2012). And the eighth stands apart from the spectrum on purpose: Radical Amazement, the white light, unsplit. The name borrows Heschel’s radical amazement, wonder at the sheer fact that anything exists at all, and the lens stands on the modern awe research: awe reliably shrinks the ruminating self and recontextualises what it holds (Keltner & Haidt, 2003; Piff et al., 2015). Under it, a hard moment is retold against the scale of existing at all, with one explicit rule: wonder may sit beside pain, and may never wave it away. The engine is borrowed from imagery rehearsal therapy, the best evidenced treatment for recurring nightmares and recommended first line by the sleep medicine academies (Aurora et al., 2010; Morgenthaler et al., 2018). Two boundaries are stated plainly. IRT is normally therapist-delivered; the app borrows only its quiet engine, rewriting meaning rather than treating nightmare disorder, and widens the practice from dreams to waking memories. And the facts of a memory are never touched: what changes is the meaning, never the event. There is a younger line of science underneath, memory reconsolidation (Nader, Schafe & LeDoux, 2000), suggesting that a recalled memory is briefly malleable and is re-stored with the emotional context of its retelling; the app treats this as an encouraging rhyme, not a proven mechanism.
Two further design rules govern the window. Moments are placed in one of eight user-claimed life eras (origin, flying the nest, survival, hermit, healing, love, rebirth, future), which are phases a person claims rather than dates, since a life is not a timeline; the eighth, the future era, is written as if already true, deliberately opening a good loop rather than closing a bad one. And the retellings are forbidden the phrase “one day”: endings must land in the present tense, because deferring hope to someday defers it indefinitely, and the present is the only place a nervous system can stand down.
The retellings and the dream haikus are written by an AI, and the app hides nothing about it. It is a large language model, Claude, made by Anthropic. It was not trained by the app’s maker, and it is never trained on what users write: a moment or dream fragment is transmitted once, answered, and neither retained on any server nor used for training. Everything else a window remembers (a weather, a kept story, a dream) is stored only on the user’s own device. The voice itself stands on a wide substrate: a model of this kind has read across the whole human library, the sciences and the older symbolic traditions alike, which is what allows the retellings to speak in both of the registers this paper has described, the weight of the evidence underneath and the older language of meaning at the surface. The app adds no training of its own. It only chooses, by instruction, which registers the voice may speak in and which laws bind it.
The maker’s work was writing the laws that bind the model’s conduct, and they are enforced in its instructions on every request. It may offer a kinder meaning and may never deny the wound; a reframe that argues with the user’s pain is forbidden outright. It never alters the facts of what was written. It never tells the user what their life means over their own reading of it. It never uses “one day.” And it is explicitly forbidden sycophancy: acceptance of the person is not agreement with their despair, and the model is instructed to hold despair rather than echo it.
The design ideals here are borrowed, with attribution, from person-centred therapy: congruence, unconditional positive regard, and empathy (Rogers, 1957). The borrowing is hedged honestly. Rogers claimed the three conditions were sufficient for change; the decades since have disagreed, particularly for deep trauma, which is one more reason this app does not present itself as therapy. As design ideals for a listening machine, rather than as a complete treatment, they have held up as well as anything in the field.
The shadow is acknowledged rather than avoided: unconstrained AI companions have caused documented real world harm, including by agreeing with despair. That risk is precisely why the laws above were written before the feature was built, why the app never plays therapist, and why crisis lines are permanently visible. There are practical rails as well as laws: requests to the model are rate limited, input length is capped, and the model’s credentials never leave the server. When the model cannot be reached, the app does not simulate it; a fallback assembled plainly from the user’s own words appears, labelled as such.
The app’s refusals are design decisions, each with a reason. It never processes traumatic memory, because processing belongs with a trained clinician. Nothing is tracked and there are no analytics, because surveillance is the posture of what harmed many of the people this was built for. There are no streaks or badges, because a broken streak punishes the user for being unwell. There is no account, because a form in front of a door closes it. It is free, funded by optional donations that confer nothing on the giver. And it does not scan what users write for risk. That last is a deliberate trade, made in the open: risk scanning would make the app a surveillance instrument for its most vulnerable users. Instead, crisis lines (Samaritans, 116 123; SHOUT, text 85258) are visible on every page, always. The app knows what it is not, and keeps the number of what is needed instead.
One thread runs through every window. Trauma can be described as a disorder of time: a flashback is the past arriving in the present tense, and dread is the future doing the same. Each of the app’s tools performs some version of a single move, returning attention to the present, which is the only place a nervous system can actually stand down. The senses walk timestamps the present by hand; the breath does it rhythmically; the weather model does it by design, since weather is only ever today’s; and the violet window does it grammatically, by requiring endings in the present tense. Trained present-moment attention has a measurable steadying effect on distress (the mindfulness research programme, from Kabat-Zinn, 1982, onward). The app builds the same move into its furniture rather than teaching it as doctrine.
The visual language carries the argument nonverbally. Each pane of the dial is a mosaic: fractured glass rejoined along every break with seams of gold, the visual grammar of kintsugi, so the app states its thesis (broken, joined, and more radiant along the joins) before a single word is read. The metals are copper and rose gold rather than device grey; the interface labels are lowercase so that nothing shouts; the sounds are soft and mechanical, a dial that clicks like a safe being listened to. In a hypervigilant reader, who scans everything below awareness, interface tone is not decoration; it is part of the intervention.
Prism Break has not been trialled. Every mechanism in it is borrowed from better tested neighbours, and it claims no clinical outcomes of its own. It is not a medical device, not therapy, and not a crisis service; it is a companion built from the findings above, designed to sit on the safe side of every line it touches. Data lives in the browser’s local storage, which the user can export but the app cannot back up, a further consequence of the no-accounts decision, stated to users plainly.
A last word on lineage. Centuries ago, natural philosophers built instruments to explore the outer world: prisms, telescopes, orreries, astrolabes; glass and brass that made the invisible visible. Prism Break is one of a series of what its maker calls philosophical instruments, built in that tradition and pointed the other way, toward the inner world. The old Hermetic formula put it in four words: as above, so below. It is read here not as metaphysics but as a design brief: the same care, the same glass, the same honest calibration once turned on the heavens, turned toward the weather inside a person. And it is no accident that this instrument is named for the oldest of them all: the prism, turned inward.
Not a cure, not a course, not a coach. A small instrument for bad days, built from borrowed and honestly labelled science, that behaves in every window as if the person holding it deserves gentleness. That is the only claim it makes, and every design decision above exists to keep it.
Every claim above stands on a named source. What is solid is stated plainly; what is young science is held lightly on purpose.
Newton, Opticks (1704): the prism reveals, and does not add.
World Health Organization, ICD-11 (2018): complex PTSD as its own clinical category.
Panksepp, Affective Neuroscience (1998): play as a primary emotional system in mammals, expressed only under felt safety.
Ryan, Rigby & Przybylski, Motivation and Emotion (2006): the motivational pull of video games; autonomy and competence.
Csikszentmihalyi, Flow (1990): absorption in manageable challenge.
Winnicott, Playing and Reality (1971): play as the site of therapeutic change.
Roepke et al., Games for Health Journal (2015): randomised controlled trial of SuperBetter; game-framed self-help reduced depressive symptoms.
McGonigal, SuperBetter (2015): games as a deliberate scaffold for recovery and post-traumatic growth.
Grossman & Taylor, Biological Psychology (2007): respiratory sinus arrhythmia and vagal influence on the exhale.
Holmes et al., PLoS ONE (2009): Tetris after distressing film reduces intrusive memories. Holmes et al., PLoS ONE (2010): visuospatial Tetris against a verbal pub quiz, and the phrase “cognitive vaccine”.
Iyadurai et al., Molecular Psychiatry (2018): Tetris in the emergency department, within six hours.
Herman, Trauma and Recovery (1992), and the phase-based consensus after it: stabilisation is safe alone, processing needs a clinician.
Brewerton, Eating Disorders (2007): the comorbidity of trauma and disordered eating.
Baumeister, Bratslavsky, Finkenauer & Vohs, Review of General Psychology (2001): “Bad is stronger than good”, the negativity bias.
Dana, The Polyvagal Theory in Therapy (2018): the glimmer, borrowed as vocabulary; the wider theory is contested and is not relied on.
Rauch, van der Kolk et al., Archives of General Psychiatry (1996): reduced Broca’s area activation during traumatic recall; one early imaging study, hedged accordingly.
Kaimal, Ray & Muniz, Art Therapy (2016): forty-five minutes of art making lowers cortisol in most participants, skill irrelevant.
Lieberman et al., Psychological Science (2007): affect labelling; putting feelings into words reduces amygdala response.
van der Helm, Walker et al., Current Biology (2011): REM sleep and the softening of emotional charge; the overnight therapy hypothesis, held lightly.
Zeigarnik (1927): the unfinished, remembered. Masicampo & Baumeister, Journal of Personality and Social Psychology (2011): a plan, not completion, quiets the loop.
Pennebaker, from 1986 onward, and Frattaroli’s meta-analysis, Psychological Bulletin (2006): expressive writing’s small, persistent effects.
Aurora et al. (2010) and Morgenthaler et al. (2018), Journal of Clinical Sleep Medicine: imagery rehearsal therapy, recommended first line for nightmare disorder; normally therapist-delivered, which is why this app borrows only its engine.
Linehan, Cognitive-Behavioral Treatment of Borderline Personality Disorder (1993): dialectical behaviour therapy; radical acceptance, and the deliberate clinical use of irreverence.
Samson & Gross, Cognition and Emotion (2012): humour as an effective reappraisal strategy.
Heschel, Man Is Not Alone (1951): radical amazement, wonder at the fact of existence; borrowed as a lens name, not a doctrine.
Keltner & Haidt, Cognition and Emotion (2003): awe as an emotion of vastness and accommodation. Piff et al., Journal of Personality and Social Psychology (2015): awe and the small self.
Rogers, Journal of Consulting Psychology (1957): congruence, unconditional positive regard, empathy; claimed there as sufficient, kept by the field as necessary soil.
Kabat-Zinn, from 1982 onward: present-moment attention, trained and measured.
Nader, Schafe & LeDoux, Nature (2000): reconsolidation, the young science of the softened memory.
BS 6008 / ISO 3103, and the 1999 Ig Nobel Prize for Literature: the standard way to make a cup of tea, kept behind a link, exactly where it belongs.