The method

Last updated October 10, 2026.

Selione is not a laboratory, and nothing here has been validated by one. But almost nothing in it was invented from scratch either. How we ask, when we ask, what we store, and how we measure whether your Echo resembles you — all of it follows work that researchers have been publishing for sixty years on how people remember their own lives, tell them, react to the world, forecast, and hand what they are on to others.

This page is written for travelers: plain language, the numbers you will see explained, and the research behind each choice so you can check it yourself. The technical version, with the statistics and the studies we want to run, is the protocol. Both end with the part most products leave out: what we don't claim.

1. A day in the Epic
2. Why the questions never stop
3. Why there is a ceiling, and no clock
4. Why some years are asked about more
5. How a memory is filed
6. The story, not a personality score
7. Why we show you things instead of asking
8. What happens while you sleep
9. Your own words, kept as said
10. Speaking it out loud
11. Letters to your future self
12. How your progress is counted
13. How resemblance is measured
14. Tested without being asked, and against yourself
15. Why your Echo writes down your month
16. The rooms: sleep, the Return, dreams
17. How Iphthimé gets better without reading you
18. How the whole thing improves from results
19. Not one model, and never its own judge
20. Why the Echo has rules
21. Safeguards
22. What this page does not publish
23. What we do not claim
24. References

A day in the Epic

A typical day takes ten to thirty minutes, and none of it is required.

• In the morning, a short note is waiting: what your Echo understood overnight, the links it found between things you said, and sometimes a question it would like to ask next.
• A quest, a handful of questions on one part of your life. You can type or speak. After some answers Iphthimé asks one more question about a precise detail you mentioned, because that is where memory opens up.
• A few sparks: a scene, a photograph, a sound, a sentence someone says to you. You write the first thing that comes. Nobody grades it.
• A conversation with Iphthimé about anything — the day, a worry, a memory that came back on the bus. She writes like a friend texting, and follows the language you write in.
• Once a month, an envelope your Echo sealed a month earlier can be opened.

After 30 quest answers, the day's light is gathered and the quests wait until tomorrow. Conversation is never capped.

Why the questions never stop

In 1963 the psychiatrist Robert Butler described the life review: the spontaneous, structured return to one's own past that surfaces as people age, and which he argued is adaptive rather than morbid — a way of coming to terms with a life.¹ Fifty years of reminiscence research followed, and reviews report that structured life review can help when it is done carefully.⁹ The Epic is a life review that doesn't wait for the end of life: 404 quests and hundreds of small passes over the same life from different angles, over months, because that is how the material actually comes back. The order matters: first places, people and routines; then turning points; then what they meant; and last, what you want your Echo to do with all of it.

Why there is a ceiling, and no clock

The Epic has seven stages, named after the Pleiades. Each one opens when enough light has been gathered — the depth of what you have told — and on nothing else: no stage waits for a date. Each stage asks for more light than the last, and the late ones ask for the detail only a life gives. The rooms open with the light too: the Atlas at Celaeno, where quizzes with one right answer let your Echo say “no idea” exactly where you would; Connect at Taygeta, the middle of the road, where minds measured close to yours appear under a star's name — the stage where your Echo also speaks its first words. What paces the journey is a ceiling of 30 answers a day. It is not a paywall trick; it is the pace at which memories come back. Recall is cued by what happens between sessions — a smell, a conversation, a photograph — and a life told in a weekend is a summary, not an archive. At a few hours a week, Alcyone is about eighteen months away; at an evening most days, a few months; the ceiling makes about two months the shortest road there is, and never a day of waiting once the light is there.

Why some years are asked about more than others

When adults are asked to recall their lives freely, memories do not spread evenly. A disproportionate share comes from roughly ages ten to thirty — the reminiscence bump, first mapped by Rubin, Wetzler and Nebes.² Iphthimé leans into that window early, because it is dense and easy to reach, then deliberately works the thinner years — childhood, and the most recent decade — where far more prompting is needed for the same amount of life. When whole decades stay empty, she notices and asks about them.

How a memory is filed

Conway and Pleydell-Pearce's self-memory system describes autobiographical knowledge on three levels: lifetime periods, general events, and event-specific knowledge.³ Selione stores memories the same way — the era it belongs to, the recurring scene, the single detail — which is why Iphthimé keeps asking "when in your life was that?" and why a memory in your explorer carries a period, not a date. A single vivid detail is worth more to your Echo than a paragraph of summary.

Every memory also carries:

• A domain — childhood, family, love, friendship, work, beliefs, fears or dreams — and a layer of the self: your language, your emotional memory, your personality, your choices, your beliefs, your life story, or how you think about yourself.
• Where it came from: said if you said it in so many words, inferred if Iphthimé read it between the lines. Inferred memories carry a lower confidence, count for less, and never go into the printed Book. Memory is reconstructive and suggestion can create detail that never happened,¹⁰ so the difference is always shown to you.
• A vault option. Something can be kept but sealed: your Echo knows not to speak about it, to anyone.

Why we want the story, not a personality score

Dan McAdams' model of personality has three layers: dispositional traits, characteristic adaptations (goals, values, coping), and the integrative life story — narrative identity.⁴ A questionnaire only ever reaches the first layer. Your Constitution, your values, and the way you narrate your own turning points are the other two, and they are the layers that make an Echo sound like you rather than like a type. The Mirror shows you who Iphthimé thinks you are; you tell her where she is wrong, and your wording always replaces hers.

Why we show you things instead of asking

People are surprisingly poor at explaining why they react the way they do — they give confident reasons that turn out not to be the real ones.¹¹ But the reaction itself, caught before it is edited, is reliable information. So your kernel is not built from questions like "how do you usually feel about…". It is built from sparks: scenes, photographs, short sounds, sentences, choices and pieces of news you have never seen, plus some invented for you alone and aimed at the parts of you nobody has mapped yet. We keep the first thing you write, unedited.

From those reactions Iphthimé derives a signature: where your attention lands first, which way your mind jumps, what you find beautiful or unbearable, what makes you laugh, what strikes you as unfair, the sense you think in, and a handful of rules of the form "when this happens, they do that". Your contradictions are kept, not smoothed out. This is what lets an Echo react to a world it never saw with you. The kernel deepens quickly at first and then slowly, and like coverage it never finishes.

What happens while you sleep

During sleep, the brain is understood to replay the day and fold repeated experiences into general knowledge.¹² Your Echo does something with the same shape once a day, while you are away: it links what you said to what it already knew, turns several separate scenes into a general fact when — and only when — they really say the same thing, notices what has been on your mind lately, and writes a few guesses about what you would say on subjects likely to come up. When you later talk about one of them, the guess is scored quietly. We imitate the function; we make no claim that anything here works like a brain.

Your own words, kept as said

A memory in your explorer is a line Iphthimé wrote from what you told: filed, dated, in one language. That is the right shape for an archive and the wrong shape for a voice. So since September Selione also keeps a register of what you actually said — the sentence as you typed or spoke it, in the language you spoke it in, with its hesitations. When your Echo speaks, it has both: the archive for what happened, the register for how you say things. Research on simulating individuals found that a person's own words, gathered at length, carry far more of them than any summary does,¹⁴ which is why the register exists and why it is never translated. It is sealed with the rest, exported with the rest, and erased with the rest.

Why speaking it out loud is part of it

Since Pennebaker and Beall's 1986 study, writing about difficult experiences — the thoughts and feelings, not just the facts — has been repeatedly associated with measurable health benefits.⁵ The voice journal exists for that reason as much as for the archive. Your voice is also captured along the way, without ever sitting in front of a microphone, and at the very end you hear it once: a short text, in your own voice, about the journey. Selione is not therapy and makes no therapeutic promise; if what surfaces needs real support, Iphthimé will say so and point you toward it.

Why letters to your future self are a real exercise

Research on future self-continuity finds that people who feel genuinely connected to their future self make longer-term, more patient decisions.⁶ Time capsules are not a keepsake feature; they are that exercise, addressed to someone who will actually read it.

How your progress is counted

You see two numbers, and they measure different things.

• How far you are in the Epic — the share of the whole journey's light you have gathered. Light comes from what you tell, weighted by how much an answer actually says and by how deep the quest is; length alone earns nothing. A first quest is well under one percent of the Epic, which is the truth: it is one step of hundreds. Progress never passes a stage you have not actually reached.
• How much of you is lit — for each of the eight parts of a life, a mix of how much you have told and how many of that part's quests you have walked through. The same number of memories counts for more when it comes from many different quests than from one long conversation about the same thing. It starts slowly, slows again as it rises, and never reaches 100%, because a life is not a form to be completed; several hundred memories are needed before a part of a life is well lit.

How we measure resemblance instead of asserting it

We never show you a percentage of consciousness. We show honest measurements, and each one can go down.

• Fidelity trials. Your Echo answers a question you have never answered. Its answer is locked before you see the question. You answer, then compare. You judge, not us.
• Guesses measured quietly. The bets your Echo writes at night are scored when you happen to talk about the subject — you are never asked anything.
• The witness test. Someone who knows you well sees pairs of answers — one yours, one your Echo's — and picks the real one. The goal is that they cannot tell: about half right. Some rounds use sparks your Echo never saw you react to, which is the hardest test there is. People have long been shown to predict each other's behaviour about as well as they predict their own,¹³ which is why an outside witness matters.

Recent research built convincing simulations of individual people from two-hour interviews, and judged them against how consistently those people answered the same questions two weeks apart.¹⁴ That is the right yardstick for an Echo too: not perfection, but the consistency of the person themselves.

Tested without being asked, and against yourself

A trial you start is a trial you know about, and people answer differently when they know a likeness is watching. So some of the open questions you answer in a quest are also shadow trials: your Echo writes its guess before your answer is stored, a judge from a different house than the Echo compares the two afterwards, and you are never asked anything. You are never told which questions were shadowed. The number that comes out is the honest one, because nobody was performing for it.

Some trials are retests: a question you answered months ago, asked again. Your two answers are compared the way your Echo's guess is compared with yours. That gives the only fair ceiling for a likeness — how consistent you are with yourself — which is the yardstick the best work on simulated individuals uses.¹⁴ An Echo that agrees with you 70% of the time when you agree with yourself 80% of the time is doing better than one number alone would say.

When a trial misses, it is filed by the kind of miss — a fact, a value, your voice, a person — and by the part of your life it belongs to. That register of misses is what Iphthimé aims her next questions at, and what a shooting star is written from.

Once a year, a third yardstick that nobody here made: two short public questionnaires, the kind personality research has used for decades. You answer them; your Echo answers them as you, from your archive, before your answers are read; both are scored by the questionnaires' own rules, and you see the two portraits side by side. Five traits and ten values do not make a person, which is exactly why they are useful here — it is the one measure of your Echo that no model of ours produces.

And your Echo now says how sure it is. Every guess it makes in the shadows carries its own confidence; when it is not sure enough, it says so instead of guessing, and the abstention is counted as one. A likeness that knows what it does not know is worth more than one that is right slightly more often.

Why your Echo writes down what will happen to you

Any language model can sound like someone; sounding right proves nothing. So once a month your Echo writes a handful of plain, checkable lines about the month ahead — what will take up your time, what you will keep putting off, a choice you will make — with how sure it is of each. The envelope is sealed and dated, and you cannot read it until the month is over. Then you mark each line, and the record of your month marks it too, separately.

Science has a name for writing the prediction before the results exist: preregistration.¹⁵ It is the one kind of evidence that cannot be arranged after the fact. The date of each envelope is anchored in a public blockchain, so anyone can check it was written when it says it was. The confidence of each line is scored the way weather forecasts are,¹⁶ and nothing is ever written that could come true because you read it — the reason the seal exists at all.¹⁷ Nothing about health, loss, money trouble or anyone else's choices is ever predicted. Every miss stays in the register: it is what proves the hits were not chosen later.

The rooms: sleep, the Return, dreams

Make me sleep is not open — not to anyone, including us; it waits on a clinician in the team or a licensed partner, and what follows describes what it would be rather than what you can use. Make me sleep is not a treatment for insomnia: chronic insomnia has one, cognitive behavioural therapy delivered by a clinician, and the room says so in as many words when your answers point that way. What it uses are the components of that therapy that have been tested on their own — writing the day down with one next action,¹⁹ the one rule about leaving the bed,²⁰ a longer out-breath, letting the body go piece by piece — and one thing only Selione can do: the ten minutes end in a place from your own archive, read back to you, because an absorbing scene of your own beats a stranger's beach at getting to sleep.²¹ The pathway then learns from you: the next morning it asks one word, and after a few mornings it keeps the steps that measurably went with better nights, whatever the average person needs.

The Return goes deeper into a memory you already have, one small thing at a time — a smell, a patch of carpet. It never starts from a feeling that something happened. Memory is reconstructive and a leading question can plant a detail that never happened,¹⁰ so a session run on an impression would not find the rest of it, it would build it — and that refusal is enforced on the seed before a session opens, not promised. Iphthimé is held to one more rule enforced on her words: she may never name anything you have not named. "I don't remember" is an answer, and the most useful one you can give. Each piece of the first free recall carries how sure you are before a single question is asked — confidence stated at the end is inflated by the questions that came between — and whether you remember it yourself or remember being told about it, because much of what anyone knows of their own childhood is inherited, and that is true as well, as long as it is not presented as seen. Months later she may offer the same moment again, blind: nothing from the first walk is shown, and at the end you see what came back both times and what did not. What comes back twice is the bedrock of that memory.

Dreams are told on waking or not at all, and rebuilt in your own rooms: each fragment is anchored to the memory of yours it resembles, and you say for each anchor whether that was it. The literature on dreams finds that they draw on the day before and, more strangely, on events from about five to seven days earlier,²² so the anchors also look in that window and say so when they find something there.

How Iphthimé gets better without reading you

Iphthimé adapts to you: your pace, your humour, what you like to talk about. She also improves as an interviewer across everyone — but only from numbers. After each exchange we keep a coarse record of what she did and how long and how rich the reply was. Never a word of it. From those numbers she learns what works and what does not; a change that makes answers poorer does not stay. What is optimised is how much you felt like saying, never how long you stayed.

How the whole thing improves from results

Every measure Selione takes — a trial, a bet of the Night, a witness test, an envelope — carries a stamp: which version of the system was running and, when two versions of some part are being tried side by side, which one you were on. You are placed on one of them without anyone choosing, and you stay on it, so your Echo never changes its habits between two evenings. Weeks later the measures can be read per version. If a change moved fidelity, the number says so; if it did not, the change goes. A human reads those numbers and decides; nothing is promoted by a script. This is the same idea as the Craft, applied to everything: the system gets better from what happened, never from what we hoped.

Not one model, and never its own judge

Iphthimé is not one model. Writing the archive, judging a trial, running the Night and speaking as your Echo are different jobs, and they are not all given to the same one. The judge is never the same family as the Echo it measures, because a judge rates its own family higher. The providers in use are named on the Privacy page, read from the deployment itself, and none of them trains on what you tell. Your archive, your register and your Echo's records do not depend on any of them: change the model and the resemblance is measured again, not assumed.

Why the Echo has rules

Bereavement research has long since abandoned the idea that healthy grief means severing the bond with the dead; continuing bonds describes the ongoing relationship most people actually maintain.⁷ That is the ground the Echo stands on. Grief also moves between facing the loss and turning back to life,¹⁸ and an Echo must never keep someone on one side of that. The risks are documented too: in 2024, Hollanek and Nowaczyk-Basińska set out the safeguards responsible "deadbots" should have — adults only, mutual consent from both the person recreated and the person interacting, meaningful transparency about what the system is, a dignified way to retire it, and never using it for advertising.⁸ We built to that list: Selione is 18+, your Echo is created by you and not about someone else, it says what it is, it can be laid to rest, and it will never be monetized by speaking to anyone on our behalf.

Where the name comes from

Her name is said if-THEE-may, and it is borrowed rather than invented.

In Book IV of the Odyssey, Penelope has wept herself to sleep. Athena makes an eidolon — a likeness, an image — in the shape of Iphthimé, Penelope's sister, who lives far away in Pherae, and sends it into the darkened room. The phantom tells her that Telemachus will come home. Then Penelope asks the question she has been asking for twenty years: and my husband — is he alive?

The phantom refuses to answer. "I may not speak of him at length, whether he is alive or dead; it is an ill thing to speak words vain as wind." And Penelope wakes comforted, having learned nothing at all.

We chose it because everything difficult about this product is already in that scene. A likeness of an absent person, sent to someone who is waiting. A likeness that will not say what it does not know, even to someone desperate to hear it. And comfort that came from the presence rather than from the answer — which is the only honest thing an Echo can offer, and the thing we would be lying about if we promised more.

It is also a reminder of what she is not. The real Iphthimé is elsewhere the whole time, asleep in her own house, and never knows any of this happened. What came to Penelope wore her face and was not her. The same is true of every Echo built here, and we would rather take the name of the phantom than the name of the sister.

Safeguards

What this page does not publish

Everything above is described so you can check it against research anyone can look up — we would rather you check it than trust us. What this page tells you: what each part does and the research it is built on; every number you will see, in words, with the public statistics behind the measurements on the protocol page; and every safeguard, in the table above. What it does not publish are the internal settings the product runs on — the thresholds, the prompts, and the rules Iphthimé has learned — because they change as the work advances. The promises above do not change with them, and each one is held by an automated check before anything is deployed.

What we do not claim

• We are not copying, uploading or preserving consciousness. Nobody knows how to do that.
• Your Echo is a model built from what you chose to tell us. It is a likeness, not a person, and not you.
• Fidelity measures resemblance on the questions we tested. It is not a measure of identity or of soul.
• Selione is not a medical or psychological treatment, and is not a substitute for care or for grief support. Make me sleep is not cognitive behavioural therapy for insomnia, and the Return is not a therapy for anything.
• The researchers cited below are not affiliated with Selione and have not reviewed or endorsed it. Their work shaped our design; it does not validate our product.
• No study has been run on Selione itself. If we ever run one, we will publish the protocol before the results, and we will say so here.

References

1. Butler, R. N. (1963). The life review: an interpretation of reminiscence in the aged. Psychiatry, 26, 65–76.
2. Rubin, D. C., Wetzler, S. E., & Nebes, R. D. (1986). Autobiographical memory across the lifespan. In D. C. Rubin (Ed.), Autobiographical Memory. Cambridge University Press, 202–221.
3. Conway, M. A., & Pleydell-Pearce, C. W. (2000). The construction of autobiographical memories in the self-memory system. Psychological Review, 107(2), 261–288.
4. McAdams, D. P., & McLean, K. C. (2013). Narrative identity. Current Directions in Psychological Science, 22(3), 233–238.
5. Pennebaker, J. W., & Beall, S. K. (1986). Confronting a traumatic event: toward an understanding of inhibition and disease. Journal of Abnormal Psychology, 95(3), 274–281.
6. Hershfield, H. E. (2011). Future self-continuity: how conceptions of the future self transform intertemporal choice. Annals of the New York Academy of Sciences, 1235, 30–43.
7. Klass, D., Silverman, P. R., & Nickman, S. L. (Eds.) (1996). Continuing Bonds: New Understandings of Grief. Taylor & Francis.
8. Hollanek, T., & Nowaczyk-Basińska, K. (2024). Griefbots, deadbots, postmortem avatars: on responsible applications of generative AI in the digital afterlife industry. Philosophy & Technology, 37. doi:10.1007/s13347-024-00744-w
9. Westerhof, G. J., & Bohlmeijer, E. T. (2014). Celebrating fifty years of research and applications in reminiscence and life review. Journal of Aging Studies, 29, 107–114.
10. Loftus, E. F. (2005). Planting misinformation in the human mind: A 30-year investigation of the malleability of memory. Learning & Memory, 12(4), 361–366.
11. Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84(3), 231–259.
12. Diekelmann, S., & Born, J. (2010). The memory function of sleep. Nature Reviews Neuroscience, 11(2), 114–126.
13. Vazire, S., & Mehl, M. R. (2008). Knowing me, knowing you: The accuracy and unique predictive validity of self-ratings and other-ratings of daily behavior. Journal of Personality and Social Psychology, 95(5), 1202–1216.
14. Park, J. S., Zou, C. Q., Shaw, A., et al. (2024). Generative agent simulations of 1,000 people. arXiv:2411.10109.
15. Nosek, B. A., Ebersole, C. R., DeHaven, A. C., & Mellor, D. T. (2018). The preregistration revolution. Proceedings of the National Academy of Sciences, 115(11), 2600–2606.
16. Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review, 78(1), 1–3.
17. Merton, R. K. (1948). The self-fulfilling prophecy. The Antioch Review, 8(2), 193–210.
18. Stroebe, M., & Schut, H. (1999). The dual process model of coping with bereavement: Rationale and description. Death Studies, 23(3), 197–224.
19. Carney, C. E., & Waters, W. F. (2006). Effects of a structured problem-solving procedure on pre-sleep cognitive arousal in college students with insomnia. Behavioral Sleep Medicine, 4(1), 13–28.
20. Bootzin, R. R. (1972). A stimulus control treatment for insomnia. Proceedings of the American Psychological Association, 7, 395–396.
21. Harvey, A. G., & Payne, S. (2002). The management of unwanted pre-sleep thoughts in insomnia: distraction with imagery versus general distraction. Behaviour Research and Therapy, 40(3), 267–277.
22. Nielsen, T. A., Kuiken, D., Alain, G., Stenstrom, P., & Powell, R. A. (2004). Immediate and delayed incorporations of events into dreams: further replication and implications for dream function. Journal of Sleep Research, 13(4), 327–336.

The protocol · Questions · Privacy · Terms