entry 00 · the claim
log in to your own mind.
archii is an independent ai psychology lab. an interface for the interior, built like a scientific instrument.
we are building a digital twin of one specific person — you, if you choose — from records and measurement. then a shell: a way to enter your own modeled mind and see what it knows — the patterns you can’t see from inside. beneath them, a simulated population of three hundred million to test every claim against. and, if the evidence ever permits it, a person-preserving upload. this page states the program plainly, including the parts that do not exist yet.
the program
the whole program is one pursuit, at increasing depth: a computational, measurable model of a specific mind. a twin estimates the state you’re in. the harder claim is how you change — a model that predicts you today but not the you tomorrow makes is not yet deep enough.
state, memories, values, goals, relationships, self-model — and the rule by which they move. that is what gets measured, modeled, and, eventually, run.
the twin
in progressa model of one specific person, sharp enough to predict them — and precise enough to be proven wrong.
the twin begins with records, not summaries. your archive, standardized assessments, instrumented cognitive tasks, the traces of a documented life — resolved into a model that predicts you before the fact, against controls: wrong person, wrong time, population baseline. the twin keeps score on itself, because a twin that can’t tell you how wrong it is shouldn’t be trusted about anything else.
today: the capture app is in open beta; the first twin loop runs end to end. no earned claim beyond that.
the shell
specified“ssh into your brain” is the ambition. what you actually enter is the model — under an access contract exact about that difference.
not a folder tree. a way to enter a mind-model, inspect its evidence, run it forward, and know how much to trust it. the shell always exposes what is measured, inferred, simulated, or unknown. ask it for something it hasn’t earned and it says so, explicitly — it never covers the gap with a plausible guess.
the contract is written; none of it is built. levels are earned in order, ssh-0 first — and today the count stands at zero.
the world
liveone mind at full depth. a population at full breadth.
the same lens that resolves one person, composed into many: a working simulation of 303,128,287 synthetic people, modeling a real population. studies run against it today, and every twin the lab validates makes the world truer.
live at world.archii.ai. and scale earns no claims here: a population is not a person.
the continuation
researchthe transfer boundary: the point at which a model of a person would have to become the person.
the destination is u7: person-preserving upload — a mind carried across substrates under an explicit theory and test of survival. no science yet says how to check, from the outside, that a person survived the crossing. that gap is an open problem inside the program, not a reason to quietly rename something lesser.
nothing in this phase is built. we are not close. no one is.
the ladder
upload is not a button, and it is not a percentage. it is a ladder of resolution claims, each with its own test — and each with things it deliberately does not claim.
current position: no earned u-level. the route runs from foundations to continuity; we are at the foundations. the mistake would be calling an impressive u1 system u7. ambition is literal; claims are earned.
the discipline
a lab aiming this high needs firewalls between what it builds and what it says. these are part of the protocol, not retrospective copywriting:
- functional is not neural.
- neural is not continuity.
- population is not person.
- interface access is not ground truth.
- believability is never evidence.
and one rule above the rest: the upload is only valuable if it knows when it is wrong.
everything is consent-gated, operation by operation — collection, training, simulation, execution — and consent to one does not imply any other. the source person is a co-investigator, not an extraction target. raw audio, camera, and health data stay on the device. under unresolved rights, the default is not to proceed.
what runs today
archii world live
the population, open for outside studies. run one against it, or bring your own data and see where you fall in it.
archii lab live
an autonomous research scientist for psychology. it reads papers, extracts claims, pre-registers hypotheses, survives its own skeptic, and keeps what holds — 726 findings so far, each one inspectable on request.
archii os beta
subject intake. the capture app: your archive, 44 standardized assessments, thirty-odd instrumented cognitive tasks. the twin is built from what you consent to share — and only that.
the twin in progress
a working end-to-end loop on the lab’s machines: its own homeserver, the model on-device. it speaks plainly, briefly, and never flatters — that sentence is its actual system prompt.
the lab
the lab is agent-operated and human-governed. the whole machine: a mac running agents, a phone in the subject’s hand, and a vault of plain markdown between them. agents study, propose, implement, and review — no agent reviews its own work, and no agent can authorize its own experiments. a human holds the charter.
we’re quiet by design. mostly we research. now and then, what we learn turns into something we release.
the invitation
the record is open.
the instruments are live, the archive is taking subjects, and the rest will be earned in public — entry by entry.