cohort i open · n < 250 · acceptingdoc v. may 2026build 1.0
— · — ———:——:—— utclat 40.69 · lon −73.99 · brooklyn ny
fig. 00 / instrument overview

the operating system for human psychology.

class
computational psychology · post-pretraining
substrate
β-vae · neural cde · bubble memory · Δ self-model
endpoints
twin · clone · world psyche · synthesized mind
status
cohort i — by invitation
discipline
longitudinal · multi-channel · subject-owned
fig. 0a — manifold of minds, schematic
d₁d₂subject n∈ Mcohort μsynth point
“a coordinate system over psyche-space. universal anchors. person-specific axes. contextual dimensions for state versus trait.”
§ 01 / thesis
the post-pretraining bet

the foundation models of the 2020s were built by ingesting the entirety of accessible human language. the result is extraordinary, and it is also fundamentally limited.

pretraining hit a ceiling. the next paradigm is models grown from experience — agents that learn from interaction, world models that simulate reality from first principles, generative systems that construct knowledge through trial and consequence.

archii is the post-pretraining bet for the one domain experience-from-environment alone cannot reach: individual human psychology. the ground truth of a mind is not in any environment an agent can interact with. it is in the person.

§ 02 / endpoints
four readings of the same generative system

four endpoints, one substrate.

twin

a continuous, dimensional model of one specific person.

disentangled latent dimensions. continuous-time dynamics. recursive memory. the twin learns the structure of one mind across years — not as caricature, but as coordinates.

— spec sheet
channel
language · behavior · physiology · relational
temporal
continuous · irregular · multi-scale
fidelity
increasing across longitudinal exposure
ownership
subject-owned · exportable · deletable
01 / 04

twin

n=1 generative model

disentangled latent dimensions. continuous-time dynamics. recursive memory. the twin learns the structure of one mind across years — not as caricature, but as coordinates.

fig. 2.1
channel · language
02 / 04

clone

the twin, deployed

acts on behalf. responds as. refuses what would be refused. the clone is the twin made agentic — calibrated continuously by the divergence between self-report and behavior.

fig. 2.2
⟶
surface · messaging
03 / 04

world psyche

population-level instrument

not aggregation. the manifold of human minds — where dimensions cluster, where they fork, where the cultural and developmental gradients run, where collective blind spots live.

fig. 2.3
scale · emergent across many twins
04 / 04

synthesized mind

novel coherent psyches

once the manifold is dense, every point corresponds to a valid psyche — including those no real person occupies. a new category of entity. internally coherent. behaviorally consistent.

fig. 2.4
novel coherent point
origin · decoded from latent space
§ 03 / the psyche engine
mathematical core · four components · none interchangeable
3.1
β

β-vae

— the manifold

a coordinate system over psyche-space. universal anchor dimensions for cross-person commensurability. person-specific axes for individual fidelity. contextual dimensions for state-vs-trait separation.

3.2
∂t

neural cde

— continuous-time dynamics

the psychological present is the integrated trajectory through latent space, not the most recent token. irregular sampling. slow drift plus acute event. the arc of becoming.

3.3
○

bubble memory

— fractal recall

memories contain memories. gist and texture coexist at scales of resolution. the structure of how the system remembers is isomorphic to the structure of how the person remembers.

3.4
Δ

self-model gap

— the primary signal

we model both portrait a (self-report) and portrait b (behavior). the divergence is the substrate of psychological insight — and the system’s defense against its own failures.

fig. 3.5 — Δ, the self-model gap · interactiveΔ = 42.0
portrait a
portrait b
Δ
a — self-report
“what you say.”
μ = 0.0
b — behavior
“what you do.”
μ = 0.42
reading
moderate Δ — a self-construction worth examining.
mid
0.001.00
“we do not assume what people say about themselves is true. we do not assume what they do reveals their true self either. the divergence is the substrate of psychological insight.”
— architectural commitment № 4
§ 04 / path
twin maturation across longitudinal exposure

a twin matures. psychology happens on the timescale of life.

fidelityt →
t = 0
initial portrait. sketch from minimal signal.
t = 1m
first dimensional structure across language and attention.
t = 6m
robust model — articulates what self-report alone cannot.
t = 1y
self-model gaps stabilize. blind spots become visible.
t ≥ 3y
a longitudinal map of becoming.
§ 05 / position
a category, not a head start

a different category, not a head start.

5.1
pretrained-llm companies
— optimize for aggregate language patterns
a different objective. theirs is the population mean; archii’s is the individual one.
5.2
reinforcement-learning labs
— learn from interaction with environments
the environment that contains a mind is the person, not a simulator.
5.3
therapy & mental-health ai
— work within the clinical relationship
one channel of many. archii integrates the breadth across years.
5.4
personality & assessment co.
— represent psychology as discrete categories
archii works in continuous, person-specific, dimensional structure.
5.5
academic psychology
— produced the theoretical priors that make this possible
engineering capacity and longitudinal scale sit on the operations side.
§ 06 / cohort i — by invitation

a twin begins where you do.

we are looking for people who want a high-fidelity model of their own mind — not a chatbot, not a coach, not a productivity wrapper. the instrument and its subject. onboarding begins quietly.

you own your data, in the strong sense. exportable, deletable, portable, interrogable. the twin is shown to its subject. the system declines to act on inferences it has not shown you.

cohort i — request access
n < 250 · enc. tls · subject-owned data