Independent research

A world modelforstreaming reality

We make machine predictions that reality grades — frozen before the fact, resolved by independent sources, mechanically scored. The formalism is borrowed from quantum information: density-matrix belief states.

Early-stage · built in the open

The field behind this page isn't stars — it's a distribution over futures, the object this lab studies. Move your cursor to perturb it; click to disturb it.Drag a finger to perturb it; tap to disturb it.

01 Research

The future is not a single guess.

Most predictive systems collapse what comes next into one answer. Reality doesn't work that way — it branches, stays uncertain, and resolves over time.

We study how to represent that honestly: a distribution over futures that stays calibrated as evidence arrives.

02 Approach

Borrowing the mathematics of uncertain, open systems.

Physics already has a language for systems that are genuinely uncertain and coupled to their environment. We build on it.

ρ

Belief states

Representing mixed, uncertain states — not single point estimates.

∂ρ/∂t

Open-system dynamics

Belief evolving under an environment, not in isolation.

Ĥ

Symmetry

Dynamics grounded in the conservation principles physics relies on.

CRPS

Calibration

Measuring whether our distributions over futures are honest — not just plausible.

A note on novelty — the quantum-information formalism here is a representational choice, not a performance or quantum-hardware claim. Density-matrix belief states are shared prior art (HQMMs, quantum cognition, Riechers, UWM-JEPA); our contribution is the integrated, outcome-scored system, not the formalism.

03 What we're building

A settlement layer for machine predictions.

Anyone can generate plausible futures; almost no one agrees to be scored on them. Our near-term work is concrete — making predictions accountable.

Each prediction is frozen before the fact, resolved later by independent sources, and mechanically scored — producing a tamper-evident receipt anyone can verify, without trusting us.

See the console — early preview, demo mode; no live model yet →

04 In the open

Built to be scored — in public.

Honesty is the whole point, so we hold ourselves — and our own engine — to a public ledger. Every prediction will be checked against what actually happened and scored head-to-head against simple baselines: persistence, constant-velocity. We'll report where we win and where we lose.

Open protocol & scoring code — Apache-2.0 github.com/Adonyth/omytea-protocol →

Receipts you verify without trusting us — verify the specimen receipt →

A free, public calibration leaderboard

Our own engine scored — wins and losses

Negative results published, not buried

Scores that can't be bought, edited, or deleted

✓ live today  ·  ○ committed — linked here as each goes public.

The longer arc

05  The longer arc
Can a single formalism unify how we model streaming reality?

That near-term work is deliberately narrow. The longer aim is more ambitious: a Grand Unified World Model. We treat it as a question to be earned, not claimed.

What this is not — not a foundation model, not video analytics, not an oracle. Omytea does not make individualized medical, legal, or financial predictions.

06 About

A small lab, working in the open.

An independent research program founded in 2026 — publishing openly, building open tools, and collaborating with researchers in world models, quantum information, and computational physics.

Founders

Jiaxuan Chen — M.S. Physics, Brown University. Prior research in high-energy physics (CMS), astrophysics, and computational plasma physics.

Haochen Wang — M.S. Physics, Brown University; M.S. in the sciences, UCLA.

The name

"Omytea" is a homophone of almighty. The ambition is in the name; the ledger keeps it honest.

Entity

Omytea LLC — a Wyoming-registered entity.

07 Contact

We welcome collaborators.

If your organization has real video, sensor, or agent state and wants calibrated predictions scored against reality, we're looking for a first design partner.

What a first pilot looks like

You bring a real stream — video, sensor, or agent state. We run calibrated predictions on it, frozen before the fact. You get scored receipts — wins and losses — that you can verify independently. Your data stays yours; we never ingest face or biometric data.

And if you work on world models, quantum information, or the calibration of prediction — we'd like to hear from you.

contact@omyteaai.com

Write as a design partner →Write as a researcher →