Research
UpCube is building toward an AI-native research and technology institution. Today, two research programs are active. Everything below is stated at its real evidence state.
Active programs
P1 — Agent reliability and verification
We're studying when machine work can be trusted and what one verified outcome costs. Verification of machine work is the first discipline we're building around.
P2 — Formal mathematics (Lean 4 / Mathlib)
We're studying whether AI agents, a proof checker and independent human review can formalize known mathematics at a measurable cost per verified proof. The pipeline is being built; no verified proofs are counted yet.
Evidence so far
One measured result has been published (AgentTrustBench R08): on one pinned build, 194 of 197 enumerated boundary conditions matched both the pre-registered prediction and an independent oracle, across two passes in different orders. It is a scoped measurement — one pinned build, one enumerated condition set, no population claim, no statistical test — and it has not been externally replicated.
Research Lab notes
UpCube's Ethen Research Lab publishes institution-authored research notes on agent reliability, verification, evaluation, memory and data rights. Most are proposals or protocols that design a test; they are not results. Read the Research Lab notes.
How we do research
- We pre-register experiments before collecting data, and we record negative results.
- AI systems are never listed as authors; AI involvement is disclosed.
- Nobody reviews their own work at the higher-risk tiers.
Research areas: what UpCube studies and watches · The institution · What I'm building
Shadab "Sha" Chow — founder of UpCube Technologies, journalist & researcher. Building Ethen & an AI-native research and technology institution. About Shadab →