Orientation
What this is
Polycentria builds verification infrastructure: ways to establish what a physical or AI system actually did, from its own motion record, beneath any narrative about it. The public surface of that work is deliberately small right now — we are pre-filing, so these pages describe problems and validated results, never mechanics.
Why 'beneath language'?
Logs, reports, and model outputs are all narratives — they can be wrong, edited, or simply missing. An orientation stream is closer to the physical event itself. Our work asks: how much can be verified from that record alone, and how do you make such a verification portable, signable, and auditable?
What exists today
What exists today, concretely
A working software pipeline, exercised by a test suite you can count on the dashboard; adapters for common robotics and wearable formats IN CODE; one real public dataset taken end to end MEASURED; and a set of genuinely open research questions we publish rather than hide OPEN PROBLEM.
Boundaries
What we will not tell you here
How it works. Until our patent converts, mechanics are a disclosure event. Specific measured figures travel under NDA. If a page seems to stop short of a number you expected — that is the reason, and we would rather say so than imply the number does not exist.