Strategic bets
Where we concentrate effort
These bets shape our research priorities, deployment model, and how we measure success.
Evaluation notes as trust fabric
Every public claim is anchored to an evaluation note with protocol, sample size, limitations, and source artifacts.
Sovereign deployments as default
Local-first cognition with Mavaia keeps data, telemetry, and governance within institutional boundaries.
Safety baked into architecture
Hallucination detection, verification, and confidence calibration operate inside the pipeline — not as filters after the fact.
Mechanism-first collaboration
Partnerships are designed around reproducible mechanisms and shared evaluation harnesses, not vague feature roadmaps.