Technology

An experiment-as-code runtime — declarative experiment configs compiled against the event schema, run on a sequential-statistics engine with adaptive traffic allocation and a dependency resolver between concurrent experiments.

Hard problem

Valid inference at mid-size traffic — classical A/B testing needs samples these apps lack, and naive continuous monitoring breaks the statistics.

Differentiator

Always-valid sequential inference, seeded with priors from system 6, cuts the sample a conclusion needs — auto-stop is safe by construction.