Research Center · Findings

What the engine has surfaced.

Each entry below is a single, citable observation from a single model run. We list the data window, the sample size, and the model that produced it. If the slice was smaller than ten members, you won't see it — not because it isn't interesting, but because we promised.

How a finding gets here

1. Members opt in to the research pool. Default is off.
2. An extract job builds a de-identified, bucketed warehouse from operational data.
3. A small set of well-defined models runs against the warehouse.
4. Each model emits draft findings; an admin reviews them before they appear here.
5. Anything below the k=10 threshold is silently refused at every layer.

The full contract lives at /research/methodology.

Next Steps