Research Center · Engine
When the engine last looked.
Every model in the engine reports its most recent run here: the window of data it considered, whether the run succeeded, how many candidate findings it emitted, and any message the model left behind. A model that never ran shows as never run; a model that ran and refused to publish (typically due to k<10) shows as skipped.
Checking the engine…
How to read this
- Last run. Timestamp of the most recent invocation, in UTC. Cron schedules are weekly; manual reruns happen during pilot tuning.
- Status. success = ran cleanly, with or without findings. error = the run threw; the message column explains why. skipped = the model intentionally produced nothing (typically because every candidate slice was below k=10).
- Input window. The date range of warehouse data the model considered. Models default to a rolling 6-month window, which lets the same model surface different findings as the cohort grows.
- Findings emitted. Draft findings the model wrote into the warehouse. Drafts are not the same as published findings — an admin reviews each draft before promoting it to /research/findings.