Several instruments here run on a schedule. A frozen suite against the live models, a mutation tester against published eval suites, a verifier that re-replays every claim in a public registry. These are the times one of them found something larger than the smallest change it could print by accident. Most weeks they find nothing and nothing is published, because an instrument that reports a result every time is measuring its own noise.
Every note is archived here as it is published. To get them as they go out, subscribe to Drift Notes on LinkedIn. Weekly when there is something to say, and silent when there is not.