See PHI1280. An install is seeing slow (~20s transcript total time) evaluation from Herald but we currently don't provide the tools to dig into this and understand it.

These tools have to be a bit involved: we can't just time each rule, because rules share conditions, and building the field value for a condition is often the most expensive part. Conditions are also evaluated lazily, and rule and condition evaluation may trigger evaluation of other rules or conditions via [ Another Herald rule ].