The scheduler runs it every minute, which is cron's finest granularity and
still up to sixty seconds of silence after a build goes red. That is the
single largest delay between a failure happening and the system visibly
reacting to it, and no amount of polling on the client side shortens it.
The same tick can now be run immediately. It is idempotent - incidents dedupe
on job and build number - so an extra run can only be a no-op, never a second
incident for the same build. Both the authenticated admin route and the
internal one follow the shape the testing-triage routes already established.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
