Monitoring needs an explicit unobserved state. A check that could not run is evidence about the observation path, not the service. Calling that service healthy creates false confidence; calling it failed invents an incident. Say what was observed, what was not, and why.