It helps that we deploy all new changes first in our internal community – the one we use for work every day, and also meta.discourse.org.
The Discourse team gets the newest and shiniest changes … and bugs. Then you folks on meta get the brunt of it too.
Contrary to popular belief, we don’t deploy all changes all the time to all sites; we do deploy sites on a regular basis. You are right that breaking a site is a risk. We have to strike a good balance between keeping sites safe and updated vs never updating them due to worry of breakages.
For tests, you can take a look at our repositories. We also have additional internal build+test processes
It is an ideal scenario for a third-party plugin to have tests as well, and also good to note that we do not support them on our hosting due to not being able to manage incompatibility and security.