No node reboots is a dramatic difference for measuring disruption. Today these are lumped into the results for micro updates that do happen to have a CoreOS update. This means day to day, the results can skew dramatically.

We need to distinguish if a CoreOS update and node reboot has taken place, and if so track this separately with a new field.

It's similar to an empty FromRelease (aka e2e jobs without an upgrade), but probably different enough to track with a new flag because we do still roll all the pods images in our manifest)

How can we determine if an OS update took place most reliably in the ci-tools code which ingests disruption results?

Should we filter out "failed" upgrades?