We run multiversion for "last LTS" (major) and "last continuous" (minor), but not against "last patch".

There is a circumstance where the next x.y.z branch could be incompatible with x.y.z-1 (latest), though we only test the changes in x.y.z across x.y-1 (last continuous) and x-1.0 (last LTS) releases.

DEVPROD-33848 recently introduced such a "last patch" concept to db-contrib-tool, so this might now just a matter of task-generation logistics instead of actual implementation.