1. A cluster is on release A.
2. Admin requests update to B.
3. Update to B is partially complete.
4. Admin requests update to C.
5. Update to C completes.
6. User decides C has a bug that causes an unmanageable performance degredation, and triggers the rollback command.

A, B, and C are all in the same z stream.  Do we take them back to B (the previous history entry) or A (the most recent completed history entry)?  We know their cluster previously reconciled A, so assuming that there haven't been breaking cluster-config or environment changes, we should be able to complete A again.  But B is presumably more recent than A, so it fixes some bugs that A had (and may also introduce some regressions that A lacked).  Thoughts?