I am not yet ready to propose such a taxonomy but perhaps I will collect here a list of events that need to be distinguished.

Falling back to a previous checkpoint is not a sufficiently general characterization of how a system may cause trouble. When a system makes a checkpoint, there may be objects within in the midst of interacting with an external object using some protocol that prescribes a sequence of messages. If the state of an internal agents engaging in such a protocol is replicated and subsequently runs, the protocol can fail for reasons that protocol design cannot be held responsible for. It is generally possible to equip communication channels to send a distinctive signal to the reawakened object state that signals the time quake.

A closely related issue is the problem of communication failures and when to decide that a transaction should be aborted or abandoned.