A retry can repeat a command, a network split can separate authorities, and a replay can revisit an old decision. This practice set combines those failures using fresh workloads. The right answer states which operation is serialized and which work may remain pending.
Use a stable identity for work that may repeat. Keep a durable intent when a database change must produce an event. A compensating action needs its own completion record, especially when it calls another service. For streams, choose the ordering scope before choosing a partition count. These tasks don't introduce a new consistency model. They ask you to apply the models and recovery boundaries you have already practiced to a different set of constraints.