Time reversible events
The current state of a system might be represented by the contents of a database table. The table could have many columns of various data types, but to simplify we'll say there is only a single integer column, so our table is just a list of integers. (Each integer could in reality be a foreign key into another table holding immutable and distinct tuples, each describing a frozen configuration of a more interesting entity such as a person, a medical record and so on. So we can make this simplification without loss of generality.)
We can describe the history of the table's state by creating another table in which the rows represent events. Of course in a database we need to think about atomic transactions; some kinds of change may not make sense by themselves and must always occur atomically as part of a transaction along with certain related changes. Therefore an event always belongs to a batch, and a batch may contain multiple events. Batches occur in a definite order, so we can number them (that is, the primary key of a batch is a sequence number). A batch is also the ideal place to record the clock time that the events in the batch were applied.
Aside from the batch_id an event has just two columns: