The physics data for an event is a collection of EDO's (event data objects) and the history for each. The EDO is an ATLAS data object and so its interface is defined by the ATLAS persistency system. It is likely that the history data for an EDO will be stored separate from the EDO itself.
I expect a common usage pattern will be for a user to request the EDO whose history matches or comes closest to matching some specification. For this it will be necessary to gather all the history records for an event so they can be searched and compared. One simple and intuitive interface would have the event return a list of history objects.
There will be many different criteria for selecting the best EDO and so this selection cannot be part of the event. It might be done by the user, by some intermediate agent or by passing a selector to the event.
Once the desired history is selected, the user needs to access the associated data object. It would be natural for the history to provide this with a lazy pointer or key.