For "error-based" sampling, we buffer the replay events until an error occurs. After an error occurs, we upload the replay and continue to record the replay and treat the rest of the replay as a "session-based" replay. After this session expires, we stop replay completely. In order to capture replays again on that browser tab, two things must occur: 1) session in session storage has to be expired and 2) user has to reload the tab.
Session-based replays will get resampled after they expire, so I think it makes sense to restart buffering mode after an error-based replay is completed.