Each new result (independent of how many rules match) is counted as 1 credit. Updated documents come at no cost. The field
matched provides this information.
More on Result chunk
If no credits are left, the data access is stopped and a control chunk containing the
resume_offset of the end (needed for resuming) is sent.
API calls which don't return any results are not counted.
The documents are billed after every completed chunk.
If a stream gets disconnected, a non-completed chunk will not be billed, since it needs to be restarted when resuming.
If an export task gets interrupted, all exported results are counted. Repeating the task means exporting duplicates, which count both times as 1 credit.