My understanding is that for Pipeline Builder batch pipelines, unless a sampling strategy is configured, the full input dataset is always in scope for the calculation of output previews.
Do output previews for streaming pipelines work the same way? I’m asking this question because we have a streaming pipeline that just filters data to rows for which a string column starts with a particular substring, and the Pipeline Builder output preview shows zero rows, but actually running the stream produces a nonzero amount of rows. This seems inconsistent with the documented behavior for batch pipelines, and we’re trying to determine if this is within the scope of expected behavior for streaming pipelines.