Guan, one thing you could do to reduce the need for purging is to copy the way Radio does timestamps. Namely, put an absolute timestamp in the source chunk header indicating when the new items in that chunk were first discovered by the scanner. The timestamp needs to be absolute, not relative, so that it stays the same when you refresh the page. This UI facilitates a "hit refresh, scan down to last familiar timestamp" workflow that I use in lieu of purging (
http://grumet.net/weblog/archives/2003/08/19/more_on_deletionless_aggregator_reading.html ).
