The test relies on statistics which can be non-deterministic.
The stats can be incorrect, with the pre-images still getting deleted, if the following occurs:
(1) Pre-images exist for nsUUID0 when we install the initial truncate markers in the truncate map.
(2) More pre-images are inserted for nsUUID0 before refreshMarkers() is issued.
(3) refreshMarkers() updates the highest seen recordId, but doesn't know how many pre-images were inserted since the initial markers were created, so it reports 0 additional pre-images.
(4) The 'stats' rely on the truncate markers being correct with their numRecords count. Initial truncate markers are estimates, and may not have a reliable view of the number of records within the marker's range.