I try to explain further:

There is NO duplicates on kornid. I have doublechecked. The table looks perfect. But the parallell log shows that the ID's went from 12316 to 12612 before it started again as if the table was restored from a backup. This was what I first checked, but it can't be the case as many record before 12316 is updated during the day.

Flocing the other table is to make sure nobody else is adding a new kornid at the same time, resulting in duplicate values.

Since a TableUpdate is issued before moving to the last record with the highest kornid, there is no implicit TableUpdate.

Checking the log-files, I find that only one PC was adding 421 new records yesterday. Another user did update 1010 records. A third user updated 573 records. It was Saturday and only few people working.

I don't have a SET REFRESH, but I don't see why this could be the reason.