>I have the impression that the indexes get corrupt by a sequence of actions
>
>A couple of things are worth trying
>
>(1) limit the cache
>

>=sys(3050,1, 64 * 1024 * 1024)
>=sys(3050,2, 16 * 1024 * 1024)
>

>
>(2) after every insert/update
>

>flush
>=sys(1104)
>

>
>(3) go top, then seek
>
>
>You'll quickly find out whether the above helps

Haven't yet tried your idea. I cleaned indexes in Stage tables in my local copy (which already had one Invalid seek offset) and then try to run it again. I got:
Error: 1103
LineNo: 1099
Message: Invalid seek offset.
Procedure: runprocess
Details:
StackLevel: 2