It seems my sysindexes table is inaccurate on a nonclustered index. In my case the rowcount (rows and rowcnt) do not match the actual rowcount of the table. The command UPDATE STATISTICS doesn't change the rows or rowcount, adding 'FULSLCAN' won't budge rowcount either.
After I did a dbcc reindex, the number of rows matched, however, upon adding rows in the table both rows and rowcount are out of sync again.
It's a fairly straightforward table, no triggers, no computed fields, only integer, datetime, varchar and bigint columns. There's a clustered index on a bigint column and a nonclustered index on a integer column.
dbcc show_statistics show that the nonclustered index is updated and it's rows and rows sampled match the number of rows in the table (not in the sysindexes-table).
I'd like to know if I'm chasing ghosts here or if there's something very wrong here. What could be causing the counts being inaccurate? Anyone who could shed some light?