Gregory
>A go recno() instead of a go top solves the problem as well.
>Speed does not matter here, it's only 500 recs in a cursor

Speed is a matter for my large datasets... But if the go recno works, this makes me think it is a update/caching problem. In case I am correct, RLock() should also make the problem disappear. There is also a distinct possibility that this bug is fixed - maybe Sergey (if he happens to read this as well <g>) can provide a direct link, as he probably oodles faster in finding such things.

>
>I'm still using vfp6, moving to vfp9 is scheduled end of this/early next year. I'll need a couple of months to do the migration
>
>But whatever the outcome of vfp9 may be, I've put the assert in. The form calculates taxes and the result better be right

understandable. Ummm... guessing that you are mostly hemmed in by large organization / deployment / licensing hassles in porting: I have been there as well. I was officially allowed to work on my laptop as well as the given company-administered desktop, and on my machine I could use vfp8/vfp9 versions long before the "official" port started. The tightening up of the compiler was not much of an effort (<1H) and it gave me a large leg up when the rubber officially met the road to have tested/developed already under the next version, even if I could use new features only in compiler switch brackets. I could solve other issues not under time pressure but "dream up" solutions for instance when driving, which also helped a lot. Just a hint from something that worked extremely well for me.

regards

thomas