A further comment on this corruption, for anyone interested.

I found an effective way to detect this problem is to commit an appended record with only the essential key fields, and then read it back and update it.

To test it, I used a variation of the method suggested in another KB, to corrupt the header by reducing the correct header count by 1.

It tested out correctly in three out of four fairly critical tables, updated in half a dozen or so places. In the fourth case, the header was corrected by VFP, after failing as it should have on the first attempt.

Not entirely correct, though. In the update which corrected the header, the last record ( which I 'discarded' ) was overwritten. In practice, this might not matter, because the corruption to the header would not be as 'clean' as in the test case.

It does, I think, suggest that the failure to report the error may well be related to VFP deciding that it has 'fixed' the problem, when in fact it hasn't.

Geoff