The Help for the SET TABLEVALIDATE Command does not communicate what happens when a table's integrity is detected as compromised.

I have been particularly interested in its nLevel setting to 2, which is designed to "Validate record count when appending or inserting records and writing them to disk.".
I see this setting as the more useful of them all given the need of nLevel values of 1 or 3 to LOCK the table header in order to process.

My limited testing of the operation of SET TABLEVALIDATE TO 2 has yielded some surprises and I think they are worth sharing. I would be most interested to hear if others obtain similar results.

Most surprising is that I was unable to intercept an error condition using either ON ERROR or TRY...CATCH...FINALLY when buffering (row or table, optimistic) was in effect.
I had expected these to trigger, but they didn't.
The only way I was able to detect the error condition was to process the non-TRUE return from TABLEUPDATE()!

In addition, and regardless of buffering style (or not), there was never an error reported when the record-count inconsistency was only 1 record, indicating strongly that the validation processing is performed after the INSERT or APPEND is completed.

When the record-count inconsistency was 2 less than expected then an error (#2091) was raised. HOWEVER, in this case it was only when there was no buffering in effect that the table's final status was legitimate (at least as I see things).
The original table had

RecCount = 5, LogicalSize = 339, PhysicalSize = 339

When the table was (artificially) corrupted to have a record count of 3 and with counts reflected as

RecCount = 3, LogicalSize = 335, PhysicalSize = 339

and then an UNbuffered single INSERT INTO... was executed to add a 2 byte record the resultant table showed values of

RecCount = 4, LogicalSize = 337, PhysicalSize = 339

and the last record (#4) contained the value INSERTed ("6").

However, the same operation done with ROW BUFFERING in effect resulted in

RecCount = 5, LogicalSize = 339, PhysicalSize = 339

and the last record (#5) did NOT contain the value INSERTed but rather the value of the last record that was 'corrupted out'. The expected result is

RecCount = 4, LogicalSize = 337, PhysicalSize = 339

With TABLE BUFFERING in effect the RecCount/LogicalSize/PhysicalSize values were the same as for row buffering but in this case the last record shown by the LIST command was #4 (as expected) and its field value was "6" as was the value INSERTed.

Admittedly there may be something amiss in my tests, so any comparisons are most welcomed.