>I've confirmed that SetFldState is NOT used in this project.

Improbable, but something like

local lcMak
lcMak = m.tcActChar + "etFldState("+m.tcParStr+")"
luRes = eval(m.lcMak)

might miss direct search for the string in project.

>>UPDATE
>>
>>A loop checking OLDVAL might reveal a field changed, but not visible to GetFldState. It's just an idea for testing.
>
>Hmm, not sure what that would look like, but again, I can't track it down.
>
>I guess I could add to the error handler some code that runs only in case a record found as changed by GetNextModified returns all 1s to GetFldState, in case this ever happens again.

Along the "valid" line of thought: transaction working on the buffer (to set dependant fields?) rolled back ? Or does transaction rollback fall under "show changed, but back to original value"?

Probably the only way to check/verify is to log writes to the buffer in separate file/table and clean on tableupdate/quit.... wrapping each replace or update is ugly - worked once on an app where table/field access was isolated GUI on via Biz, Data and some more layers and this created stack sizes I still shudder at.