I always thought that the xBestIndex invocations are made in no particular order and sqlite even free to repeat them if it wishes so even if some constraint case was not sufficient for me, I still can wait for a better one following. That's why the answer in my tables always either SQLITE_OK or SQLITE_CONSTRAINT (after version 3.26.00) and no SQLITE_ERROR since the latter will interrupt the variants enumerations. But you explanations implicitly assumes some possibly documented ordering logic allowing one to detect in a xBestIndex invocation that there won't be better one for particular column set and one is safe to fire SQLITE_ERROR and breaks the loop. Can you confirm this?