row N values differ from index X — For non-BINARY collation indexes, verifies that the index actually stores the correct values. Without this, an index using e.g. NOCASE collation could have stale data that Found wouldn't catch (since Found uses the index's collation).