This is clever! Expanding on this, though, you can improve the average result against an arbitrarily-powerful test-writer by first XOR'ing the test answers wit

This is clever! Expanding on this, though, you can improve the average result against an arbitrarily-powerful test-writer by first XOR'ing the test answers with a pre-shared value (key); the resulting value is random (even to very powerful test-writers), so e.g. encoding possible answers into the unused values improves your expected result (i.e. data rate.)