The correct interpretation there is that for any given R-S code, the tolerance for correctable errors is half as many bits as the tolerance for uncorrectable but still detectable errors. If you have an R-S code constructed to correct 6 bits or detect 12-bit errors, then for a given message:

if the number of corrupted bits is no more than six, you can reconstruct the original message

if the number of corrupted bits is between 6 and 12, you will detect the corruption

if the number of corrupted bits is larger than 12, you may or may not detect the corruption.