This probably still isn't a great idea. To simplify let's assume that the secret data is completely random. Let's also assume that "non-matching parts" is bit-b

This probably still isn't a great idea. To simplify let's assume that the secret data is completely random. Let's also assume that "non-matching parts" is bit-by-bit. This would mean that you accidentally reveal 1/2 of your private details because the random values match. With redundancy in the keys as mentioned in the article it seems like solving for the missing data shouldn't be too hard. Of course the bit-by-bit assumption makes you reveal much more data than buye-by-byte or base64 characters but the point still stands.