The first caveat raised by avalanche effects is in regard to our earlier discussion of how particular plain text bits jump around to specific new positions in the cipher text. This simplification is not really correct. The information in one individual bit of plain text input is not simply moved to a new location in the cipher text, but rather one bit of information is diffused into the entire cipher text. In a very real sense, each bit of cipher text contains, for example, 1/64th bit of information about bit-one of the plain text. It may seem odd to talk about less than one bit of information, but that is fundamentally what we have with cryptographic diffusion.

The second caveat raised by avalanche effects is in regard to S-boxes. The tutorial describes S-boxes as having the same input- and output-block sizes to preserve a one-to-one relation between inputs and outputs. Well, that description is basically true, but may not be how you see S-boxes described elsewhere. For example, DES uses S-boxes that are often described as taking 6-bit inputs and producing 4-bit outputs. On the face of it, anything that does that is necessarily not fully reversible (so no decryption). But the lookup table for DES S-boxes really does have 64 (2^6) entries, and really does only have 4-bit outputs listed for each entry!