Delimited data. Data file formats are, in most cases, even more rigidly structured than source code. Delimited data is a good example, but the same line of thought applies to many other data formats (XML, however, has a lot of optional white space, which could make for a good subchannel). At the level of content, however, data file formats have non-predictable variation by definition. After all, the point of actually sending a data file is to convey the information in it that the recipient does not know. For example, a row record for a person might have a first name, last name, and social security number, each of which must look a fairly specific way. But the actual SSN a person has is not predictable from the other information. A possible subchannel exists in subtly varying this data content. However, a danger of revelation exists if an attacker has independent ways of correlating data (if no one in your data file has the true match between name and SSN, this looks suspicious to an attacker). Finding this kind of subchannel requires specific knowledge of the data format and content being used.