Looks like you're getting the hang of it: good steganography relies on having better statistical models than the opposition.

Things are weighted in favour of the concealer, however, as (unlike the attacker) he doesn't have to worry about false positives/negatives