A 'data avalanche' refers to starting with a bit of data - say, the name of a function - and using that to get more data - say, the function's arguments can be 'tagged' with types. Then once you know the types, perhaps those arguments go into another function, and we can decompile that function knowing the arguments.