Okay, the title is kind of cryptic because I'm lacking terminology (part of the problem).
Let's say you want to generate data corresponding to one specific, standardized paper document. But you have millions of people filling out this document, some of them are lying -- but there's several different ways (possibly hundreds) of lying, and many of the document's fields are interdependent. I want to generate all these types of lies via some randomized distribution.
Additionally, with respect to this form, it's impossible to go through each field of the form in a sequential order and capture the nuance and inter-dependency of the data/lie relationship. Instead, you could (the path I see) opt for a method where each type of fraudulent form has hard and fast rules for its generation -- logic and static values contained in some config file -- and then just boil it down to copy and pasting the function into function_1, function_2 etc, which are almost identical besides the small changes which capture the characteristic of whatever lie they correspond to.
But then it's sort of ridiculous to write a function for each of those "lie rules", even if you're almost copy-pasting. But I've never encountered a situation where there's so many and unique cases (some with large differences and logic) for creating the same piece of data. Is there well-known approach/name to this problem? Or does anybody have any suggestions?