Maybe because startups are small enough and therefore simple enough (in terms of variables affecting success) for something like this to work, but large public companies wouldn't be.

But really I'd be skeptical about data in this first batch of tests. I'd bet that YouNoodle used these widely known startups as a training set for its algorithm, or at least as a test set...I imagine if one of their unit tests came back and said "Facebook is gonna be worthless", someone would tinker with the algorithms until it didn't say that anymore. Usually you deal with this problem by setting aside some data points until the very end for you to evaluate your accuracy, but without knowledge of where they got their training set, you can't really do this yourself