So yes, after reading this post, you too should be able to rank in the top ten or so. It landed where it did because, just for giggles and grins, we blended results (50/50) with Jetrays who had a similar score to us at the time.
The rotation evaluation can also be performed by the execution of μ-rotations such that the complete SVD-updating algorithm can be expressed in terms of orthonormal μ-rotations.
Now for the method to the mathness, begining with a review of the problem: Netflix provided 100M ratings (from 1 to 5) of 17K movies by 500K users. For visualizing the problem, it makes sense to think of the data as a big sparsely filled matrix, with users across the top and movies down the side (or vice versa if you feel like transposing everything I say henceforth), and each cell in the matrix either contains an observed rating (1-5) for that movie (row) by that user (column), or is blank meaning you don't know.
These essentially arrive in the form of a triplet of numbers: (User, Movie, Rating). To quantify "big", sticking with the round numbers, this matrix would have about 8.5 billion entries (number of users times number of movies).
Note also that this means you are only given values for one in eighty five of the cells. Netflix has then posed a "quiz" which consists of a bunch of question marks plopped into previously blank slots, and your job is to fill in best-guess ratings in their place.
They have chosen mean squared error as the measure of accuracy, which means if you guess 1.5 and the actual rating was 2, you get docked for (2-1.5)^2 points, or 0.25.