Brent Sampson has figured it out, and it’s remarkably unlike how one expects it to be calculated. Which in turn means that what it means is different than what one would suppose it to mean.

In actuality, the process is somewhat more convoluted than they let on. Only the top 10,000 books are updated every hour and the ranking does not depend upon the actual number of books sold, but rather, on a comparison against the sales figures of the other 9,999 books within that same hour. Simultaneously, a trending calculation is applied to arrive at a computerized sales trajectory. So, hypothetically, a book that held a ranking of 2,000 at 2pm and 3,000 at 3pm, might hold a 4,000 ranking at 4pm, even if it actually sold MORE books between 3-4 than it did between 2-3.

Books with rankings between 10,000 and 100,000 are recalculated once a day, rather than once an hour. Current projections, as well as historic sales information play a key role in these calculations. In fact, the predictive nature of the Amazon ranking system is what makes it possible for a newly-released book to outrank an older established title, even though the actual sales figures for the latter far exceed the former.

(Via Bookninja.)