Netflix‘s recommendation engine could influence entertainment’s future more than many execs realize. That’s why anyone who cares about the business should check out Alexis Madrigal’s fascinating article — “How Netflix Reverse Engineered Hollywood” — posted today by The Atlantic. Perry MasonThe company “has meticulously analyzed and tagged every movie and TV show imaginable,” the article says, giving Netflix “a stockpile of data about Hollywood entertainment that is absolutely unprecedented.” The people it uses to tag content work from a 36-page training document. “Every movie’s ending is rated from happy to sad, passing through ambiguous. Every plot is tagged. Lead characters’ jobs are tagged. Movie locations are tagged. Everything. Everyone.” They come together to produce 76,897 genres including “Scary Cult Mad-Scientist Movies from the 1970s” and “Feel-good Foreign Comedies for Hopeless Romantics.” Others note whether a film is an Oscar winner, or has a strong female lead. While the possibilities are limitless, as a practical matter “there are no genres that have more than five descriptors” — and the most popular ones have three or two. The information influences recommendations more than users realize. “We’re gonna tag how much romance is in a movie,” VP of Product Todd Yellin tells Madrigal. “We’re not gonna tell you how much romance is in it, but we’re gonna recommend it. You’re gonna get an action row and it may have more or less romance in it based on what we know about you.”

But some things Netflix can’t explain — including why Raymond Burr is users’ favorite actor (ahead of Bruce Willis, George Carlin, and Jackie Chan) and his co-star from the courtroom drama Perry Mason, Barbara Hale, is No. 7 (ahead of Clint Eastwood and Elvis Presley). “The more complexity you add to a machine world, you’re adding serendipity that you couldn’t imagine,” Yellin says. “Perry Mason is going to happen. These ghosts in the machine are always going to by-product of the complexity. And sometimes we call it a bug and sometimes we call it a feature.”