HackerBooks counts the arguments too

HackerBooks went up on Hacker News this morning and pulled 661 points — a scraper that trawls the comment threads for book titles and ranks them, pitched as “HackerTube, but for books”. So that shape is a genre now: take a decade of arguing, extract one class of object, sort by frequency. What falls out the far end is a canon, and nobody wrote it.

The part a quick glance misses is that a counter has no ear for tone. “Seeing Like a State rearranged how I look at infrastructure” and “Seeing Like a State spends four hundred pages saying the map is not the territory” both add exactly one. A warning is a mention. “I couldn’t get past chapter three” is a mention. What the ranking actually measures is friction — which books are long enough to be contested and famous enough to work as shorthand in a reply you’re writing at speed. Affection doesn’t register at all.

There’s a second bias baked into the matching. A title has to be weird enough to catch. Gödel, Escher, Bach has an umlaut and two commas in it; it is unmistakable in a block of text and it will sit near the top of any list built this way, forever. Meanwhile a novel called Go is simply unrankable on a site where go is a programming language, a verb, and a board game. The C Programming Language survives only because the phrase is long. So the canon tilts toward books with awkward names, which is not a quality anyone has ever read for.

If someone made me put money down, I’d bet on Hofstadter being in that top ten and I’d bet comfortably. I would not put a dollar on most of the people who typed it having made it past the Crab Canon. The honest version of the site has a second column beside the count: how many of those mentions were somebody putting the book down.