Hacker News held its character through 14x growth by policing behavior, not users

Hacker News launched in February 2007 as a side project: an application to sharpen Paul Graham's Arc language and a place for Y Combinator founders to exchange news. Weekday traffic ran about 1,600 daily uniques, and by the second birthday in February 2009 it had grown roughly fourteenfold, to about 22,000. The growth pleased and worried Graham at once, because user-voted link boards had a history of being ruined by scale: he had watched Reddit — built for maximum growth, censoring nothing but spam — decay into a free-for-all.

The obvious cures were the wrong ones: chase growth like Reddit, or defend the gates with barriers and bans on people. Graham concluded that dilution is measured in behavior, not users, and behavior is malleable — people who are expected to behave well generally do. Forbidding bad behavior also filters softly: people who cannot follow norms feel constrained and leave on their own. He applied the broken-windows idea from criminology, treating small tolerated disorder as the seed of worse, and used Reddit's opposite trajectory as the control experiment.

The move was a stack of low-tech policies. Links easiest to judge take over a voting site — Graham called this the Fluff Principle — so HN banned easy upvote bait like kitten pictures and political diatribes, let editors kill empty rants and rewrite linkbait titles, and banned domains that habitually linkjacked others' work. A showdead switch let any user inspect killed submissions, keeping moderation honest. Graphics stayed plain and titles sober, leaving ideas as the only draw. A comment-score test (orange usernames) split the culture into haves and have-nots and was withdrawn within days.

At two years the mechanism looked to be holding: front-page stories were still roughly the ones that would have been there at launch, and performance stayed consistently mediocre through fourteenfold growth because Graham removed the biggest bottleneck with the least code. A week before the essay, the birthday thread had been full of members crediting the community and its moderators, while some worried growth would eventually dilute it. The essay itself drew 381 points and 119 comments on Hacker News.

Police behavior, not users, to keep a community's character

Graham watched the same people post flame versions on Reddit and subdued versions on Hacker News — behavior follows the room, so norms plus light enforcement filter better than banning people.

Content that is easiest to judge — cartoons, rallying rants — crowds out content that needs thought, so HN banned fluff categories at the source instead of trusting votes to sort them.

A showdead switch let any user inspect what editors had killed, so the community could verify moderation and trust editors enough to let them keep acting.

Plainness is a recruiting filter: with no graphics and sober titles, the only reason to arrive is the ideas, which attracts thoughtful users and repels spectacle-seekers without a wall.

Reversible experiments kept mistakes cheap: the orange-username ranking split the culture into haves and have-nots within days and was pulled immediately, so one wrong call never became entrenched.

Dilution is a behavior problem, not a headcount one: set explicit norms, ban the easiest content to upvote, keep the surface plain enough to repel the wrong users, and revert social experiments fast.

By the second anniversary weekday traffic stood near 22,000 daily uniques, and Graham called the growth rate a little higher than he would like. The orange-username experiment stayed dead — if you are going to distinguish between people, you had better be sure you do it right — while the fluff bans, plain design and editor transparency remained. Graham kept treating the site as an experiment in a young field, kept performance mediocre by removing the largest bottleneck with the least code, and named the addictiveness of social applications as the next unsolved problem.