How Hacker News ranking works: scoring, controversy, and penalties (2013)

A forensic analysis of Hacker News ranking shows that story order is driven by a published score formula - score = (votes−1)^0.8 / (age_hours+2)^1.8 - plus a variety of penalties and operational quirks. Because the time exponent ("gravity") is larger than the votes exponent, stories peak quickly and decay, and rankings are updated lazily: a story is reranked when it receives a vote, one random top-50 item is reranked every 30 seconds, and pages may be cached for ~90 seconds. Plotting raw scores over a day reveals rapid early vote growth followed by gradual decline, occasional very high-scoring outliers, and frequent mismatches between raw-score leaders and the displayed #1 caused by penalties.

Substantial automated and manual penalties alter what reaches the front page. Titles containing "NSA" were observed to incur ~0.4 penalties (later removed), and many popular domains showed penalties in the 0.25-0.8 range. A severe "controversy" penalty activates empirically when comments exceed upvotes and hit about 40 comments (published code suggested >20 and a (votes/comments)^2 scale, but observed behavior implies a higher exponent and a 40-comment trigger), causing posts to plummet or disappear. Penalty magnitudes cluster near 0.2, 0.4 and 0.8-0.9; roughly 20% of front-page and 38% of second-page stories were penalized. The findings derive from minute-by-minute crawls and comparing computed raw scores to observed ranks to infer penalty factors.