James Somers's Hacker Classics mined "(1999)" titles to surface HN's best old posts

Hacker News, like most of today's web, is built around what is new. Old articles still surface when a reader resubmits one — usually with the year it was written in parentheses, like "(1999)" — but those resurrections are rare and scattered, so the site's genuinely timeless posts sit in its archives most of the time.

A recommendation engine or a hand-picked editorial list would have been the usual fix, and both cost far more than a hobbyist page. James Somers noticed that the community had already solved the labeling problem: over the years, date-stamped submissions such as "(1998)" tended to be the best posts on the site, because on a site devoted to news, an article posted long after it was published has to be especially good to survive.

So he automated the convention. A script queried the Algolia Hacker News search API for every story with more than 40 points whose title contains a parenthesized year from 1900 to 2010, removed duplicate resubmissions of the same title, sorted the survivors chronologically, and stored them in a stories.json file that the page renders client-side. The result was The Hacker Classics, a browsable shelf of roughly two thousand reposts instead of a newspaper rack.

Posted as "Show HN" on February 23, 2018, the page drew 864 points and 107 comments. Commenters confirmed the mechanism's premise: a handful of submitters — Tomte, luu, tosh and others — had been surfacing most of the classics, effectively serving as the community's volunteer librarians, and readers suggested refinements such as deduplication, sorting by points and Wayback Machine links for dead URLs.

Reposts with '(year)' in the title are pre-vetted classics

The year-in-title convention is a label the community already applies by hand, so the filter harvests accumulated human curation instead of trying to detect quality from scratch.

The vote floor above 40 removes resubmissions that sank, keeping only articles that survived both time and the community's judgment.

An old article reposted long after publication is self-selecting: on a news site it has to be exceptional to reach the front page at all.

Sorting by original year turns a list into a library: readers wander through decades of the stacks rather than skimming another ranking.

When a community marks quality with an implicit convention — '(1999)' on resurrected old articles — treat it as labeled data and extract it automatically instead of curating by hand.

The Hacker Classics stayed online and still loads in 2026, and the GitHub repository documents how the list is built so anyone can regenerate it. The February 2018 discussion made the social layer behind the convention visible: roughly a dozen users had submitted a large share of the year-stamped classics, so the curation the page automated had been happening all along. Readers also asked for point-based sorting, duplicate removal (already handled in the source) and automatic Wayback links for submissions whose original pages had died.