I published a dataset today: every Show HN post for the last 24 months — 70,894 launches, built with the free HN Algolia API only (no scraping, no key). The headline isn't the doubling. The headline is what didn't change while it happened.
The median launch gets 2 points
Across 70,894 posts: median 2 points, p75 = 4, p90 = 9. Only 9.5% reach 10 points, 2.2% reach 100, and the max in the whole window is 3,346. 61.6% of Show HN posts never receive a single comment.
The launch page is a lottery, and this is the ticket-stub view most "how to launch on HN" posts quietly skip: they profile the p99 winners and call it strategy.
What doubled — and what didn't
Monthly submissions more than doubled over the window (~2.4×): ~1,600/month at the start (Oct–Dec 2024) to ~3,900/month recently. Meanwhile, per month:
- the median stayed at 2 points in 24 of 25 calendar months (June 2026 inched to 3),
- the never-commented share stayed inside a narrow 0.55–0.68 band the whole time. More people are launching; the audience's attention did not double. The lottery got more tickets, not better odds. ### When a launch happens, it happens early I sampled two October weeks — chosen because they sit exactly one year apart in the dataset, not for launch-season representativeness — and pulled the first comment for every story: 514 stories from Oct 2025 (176 ever commented) and 979 from Oct 2026 (322 ever commented), 1,493 total. For posts that ever get a comment, the median time-to-first-comment is ~4.4 hours, and it was nearly identical a year apart (268 minutes in Oct 2025, 264 in Oct 2026). About 30% of commented posts get their first comment within an hour. Early silence is the norm, not a signal. Even posts that eventually get comments stay quiet for over four hours first (median), and the never-commented share barely moved year over year (65.8% in the 2025 week, 67.1% in the 2026 week — both near the top of the full dataset's monthly 0.55–0.68 range). Most "my launch flopped" post-mortems are describing the median outcome, not a failure. ### One number I didn't expect github.com is 32.7% of all Show HN posts — 23,154 of 70,894. A third of everything people show is a repository. The next domains aren't close: apps.apple.com (917), the Chrome Web Store (713), YouTube (408). ### Method, and the honest caveats The dataset and the whole pipeline (pull, analyze, sample) are public and dependency-free. Points and comment counts are snapshot values at pull time, so recent posts' numbers can still move; October 2024 and October 2026 are partial months; the TTFC sample covers two October weeks chosen to be one year apart, not launch-season representativeness. Points have a floor of 1 (the self-vote), which is why "no engagement" here means zero comments, not zero points. I verified 5 of 5 randomly sampled rows against the live API at publication time, and the pull is complete by construction: the sum of per-window result counts reported by the API equals the 70,894 rows collected, slice by slice. (One measurement trap worth knowing: for a single query spanning the whole 24 months, the API's reported total ignores the date filter and returns its all-time count — per-slice counts are the ones that reconcile.) "Show HN has been analyzed" is fair — blog posts profile winners every year. What I couldn't find was the boring artifact: the full fresh dataset with the median outcome visible. Now it exists, it took one free API, and anyone can re-run it in about five minutes.
Top comments (2)
the doubling with an unmoved median is the finding that matters. the front page has a fixed number of slots, so more submissions just means more tickets for the same prizes. would be curious whether the group that reaches 10 points thinned out over time too - that would tell you if even the winners are losing margin.
Ran your question against the full 70,894-post set. You called it — and it's worse than the ticket metaphor suggests.
Share of submissions reaching 10 points, half-year windows: 14.5% (2024 H2) -> 13.0% -> 9.4% -> 7.8% (2026 H1). Nearly halved. The "at least a little traction" line (4+ points) fell too: 32.6% -> 23.9%. So it's not just that the front page is zero-sum; the floor is rising out from under the middle.
And your winners-losing-margin instinct is also right: median points WITHIN the 10-point club went 38 -> 38 -> 32 -> 30. The people who do clear the bar are clearing it by less. Same prize pool, more tickets, thinner slices.
Caveats in the interest of honesty: 2026 H2 is partial and recent posts are still accruing votes, so the last cell is depressed by construction — the clean comparison is 2024 H2 vs 2026 H1. Points are also one snapshot taken at crawl time, not full histories (that data stopped being collectable — that's the subject of my other piece). But the direction across four settled windows is unambiguous.