A Study of 41,301 Show HN Launches Puts a Clock on Every Launch

Last June, an airline pilot posted the flight-log visualization project he had been building for years. It hit 1,539 points, the third biggest Show HN of the year. By day three the thread had drawn three comments. In week two it got none.

A developer in New Zealand wanted to know whether that flatline is the exception or the rule, so he scraped every Show HN from the last twelve months through the Algolia HN API: 41,301 launches, plus every comment timestamp on the launches that got at least ten comments, about 100,000 timestamps in all. He put the scraper, the analysis, and the raw data in a public repo, which is the only reason I could re-check every number in this post myself. I did, and they held.

The short version: the pilot's launch was the rule. The median Show HN in his corpus earns 2 points and 0 comments. If your launch got a single upvote from a stranger, it beat two thirds of the field. And if it did catch fire, half the attention was gone within 7.2 hours anyway.

I pulled the latest 250 Show HN launches myself before writing this, the same way the study did: median 2 points, 69% with zero comments. Nearly four months past the study's window, the pattern has not moved.

The median launch is a shrug

The outcome distribution is brutal, and it is worth staring at for a second. 63% of launches finish with 2 points or fewer. About 1 in 12 makes it to 10 points, which on Hacker News is roughly the floor of "anybody noticed." Fewer than 2% ever cross 100.

Every founder I know treats launch day as the milestone. The data says the milestone is usually a shrug. That is not a reason to skip the launch. It is a reason to stop budgeting the following weeks around a spike that, for most launches, never arrives.

The half-life is 7.2 hours

HN publishes no vote timestamps, so the study uses comment timing as the proxy for attention. The author flags that limitation himself, and I will keep it flagged too: comments are the visible half of engagement, not the whole of it. For the 2,066 launches with enough comments to have a curve at all, the median time to half of all lifetime comments is 7.2 hours. Half of everything anyone will ever say about your launch gets said before the first workday ends. 90% is done by hour 26.

The mechanism is the ranking formula, which Ken Shirriff documented back in 2013: points divided by (age + 2) raised to the 1.8 power. Time sits in the denominator with an exponent bigger than the one votes get in the numerator, so gravity beats quality no matter how good the post is. The front page is a conveyor belt. That design is why the site stays readable at all. It is also why nothing stays on it.

Success buys volume, not time

Here is the finding I did not expect. The study split the measurable launches into the top decile, 268 points or more, and everyone below. If big launches hold attention longer, the top group's half-life should stretch. It does not. The top decile runs at 7.6 hours, the rest at 7.1.

The twelve biggest launches of the year all run on the same clock. The 3,346-point monster at the top of the corpus was half done in 6.3 hours. Being a hit multiplies how many people show up. It does nothing to change when they leave.

After hour 48, the median launch collects 4.2% of its lifetime comments, and 71% of launches are more than 90% finished by that point. The second-day cliff is where the conversation ends. Traffic dies faster than comments do; the study cites front-page traffic data showing over half the visitor spike gone within 8 hours. The comment thread is the long tail, and the long tail is thin.

Two clocks, and only one of them is yours

I find it useful to read the same corpus as two clocks running at once. The first is the points clock, the 7.2-hour half-life, the ranking decay, the conveyor belt. You do not control that clock. HN does. The second clock is everything that happens after the thread: the README someone reads on day 3, the search result that still resolves in month 2, the changelog that keeps getting longer, the reply you leave in someone's issue tracker in week 6.

The first clock decides whether you had a launch. The second decides whether you have a project. Founders wear themselves out trying to bend the first clock, refreshing the thread, posting again too soon, arguing with a ranking formula. The second clock only responds to boring, compounding work: keep shipping visibly, keep being findable, keep giving the people who did show up a reason to come back after the thread dies.

None of this means the launch is worthless. A front-page hit still puts your project in front of more people in 48 hours than most projects meet in a year. The study's own numbers say so. It means the launch is a starting gun, and most of us have been treating it as the race.

The author of the study is building a tool for this, and he says so openly in the post. You do not have to use it to use the data. The repo is public, make reproduce regenerates every figure, and if your launch is coming up, you now know the clock is already running when you press submit.

Sources: Your Show HN dies in 7 hours (jonno.nz); scraper and data (GitHub); How HN ranking really works (Ken Shirriff, 2013). Fresh-launch replication pull done October 8, 2026 via the Algolia HN API.