Yeah the headline oversells it a bit. The part I actually found interesting was that 500+ point posts die just as fast as 50 point posts. HN's algorithm doesn't care how popular you are. Sections 4-6 have the more useful stuff if you're curious.
interesting, that actually explains a lot. If HN rewrites the timestamp on resubmission/boost, that's exactly why powera saw "14 days ago" for a post our scraper first picked up Dec 8. We were just grabbing whatever hit the top 50 every 30 min, so we caught the original appearance regardless of what the timestamp says now. Good to know.
The "44 days" outlier (Grov) appeared in our snapshots on 12 unique dates spread across 47 calendar days, not continuously. Our methodology measured time between first and last appearance in the top 50, which collapses resubmissions into one window. That's a real limitation we should have flagged.
The raw data: Grov first appeared Dec 8, then clusters on Dec 17-22, Dec 29, and Jan 16-24. Big gaps in between. It wasn't sitting on the front page for 44 days straight — it kept reappearing.
We're updating the study to add a "continuous visibility" metric alongside the existing one. The core finding still holds (99% of Show HNs are gone within days, and the median post gets a single 30-min window), but the outlier framing was misleading.
Fair point, so you think it would be more useful to keep it developer centric? I do like the idea and actually might shift it in that direction. Thank you for the input.
Yeah fair point on the medical MLLM stuff. FDA and modular approaches have never been best friends. That's actually why I'm most curious to see how it plays out - the ArXiv activity around safety grafting has been picking up but whether regulators actually buy it vs demanding full retraining... we'll see.
The OpenAI platformization one I'll admit feels safer. Curious though - do you think the window's already closing or is there still runway for the picks and shovels players?
You're right that most voting is headline-driven - that's definitely a limitation worth calling out.
I went with full article text because I wanted to capture what the content actually delivers, not just what the headline promises. A clickbait negative headline with a balanced article would skew results if I only looked at titles.
That said, you've got me thinking. It would be interesting to run sentiment on headlines separately and compare. If headline sentiment correlates strongly with article sentiment, your point stands. If they diverge, there might be something interesting about the gap between promise and delivery.
Might be a good follow-up analysis. Thanks for pushing on this.
159 stories that hit score 100 in my tracking, with HN points, comments, and first-seen timestamp.
Methodology:
- Snapshots every 30 minutes (1,576 total)
- Filtered to score=100 (my tracking cap)
- Deduped by URL, kept first occurrence
- Date range: Dec 2025 - Jan 2026
For sentiment, I ran GPT-4 on the full article text with a simple positive/negative/neutral classification. Not perfect but consistent enough to see the 2:1 pattern.
Thought about this during the morning. I'm run the posts through ministral3:3b, mistral-small3.2:24b, and gpt-oss:20b this weekend to build a sentiment mapping and see what I get. I'm optimistic about ministral3:3b, but the other two are pretty good at this type of stuff.
The interesting part is in sections 4-6: what the survivors did differently. Happy to hear what would make it more useful to you.