Every week a script of mine scans four sources — Show HN, new GitHub repos, Lobsters, Product Hunt — and posts the ranking to Telegram. This week I pulled the raw data back and re-derived every figure in code. The counts were right; the ranking was not. This post covers the dataset; the defects are in the next one.
The dataset
- 615 repos created after 19 October 2026 that already passed 40 stars, as of 11:12 on 26 October.
- 95,350 stars in total — 155 stars per repo in under seven days.
- One GitHub Search API query: `created:>2026-10-19 stars:>40`, sorted by stars.
Languages: Python still takes nearly a third
- Python 172 repos (28.0%) — more than TypeScript and JavaScript combined (108, or 17.6%).
- HTML 75 (12.2%): almost entirely demo pages, single-file dashboards and skin/script repos.
- Rust 30 (4.9%) and Go 24 (3.9%) — small, but this is where the structured projects are.
- 48 repos have no language at all, usually because the repo is a README or nothing but `.github`.
The point is not that Python leads, but the gap: a quarter of the week's new repos are single-file Python scripts. If you are looking for a base to run a service on, that ratio says most of this list is not products — it is scripts, demos and proofs of concept.
The trap: stars-per-day rewards the newest repo
The original script ranked by stars divided by repo age. That sounds fair but pays out for being new. A 41-star repo at 0.3 days old displays 137 stars/day; a 5,226-star repo at 5.8 days displays about 903 stars/day. The consequence is a leaderboard pushed up by repos that are hours old.
- 25 of the 615 repos (4%) are under a day old — exactly the group the formula rewards most.
- The 5,226-star repo at 5.8 days shows ~903 stars/day — yet still ranks below word-atlas: 811 stars at 0.82 days, or 985 stars/day.
- The same formula on Show HN: a one-hour-old post with 24 points shows 24 points/hour, ranking above a 938-point post at 130 hours (7.2/hour).
How to read it instead
- Raw stars tell you how hard the community reacted — use them when you want to know what is getting attention.
- Stars/day tells you speed — use it to catch things early, but filter by a minimum age first.
- By raw stars this week's top six were aerial-tv (5,226), gpu-unlock-ada (2,931), hatchzip (1,955), 3d-model-anything (1,947), php2shell-poc (1,797), day-trader-skill (1,697).
The full 100-project list of open-source worth running as a SaaS is still the roundup by cumulative stars — that data is unaffected by the speed defect, because it never divides by repo age.
See also: which of these 615 you can actually sell and what stars do not measure. Sources: GitHub Search API, HN Algolia API, data as of 26 October 2026.