How it works

Every day at 00:00 UTC, tierlist.tech reads the GitHub repositories created in the last 72 hours, removes the ones with bought stars, and has a language model judge every one that is left against a published rubric. The judgement becomes a tier from S to D with one line of reasoning. Nobody picks or edits the results by hand.

1. 1Collect

   A GitHub search for repositories created in the window that are not forks, not archived and have a README, most-starred first, up to about 300 repositories a day.

2. 2Remove bought stars, for free

   • Bought stars: when 5 or more repositories share an identical star count and were created within 60 minutes of each other, the whole batch is dropped.
   • Every other repository is rated, with no star or description minimum. A repository with no description or no stars is the model's to judge, usually as D. A cap of 1000 a day guards against a source misbehaving.

3. 3Judge

   The model reads like a demanding open-source editor: one small tool done well beats an empty platform, and stars are evidence, not the verdict. It scores four dimensions from 0 to 25, which are weighted into an internal score. That score is never published; only the tier is.

   Dimension	What it asks	Weight
   Novelty	Is the problem, or the way it is solved, one nobody has done well lately? Another wrapper, todo app, list or scaffold scores low.	×1.2
   Usability	Could a developer install it and use it today: does the description say what it does, who it is for and how to start?	×1.2
   Craft	Is the scope clear and restrained, and are the name and description honest and accurate rather than inflated?	×1.0
   Momentum	Star velocity, age and signs of real use. Evidence, never the verdict: bought stars and borrowed hype do not count.	×0.8

   Always rated D: bought stars or squatted mirrors; cracks, cheats, piracy or paywall bypasses; pure content (prompt collections, resource lists, tutorial notes); coursework and practice projects; repositories that do not say what they do.

4. 4Tier

   The lines are percentiles of this rubric's own scores, measured on 30 days of repositories: the top 2% are S, the top 10% A, the top 30% B, the top 60% C, the rest D.

   S

   Look today Rare: new, usable today and well scoped. Worth a look now.

   A

   Worth watching Clearly worth watching, with one weak spot.

   B

   Has a spark A real strength, but narrow, early or up against strong alternatives.

   C

   Ordinary Usable but not new, or new without a clear user.

   D

   Skip Not worth the time: a rewrite of something that exists, a content list, bought stars, a crack, or no clear purpose.