Some of them look good at a quick glance, polished readmes, recent activity, commit history, features lists..etc. But when you look at the code they turn surprisingly bad, surprisingly fast, low effort, low quality, unmaintainable messes of patches that break compatibility every other version. AI has made this 100x worse.
In the past, if a project had a relative high number of gh stars and activity in relation to the niche that would mean the project had some standard of quality/potential/efficacy and you could review 1-3 projects, basically it acted as a good first filter, today there are hundreds.
How do you find the signal in the noise, tips/tricks? Some heuristics I use: multiple contributors, less is more when it comes to commits per day, if the author uses AI a small signal is if they limit/block acceptance of PRs, codeberg or outside github repos and of course AGENTS/CLAUDE.md, especially CLAUDE.md.
Curious what signals you found useful, especially ones that are difficult to game and can be semi-automated so at the end you only review the fewest number of them left? loc, file count, other registries?