Every "best open source X" list has the same problem: half the projects are ghost ships. Last commit 2023, an issue queue of 900 unread, a README that still advertises a version two majors old. You find out after you've installed it.
Here's the checklist I run now. It takes about five minutes per project and has saved me more time than any comparison article.
1. pushed_at, not the last release
The release page lies. A project can cut a "v2.1.0" that's a version bump and nothing else, or ship nothing for a year while the repo is very much alive.
What you want is the last commit to the default branch. In the GitHub API that's pushed_at. If it's within a couple of weeks, someone is in the code. If it's six months old, ask why before you install.
Caveat: a mature project can legitimately go quiet. The distinction is quiet with zero open PRs (maintainer may be gone) versus quiet with 40 PRs merging (it's fine, the release cadence is just slow).
2. Open issue count relative to stars
Raw numbers are useless. The ratio tells you something.
A project with 100k stars and 200 open issues is being triaged. A project with 50k stars and 3,000 open issues is drowning — or has decided not to triage. Neither is automatically bad, but you should know which one you're adopting, because it predicts how your own bug report gets handled.
3. The license, actually read
"Open source" is not a license. I check the SPDX identifier and then look at the exceptions:
- MIT / Apache-2.0 / BSD — boring, use it
- AGPL-3.0 — fine for internal use and self-hosting, a real consideration if you're exposing it as a service to others
- Sustainable Use / BSL / "fair-code" — not OSI open source. Internal use is usually fine; competing with it isn't
The GitHub API returns NOASSERTION for licenses it doesn't recognize. That's your cue to read the file, not to assume it's fine. n8n is the classic case here: 200k+ stars, hugely capable, and deliberately not OSI-licensed.
4. Multi-platform release artifacts
Open the Releases page and count. If the latest release ships Windows .exe/.msi, macOS .dmg, and a Linux .AppImage or .deb, the maintainers care about people who aren't going to build it from source. If it's source tarballs only, you're expected to have a toolchain — which is fine, just decide now whether that's a thing you want.
5. Star velocity, not star total
Total stars measure how famous something got. Velocity measures whether it's still moving.
This is where a "fastest growing" view beats a "most starred" view. A project at 8k stars adding 400 a day is doing something right now. A project at 200k stars adding 20 a week is coasting on history.
I use GitHubPlay's rising board for this, because it separates the four questions people actually ask — 24h hot, fastest growing, newly released, all-time stars — instead of collapsing them into one "popular" list. The numbers come straight from the GitHub REST API, so it's the same data as api.github.com/repos/... just ranked.
The five-minute version
1. pushed_at within 2 weeks? → alive 2. open issues / stars < 5%? → triaged 3. SPDX is MIT/Apache/BSD/AGPL? → safe to adopt 4. latest release has my platform? → install-able 5. star velocity positive? → not coasting Four out of five is usually enough. All five failing means you found a tutorial, not a tool.
One more thing I learned the hard way
Check who's behind it. A single-maintainer project with 80k stars is a bus-factor problem, not a quality problem — but you should know. Look at the contributor graph, the org, and whether there's a company funding it. That tells you what happens when the maintainer gets a full-time job.
The meta-point: the useful comparison isn't "which project has more features". It's "which of these will still exist in two years, and will I be able to leave it if it doesn't". Most of that answer is in five API fields.
