Will the code you depend on keep getting commits and releases?
For engineers and coding agents choosing or auditing a dependency.
For each repository we cover, Anteproof publishes two dated probabilities: will it reach its commit target, and will it publish a release.
An npm, PyPI or crates.io name, or a GitHub owner/repo. No account needed.
Our list is drawn from the most depended-upon packages on npm, PyPI and crates.io, ranked on 4 September 2026. A repository gets each forecast only when it meets that question's rule.
Here is one live answer. Each question shows its probability, the day it was issued and the day it resolves.
The commit target is the repository's commit total in the 90 days before the question was issued. The forecast covers the 90 days from that issue date. Both questions resolve from the GitHub REST API.
Loading the latest forecast from the record…
Try to see the two probabilities for expressjs/express, each with the date it resolves.
Try to see the same two questions for psf/requests.
Try to see what a package with no open forecast looks like.
Check a list of packages
Paste up to 25 package names into the coverage preview to see which of them we can forecast today, and why the rest can't.
Upload a CycloneDX or SPDX file on your account page. Dependencies we cover are matched, and you can keep some of them on a watchlist.
What we do not claim
We never describe a package as safe.
A missed commit target says only that the repository made fewer commits in the 90 days after the question was issued than in the 90 days before it.
Commit activity alone does not tell you whether a package meets your needs.
Maintenance forecasts are new. Their past scores are withheld, because their outcomes were counted from a third-party archive of GitHub events we have stopped trusting.
A probability is not advice.
How to check us
Each forecast goes into an append-only log. The head of the log is anchored daily to a public repository, github.com/anteproof/log.
Scores use the Brier score. Lower is better, and always saying 50% scores 0.25. Live forecasts and point-in-time pastcasts are scored separately. With fewer than 30 resolved rows, the number is shown but says little. Until a family has resolved rows of its own, treat its probabilities as unscored. The calibration page has the details.
Coding agents get the same answer as an MCP tool. Calls without a key are for trying it out. You can get a free key on your account page. The developer docs have the setup and the quotas.