What a forecast looks like, and how to check it
Every number on this page is fetched live from the API when you load it. Nothing here is a screenshot, and nothing is written down in advance.
The worked example
None of this is a vulnerability report. A CVE feed, a scanner and a bill-of-materials entry would all be unchanged by it. They describe what is in the product. These questions ask whether the repository will keep up its commit pace and publish a release.
Read the two question rows together. Commit pace and release pace are answered separately, and the gap between them tells you the most.
Try it yourself
Why two questions
A single health score cannot say that a project commits rarely and ships every week. Two dated questions can.
Which card is the quiet-but-shipping case and which is the busy-but-rarely-shipping one depends on the numbers on the day you load the page. The packages can swap places.
The same forecast, read three ways
The card above is fetched once and read three ways, one for each kind of team we think might use it.
Tell us which one you are, or that none of them is: anteproof.com/support.
How to check we are not marking our own homework
Every forecast goes into an append-only log, and we anchor the log's head to github.com/anteproof/log once a day. A live forecast becomes checkable from outside with the first receipt after it was issued, which can be up to a day later. Anchoring started on 2 September 2026, so a forecast that resolved before then has no receipt older than its outcome. If a question resolves inside that daily gap, that one will not either. Both kinds are in the scored record, which is public.
The maintenance forecasts have no track record yet. Every past question was resolved against a
third-party archive of GitHub events that we no longer trust, so we withhold the count and the score. The response
you just fetched says so itself, in calibration_note.withheld. The current state of every
family is on the calibration page and in
/v1/proof.
What does not answer yet
These are tracked and do not carry a forecast yet. The API resolves each one and returns a plain note saying what is missing.
This list is a live query against a fixed set of well-known packages, run when you loaded the page. When it is empty it is not here at all.