Writing a useful program is half the work. The other half is getting it onto someone's computer - built for their machine, signed so they can trust it, kept up to date - and that half is usually done badly, or copied from the last project and done badly again.
The same five hundred lines, again
Two of our own tools carry nearly identical delivery machinery: cut a release, build it for two kinds of processor, prove where it was built, install it on six systems to see it works, and publish a signed package repository. Each copy drifts. A fix made in one never reaches the other. And the next project starts by copying the newest one - including its mistakes.
One conveyor, called, not copied
The conveyor does the delivering once, for everyone. A project keeps one short declaration - its name, how its releases are named, which systems it ships to, which key signs it - and calls the conveyor at a pinned version. Improving the conveyor improves every project that moves to the new version, and none of them is changed by surprise.
Like a postal serviceNobody builds their own mail van to send a letter. You write the address clearly and hand it over; the service knows the roads.
What the machines actually run
It ships what our machines use: Arch Linux packages for the lab and desktops, Alpine packages for postmarketOS phones - from one declaration, each with a signed repository. Other systems join when someone needs them.
What it does not do
It does not decide when a project releases or what its version is: that stays with each language's own tool. It cannot make a broken program work - it can only refuse to publish one that fails its checks.
Where it stands
Designed, from the delivery machinery of snapall and lab-agent. The first step is to move both onto it with nothing visible changing for anyone who installed them; then Alpine packages, and the phone's shell as their first user.