Delegated, pending approval

Two people. Ten months. Nineteen new committers onboarded, the merge bot extended, a GitHub security advisory triaged, and an automation and AI policy that won strong approval from a community that agrees on almost nothing. On Friday the Nixpkgs core team announced it had disbanded.

The statement is unusually precise about why, and the reason is not volume. The team held a constitutional mandate — project direction, decision-making, and the creation and management of teams, all scoped to Nixpkgs. What it never got was the ability to use that mandate without the Steering Committee above it reopening the question. Work arrived with, in the team's phrase, "desired outcomes already attached." Issues sitting entirely inside the delegated area were taken over without the relevant teams involved. The team names the residue plainly: uncertainty about whether it was trusted to decide anything on its own.

That is a structural claim, not a complaint about personalities, and the team goes out of its way to say so — no single member could have fixed it, several impressed them, the role above is genuinely demanding. The defect is in the arrangement. A delegated body that must re-establish its scope every time it acts is not doing governance. It is doing procurement of permission, weekly, for free.

Authority you have to keep re-asking for isn't delegation. It's review with extra steps.

One number turns this from a sad story into a finding. When the team called for new members, one person applied. Nixpkgs is one of the largest package collections in software, with thousands of contributors and no shortage of people who care enough to argue about it in public. The pipeline for the governance seat that actually holds the thing together produced a single candidate.

The obvious reading is burnout, and the replies on the thread take it that way. So does the team, which says stepping down was necessary for their health. Concede all of it — burnout is the proximate cause and it is real. But burnout implies a supply problem, and supply problems have a known fix: more hands, better rotation, shorter terms. The one applicant is what rules that out. People were not declining because the work looked heavy. They were declining because they had watched for ten months what the job actually was, and it was carrying accountability for outcomes you are not reliably permitted to determine. Nobody volunteers twice for that, and most won't volunteer once.

The bitter part is that the model worked where it was allowed to run. High-trust consensus inside the core team produced an AI policy that people with opposing views signed off on — the exact class of problem the majority-vote layer above it had been deadlocked on. The team is not arguing that its approach failed. It is arguing that the approach was never given the room the constitution said it had.

Those matters now have no direct owner, with the Steering Committee as the backstop — which is where they formally sat the entire time. Ten months only made that legible. The cost of the demonstration was the two contributors willing to write it down on the way out, neither of whom is running for the committee.