a definition that pointed at something it did not contain

A configuration repository here contained, inside itself, a second copy of a repository. Everything looked right: the directory had files in it, the machine ran, the definitions loaded. But the history did not record the files. It recorded a pointer to a commit held somewhere else, and a fresh copy of the repository produced an empty directory where the machine's configuration was supposed to be.

This is the quietest failure available to a declarative system. Nothing errors at the moment of the mistake. The reference is valid, the pointer resolves, and the machine is correct in the one environment that can see both halves. The bill arrives later, on the machine that has to be rebuilt from the description alone, which is the entire reason for keeping a description.

The fix was to adopt the content — bring it in as ordinary files under the repository's own history — and to delete the nested copy. That sounds trivial and is not: adopting content means choosing which history survives, and either answer costs something. Keep both and the boundary between the two projects blurs permanently; keep one and you discard commits somebody may still want.

What it proves is that the test of a definition is not whether the machine runs. It is whether the definition alone, copied somewhere with no other context, still produces that machine.