A copy no longer half-lands a tree

Copying a subtree — from the Copy dialog, from "copy to home", or through an agent's or the CLI's copy command — used to report the number of nodes it managed to write, and that number said nothing about whether the copy was whole.

Two things could make it short. If part of the source was not visible to you — a branch someone else has not shared, or the gated part of an installed package — the copy simply did not see those nodes, carried the rest, and reported success. And if one write in the middle was refused, the copy stopped there but the nodes already in flight landed anyway, so the result could contain a file sitting under a folder that never arrived. Either way you were handed something that looked like a finished copy and was not. One node type reached a personal workspace this way without any of its source files, and then failed to build every time the portal restarted, for four days, complaining about symbols that were simply missing.

A copy now checks itself, before and after. Before it writes anything it establishes how big the source subtree really is — not just how much of it you can see — and if it cannot carry all of it, it refuses and writes nothing at all, telling you how many nodes would have been left behind rather than handing you a broken half. (It deliberately does not name those nodes: they are the ones you are not entitled to see.) It then writes parents before their children and stops descending at the first level that fails, so what does land is always a properly rooted tree, never a file under a missing folder. And it reports every node that did not land — the ones that were refused and the ones it therefore never attempted — instead of naming one and staying silent about the rest.

Copies that were already complete behave exactly as before, including the count they report.

This covers the copy described above. Moving a node, importing an archive and installing a package each run on their own machinery and are unchanged — a move already refuses rather than relocating part of a subtree, and the other two are separate pipelines.