The worker Shipped
A single specialist reviewing a single changeset through a single specialty, in its own context, seeing nothing about who else is running.
The worker sees one lens, one changeset, and the specialty documents that define what it is looking for. It does not see the roster. It cannot know that a security specialist is also on this run, so it cannot decide that security is someone else's problem today. That blindness is the coverage guarantee.
It also cannot see why the code was written
A reviewer running in a fresh subagent context sees the diff and the criteria and nothing else — not the conversation that produced the change, not the constraint that made the odd choice the right one. That is a real limitation and it is also the source of the independence. Its known failure mode is the flip side: a reviewer prompted to find gaps will usually report some even when the work is sound, which is exactly why the worker does not get to say how confident it is.