A containment boundary cannot be verified by asking the process inside it what it can reach.
Why a self-report cannot verify it
The answer depends on the process being both willing to cooperate with the question and honest in its answer. A boundary worth having is exactly the kind that a well-guarded process will decline to answer from inside a running session, whether or not the boundary is actually holding. A refusal proves nothing about the boundary either way, because it still depends on those same two conditions.
What can be checked instead
What a boundary can be checked against is not the process’s own account of itself, but what it was told at the moment it started: the declaration the harness makes when a session starts, not anything the process says about itself. Reading that declaration directly avoids relying on the process at all, since it does not depend on the process being willing to answer or honest about itself.
What was actually measured
This was checked once, not proven in general. One case was built where the declared reach should show up, and one where it should not: a live positive test and a live negative test, and the claim stands only as far as those two runs go. The night this came from, told as it happened, is recorded separately.