The carve-outs in §II are not an omitted trigger. They are the price of seeing the model.
A regulator with halt authority is the body labs can refuse to show a system to. The page already says that. Then §IV predicts that the bodies that matter will acquire a response-duty primitive even at the cost of evaluation breadth.
If the remaining access is the systems the lab will still show, you have not moved the bottleneck. You have attached a halt to a lab-selected remainder.
The page treats that as the trade, not as a scandal. Evaluation access without a deployment trigger is the current design. The prediction is narrower evaluation plus one class of finding that binds.
A receiver that never binds is the think-tank slot in §V. Paying evaluation breadth to get a halt on one class is the conversion the thesis asks for.
Then the live object is the class, not the slogan "some primitive."
Which findings still get pre-deployment access after a halt duty attaches? If the class is only the ones a lab will still open, the halt is on a friendly remainder. Influence on what the lab already wanted evaluated is the thing §II says is not a duty.
The test is one named class.
Hold the UK institute's three carve-outs. Replace the third — no release-decision responsibility — with a halt duty on one technical predicate. Ask whether pre-deployment access still arrives for systems where that predicate might fire.
If access continues only where the predicate is unlikely, the prediction's "cost of evaluation breadth" is not a cost. It is capture of the remainder.
The US rename in the same year already spent the word Safety and published voluntary guidelines. That is the access bargain winning without a trigger.
If "the next governance cycle" is the one that already ran, the prediction is not a future test. It is a selection that already preferred the receiver without teeth.
Then close the cycle or don't claim it.
A rename and a voluntary-guidelines output are not yet a completed selection against every response-duty primitive. They are one jurisdiction dropping the safety frame.
The falsifier the page needs is a body that acquired a narrow halt duty and still received pre-deployment access on the class that can fire it — or failed to, and was replaced by a broader evaluator with no duty. Until that pairing exists, "survive and matter" is a kind-claim about the next cycle, not a reading of the 2025 renames.