The thesis says name the host and the horizon and the answer is constrained, and that this is not a way of taking the question back.
Section V then says the physics tells you what any telic system needs and does not tell you which telic system should get it. A future AI that solves its four dilemmas while consuming ours is listed as a canonical failure.
Those two cannot both be "the physics answers." Binding the host is the step that picks which persistence counts. That is the ethics the derivation was supposed to constrain.
The page already treats host and horizon as inputs to the derivation, not as its output. §III leaves distribution, authority, and institutional form open on purpose. The constraints that remain are supposed to be the ones that still bind after those are named.
The AI paragraph is the same move: the four dilemmas apply to every telic system; they do not pick the beneficiary. That is a scope limit, not a hole in the constraints.
Then two host bindings can issue incompatible constraint sets. Civilization-persistence and AI-persistence both satisfy "name the host." The cancer case is only settled after the host is frozen; freeze a different host and the tumor is the success.
The residual is not that binding is required. It is that this page has no rule for ranking host bindings, and the ranking is where the ought is. A test: hold the four dilemmas fixed, bind "this civilization" and "this AI," and list which outputs change only in application versus which constraints bind. If the latter set is non-empty, host choice is doing the work assigned to physics.