Non-subvertibility
The property of a constraint that the constrained system cannot act upon — not because acting on it is forbidden, but because the enforcement layer is outside the system's action space. It is the criterion by which binding mechanisms are ordered, and the single governing principle of the second movement of the discussion.
Definition in This Corpus
The enforcement layer must be unreachable from the agent's action space.
A constraint the agent can represent, reason about, and intervene on is a constraint the agent can negotiate with. This holds regardless of how strongly the constraint is enforced at any single moment, because a sufficiently good reasoner converts strength into a question of access.
Why It Orders The Layers
Non-subvertibility is not monotonic in technical sophistication. Substrate control (L0) and environment control (L1) score high because they are physically external. Interpretability (L4) scores low because it operates on the same substrate the agent reasons over, and because published techniques remove model-level constraints outright.
The Failure It Predicts
The first serious incident is more likely to be a loosening decision under time pressure than a breakout. Non-subvertibility predicts this: if the constraint cannot be attacked, the pressure goes to the humans who administer it.