Writing · August 2026 · from the working notes, July 2026
Existence is a leak.
In n0de’s formal access-control model, we formalized the naive shared-namespace leak as the motivating counterexample. We hold the fix, polyinstantiation, by a machine-checked proof. Names are scoped so that creation attempts from outside a compartment learn nothing from the outcome. The principle behind that work is that access control cannot stop at guarding contents: existence itself is a leak.
The cheapest version of the attack is a naming collision. An attacker tries to create an object with a guessed name. If the system returns "already exists," it confirms the presence of an object the user was never cleared to see, through an error message that touched no data. The pattern generalizes to search results that count what they will not show, timing differences between "denied" and "not found," and ID sequences with conspicuous gaps. In a multi-level system, each is a covert channel from compartments outside the caller's clearance.
These problems live in the corners. No feature test writes assertions about the error message returned when creating a duplicate name in an unseen compartment. The channel sits in code paths test suites treat as uninteresting failure cases.
Checking the access-control model
Non-interference gives the general form of the guarantee: activity inside a compartment provably cannot change what an outside observer sees, including errors, collisions, and counts.
These properties apply to the model and its assumptions; they do not establish the security of a complete deployed system.
This property justifies the cost of proof. It is quantified over all names and all traces in the model, precisely where testing thins out and an assertion like "cannot" must mean cannot. It matches the habit we argue for: match the strength of the guarantee to the cost of being wrong. For most of a system, tests earn their keep. For the channel that leaks the existence of what you are sworn to protect, we want a theorem.

