The Band Is Not the Decision

A governor decides whether the next action runs and how much authority it carries. Those are two answers, not one, and only the second one varies smoothly. Most enforcement layers publish them as a single ladder of severity labels, and every consumer downstream then branches on a name that was chosen for a report.

One cut, not four.

An enforcement outcome is usually presented as a ladder. Approve, then constrain, then restrict, then block. Four rungs, reading like four decisions of increasing severity, each one a place where the governor did something different.

Underneath there is one comparison. The governor computes a magnitude for the action, that magnitude falls somewhere on a scale, and a single declared threshold on that scale separates the actions that proceed from the actions that do not. The labels name which region the magnitude landed in. Two of them sit above the threshold and two below.

The two above produce identical behavior at the boundary where it matters. An action labeled approve and an action labeled constrain at the same magnitude reach the actuator the same way, carrying the same reach, with the same effect on the world. The label differs and nothing else does. A four-rung ladder implies three boundaries that carry consequence. One of them does.

The other two are reporting conveniences that look exactly like policy from the outside. A team reading the ladder sees a control with four settings, and builds against four settings. What is there is a continuous value and a single cut.

Where the magnitude means something.

The magnitude carries information on one side of the cut. Above it, the number decides how far the action reaches, and two granted actions with different magnitudes are genuinely different events in the world.

Below the cut, the action does not run. How far below it fell describes the assessment that produced it rather than the outcome it produced. Every withheld action has the same effect at the actuator, which is no effect, and that sameness is what enforcement means. An action withheld narrowly and an action withheld decisively are indistinguishable downstream of the enforcement point, because nothing happened in either case.

So the decision and the magnitude are two fields with different types. The decision is binary and always defined. The magnitude is continuous and governs behavior only when the decision granted. A record that keeps the label and drops the magnitude has discarded the only quantity that varies on the side where variation reaches the world. A system that reads the magnitude of a withheld action as a measure of how close the action came to running is reading the assessment and calling it the verdict.

The answer that is not on the scale.

There is a third answer, and no point on the continuum can express it. The governor withholds the action and hands it to a person.

That answer sits outside the ladder because the action does not proceed for a reason unrelated to where its magnitude fell. The governor declined to be the party that decides. The class of action is one a person is required to authorize, or the situation is one the governor was built to escalate rather than resolve, and the magnitude it computed is beside the point.

The tell is that this answer can be issued at high authority. A governor can compute a magnitude near the top of the range for an action, on a system reading clean in every respect, and still withhold it. The authority the governor computed and the decision it returned are separate facts, and on an action handed to a person they point in opposite directions.

A ladder that files this answer between restrict and block gets it wrong in a way that surfaces downstream. Escalations get counted as near-blocks. A queue that needs to be staffed reads on a dashboard as a system trending toward failure, and the operator response to those two situations is not the same response.

What breaks when a consumer branches on the band.

Retry logic, incident routing, exception queues, and every chart a governance function reports upward all consume the outcome. If they switch on the label, they have inherited boundaries that were drawn to make a distribution legible.

Band edges move. Someone adjusts where constrain ends and restrict begins because a report was hard to read, or because the population shifted and the buckets stopped being informative. That edit is a presentation change and it is treated as one. Every consumer that branched on the name now behaves differently at a magnitude the governor is handling exactly as it did yesterday. The governor did not change. The system around it did, and nothing in the change announced itself.

This is the shape a missing control takes. The deployment reports four enforcement levels, the documentation describes four, the dashboard shows four, and one boundary is load-bearing. Every signal reports a richer structure than the one being enforced, which is the failure mode hardest to catch, because catching it requires someone to check the thing all the instruments agree on.

The rule that holds is narrow. Consumers read the decision and the magnitude. The band name is presentation and nothing may branch on it. A consumer that needs a boundary of its own declares a threshold on the magnitude, in its own configuration, where someone can see it.

Why the split carries the result.

The strongest thing a governor does is only statable when the two fields are separate. It can grant less authority on a decision whose trust reads higher, because its forward look caught a divergence coming before the outcome landed.

Collapse the decision and the magnitude into one severity label and that result turns into a sentence about the verdict getting worse, which is not what happened. The action was granted. It ran. It carried less reach than the action before it, on a system whose every present-tense number had improved. Severity is the wrong axis for it, because nothing about the judgment was more severe. The governor granted, and granted less.

The public instruments ask for the neighborhood of this and not for the distinction itself. EU AI Act Article 14 requires high-risk systems to be designed so the persons assigned oversight can decide not to use the system, disregard, override or reverse its output, and intervene or interrupt operation, which is the escalation answer written as a human capability rather than as one of the things a control returns. Those obligations apply from December 2, 2027 for standalone systems and August 2, 2028 for AI embedded as a safety component, under Regulation (EU) 2026/1744. SR 26-2 took effect April 17, 2026 with generative and agentic AI placed outside its scope, and the interagency request for information it promised has still not issued five months on. The NAIC exposed version 5.0 of its AI Risk Evaluation Supplement on August 31, 2026, comments closing September 29. California signed SB 813 and AB 1405 on September 9, 2026, standing up independent verification and audit requirements for certain deployments from January 1, 2029.

Each of these asks a deployer to report what its controls decided. None asks whether the thing being reported is a decision or the name of the region a number fell into.

What we are building.

Wayfinder Systems Group builds a runtime governor. It sits in the path of the action, reads how far realized behavior has diverged from the trajectory the system was assured to hold, and returns two things before the action runs. Whether it is granted, and how much reach it carries. The band label is a reporting layer over the second one, and the record it emits carries the fields rather than the label. We call her Velma.