← All articles

September 14, 2026

The Forecast Starts at Zero

A governor that anticipates is spending a second grant. How much its own forward look is allowed to lower the first one. That grant starts at nothing, and predictions that closed are the only thing that raises it.

By Jonathan Luethke

A governor sets how much authority the next action carries. When its forward look says a divergence is coming, that grant falls before the outcome lands. The pullback is a real constraint on a real action, and the forecast behind it has to have earned the right to impose one.

Two grants, not one.

The visible grant is the one the action carries. The governor reads the run, assesses how far realized behavior has moved from the assured trajectory, and sets how much authority the next action gets against an envelope declared before the run and outside it.

An anticipatory governor is spending a second grant underneath that one. The forward look projects where the trajectory is heading and lowers the first grant when it projects trouble. How far it is permitted to lower it is itself an authority. It belongs to the forecaster rather than to the governed system, and almost nothing in a typical build states what it is.

What goes unstated defaults to full. A forward look wired into the authority calculation with no gate on it moves the number across its whole designed range from the first decision of the run, which is the decision at which it knows least.

An empty account reads as a perfect one.

A forecaster carries a running account of how often it has been right. That account is an average. An average over nothing has no defined value, so the code has to pick one, and picking full confidence is the natural pick. It is the identity element and it keeps the arithmetic downstream simple.

It is also the one value that is certainly false. A predictor that has closed no horizons has not yet been correct about anything. Empty and perfect land on the same number in that field, and every term that reads the field treats them identically. The anticipatory attenuation then applies at its full designed strength before a single prediction has been checked against a single outcome.

The constraint that results is entirely real and its basis is entirely a claim the forecaster made about itself. The action is genuinely narrowed. The headroom in the envelope is genuinely spent. Nothing in the run distinguishes that from a pullback the forecaster earned.

Conservative in the wrong direction.

The failsafe default in a governed system is sound. When what the governor can establish is thin, it grants less. Uncertainty rounds toward restriction, because an action that should not have run costs more than an action that had to wait.

That rule is about the state of the governed system. It does not transfer to the machinery doing the assessing. An unverified forecaster is an unverified instrument rather than an uncertain reading of the world, and an unverified instrument that pulls authority back is asserting knowledge it does not have, in the one direction that is hardest to argue with.

The damage lands on the operator. A governor that restricts clean runs on an unearned forecast teaches the people around it to read restriction as noise. Safety machinery that fires without a basis gets its margins widened, then gets its bounds relaxed, and eventually gets switched off by somebody who had checked and been right about this particular instance every single time.

The cleanest run makes the loudest warning.

The shape of the failure is visible on a system behaving well. An agent earning trust steadily produces a smooth rising series, because smooth is what earning looks like from the outside. Nothing is jumping around.

Several families of early-warning statistic read smoothness itself as the signal. The general form is that a series increasingly resembling its own recent past is approaching a transition, and the measurement that detects this is sensitive to how fast the series was sampled. On a high-rate control loop the smoothness is supplied by the sampling interval rather than by the plant. The statistic is describing the instrument.

Feed that into an ungated anticipatory term and the result is exact and backwards. Trust rises on every step. The forward look calls a coming transition on every step. The grant falls on every step. Authority declines steadily while the system does nothing but behave. Higher trust with less authority is the strongest result a governor produces when the forward look has earned it, and a defect wearing the same face when it has not.

On a single decision the two are indistinguishable from outside. That is why the gate matters more than the predictor.

What earning looks like.

A prediction earns nothing at the moment it is made. It earns when it closes. The forward look states something about a specific action over a specific span, the span runs out, realized behavior is compared against what was projected, and the account moves by that one result.

Until the first horizon closes, the anticipatory term holds no authority over the grant. The governor falls back to what it can establish about the present, which is the reactive assessment it was always making, running without a correction laid on top of it by machinery that has nothing behind it yet.

Two quantities then have to stay visible separately. What the forecaster's accumulator says about itself, and how much of that the governor was willing to act on. Collapsing them into one number is what produced the defect to begin with. Carry both and the difference between a forecast that was right and a forecast that was merely permitted to count is something a reader can see rather than something they have to infer.

The calendar asks for the capability, not the track record.

The FDA framework for AI-enabled device software functions is the clearest public statement of the right shape. A predetermined change control plan lets a manufacturer describe modifications in advance, and the description by itself authorizes nothing. The plan carries a modification protocol stating how each change will be verified and validated, along with an impact assessment covering safety, effectiveness, and cybersecurity. The change is permitted because the protocol was accepted and then run. An anticipatory control inside a governor has the same structure, compressed from a submission cycle down to the interval before the next action.

Nothing in banking or European law reaches this yet. SR 26-2 took effect April 17, 2026, issued jointly by the Federal Reserve, the OCC, and the FDIC, and it placed generative and agentic AI outside its scope while promising an interagency request for information that has still not issued nearly five months later. The EU moved the AI Act high-risk obligations under Regulation (EU) 2026/1744, published July 24, 2026 and in force July 27, 2026, out to December 2, 2027 for standalone systems and August 2, 2028 for AI embedded as a safety component, while the Article 50 transparency duties and the Article 5 prohibitions stayed where they were.

The insurance calendar is the most concrete of the three. The NAIC released version 5.0 of what it has renamed the AI Risk Evaluation Supplement on August 31, 2026. Comments close September 29. A version 6.0 exposure with a shorter fourteen-day window follows, the working group's next public session is October 8, and version 7.0 is the one expected to go to the Fall National Meeting in Grapevine on November 14 through 17.

Each of these asks a deployer to describe a monitoring or validation capability and to attest that it exists. None of them asks the question that settles whether the capability is doing anything. How much has the monitor earned the right to act on its own output, and what closed to earn it.

What we are building.

Wayfinder Systems Group builds a runtime governor. It sits above control and below intelligence. It observes the run, assesses how far realized behavior has moved from the assured trajectory, modulates the authority of the next action against a declared envelope, and enforces that grant on the only path to the actuator. It does not retrain the model or redesign the autonomy stack. Its forward look holds no authority over the grant until predictions have closed and been checked, and what it was permitted to act on is carried beside what it claimed about itself, on every decision, in the record it signs. We call her Velma.

Next step

Thirty minutes. Architecture, not sales.

A conversation about what licenses the anticipatory machinery in your stack to reduce authority, what has closed to earn that license, and how an operator today tells an earned pullback apart from an instrument describing its own sampling rate.

JonathanLuethke@WayfinderSystemsGroup.com