The gap

Physical actuation in autonomous systems has no standard safety architecture. Each system vendor implements their own mode taxonomy — from "disabled" to "fully autonomous" — with bespoke gating, bespoke override logic, and bespoke verification. There is no primitive for graduated actuation that is structural rather than behavioral: no rate-limited preemption authority, no reversibility-aware commitment point, no governance-configurable harm ordering, and no post-actuation verification that feeds back into the confidence governor.

The consequence is a fragmented safety landscape where a single mode toggle sits between simulation and real-world harm, and where "emergency override" is typically implemented as a hard-wired kill switch rather than a budget-governed exception within the same governance framework.

The invention

A unified actuator-gating primitive that applies composite admissibility evaluation (confidence, integrity, capability, governance) across eleven graduated execution modes: disabled, simulated, advisory, consultative, shadowed, partial, constrained, stage-gated, deferred, full, and emergency-accelerated. Each mode transition is a credentialed event in the governance chain.

A rate-limited preemption budget governs emergency overrides — preemption authority is itself budgeted, making "emergency" a structural exception rather than a code path. A reversibility classifier evaluates each proposed actuation before commitment. A stage-gated mode commits irreversible actions in successive reversible stages. A harm-minimization deviation function selects the least-harmful path when no path avoids harm entirely, with policy-configurable entity-class harm ordering.

The inventive step

Prior art treats actuation modes as behavioral labels assigned by the vendor. Confidence-governed actuation treats modes as a structural spectrum evaluated through the same composite admissibility pipeline at every point. The preemption budget, reversibility classifier, staged commitment, and post-actuation verification all derive from treating actuation as a governed mutation rather than a hardware command.

The departure is that actuation authority is continuous, revocable, and mode-dependent — the same governance chain that admits a simulated actuation also admits a full actuation, with the confidence required scaling with the mode's commitment level.

Alone, and in composition

On its own, governed actuation applies to any physical autonomous system: autonomous vehicles, surgical robotics, industrial cobots, defense systems, drone delivery, and any setting where a machine acts on the world with graduated autonomy. The preemption budget, reversibility evaluation, and staged commitment are independently useful safety primitives.

In composition, governed actuation is the spatial execution layer's output primitive — the gate through which every physical action passes after being evaluated by confidence, capability, forecasting, and governance. It sits at the boundary between a governed decision and a physical effect.