The Non-Scalar Viability Kernel and Adaptive Navigation in Open Fields

Control architectures for autonomous nodes are conventionally organised around a terminal objective function: a scalar U whose maximisation defines the top of a decision hierarchy. We propose an alternative organisation in which the primitive object is not an objective but a rolling, set-valued, multi-metric viability kernel Kₜ. The node maintains a vector of condition indicators V_N(t) = (v₁, …, vₘ), each measuring viability along one dimension (boundary integrity, entropy export, relational coupling, dynamical optionality), and acts so as to remain within a kernel built on fixed ontological floors bᵢᵐⁱⁿ, while aiming at a navigational safe set Sₜ defined by adaptive thresholds bᵢ(t). Arbitration among feasible actions proceeds by Pareto dominance, a critical-set restriction, and a context-indexed tie-break; the construction of any scalar aggregate U = Σᵢ wᵢvᵢ is prohibited by design. Thresholds are updated by a quantile-plus-margin rule over a rolling window, projected into fixed bounds, and remain open to external correction. We discuss conditions under which the kernel remains nonempty and the threshold dynamics remain bounded, and we formulate the resulting policy as a ten-step rolling update. The construction is presented as a formal architecture under explicit assumptions; the interpretation of its components as physical or informational quantities in interacting fields is offered as a hypothesis, not as a derived identity.