Before Delegation: Specifying the Decision Surface in Institutional and AI Governance

Governance is typically introduced after decisions, delegations and incentives have already been structured. This paper argues that the resulting reliance on control and enforcement is not a failure of rule design but a consequence of sequence: rules retrofitted onto a system whose underlying meaning is underspecified can compensate for ambiguity but cannot remove it. We develop pre-governing — the specification of intent, constraints and invariants prior to delegation, such that agents, human or artificial, can act autonomously without undermining institutional purpose — and formalise the object it acts on: the governed decision surface, the set of actions authorised to be chosen at the moment of decision. Governing that surface is distinguished from filtering: scoring, reviewing or rejecting options after they have been generated. The central claim is that filtering expands the surface it is meant to contain. A rejected option that remains available remains thinkable, and is eventually selected — so each round of ex post control enlarges the space of what can be chosen while appearing to narrow it. That mechanism yields an account of several familiar pathologies: why compliance regimes expand without producing trust, why risk management escalates without reducing fragility, why governing bodies lose effective control as the surface beneath them grows, and why alignment degrades under delegation. The contribution is conceptual and architectural. The paper supplies a criterion for when governance must be specified and what must be fixed before control and enforcement become necessary, operating at the level of meaning rather than behaviour, and drawing on institutional theory, systems design and work on semantic architecture. Empirical evaluation of the framework is left to subsequent work.