The Maintenance Constraint: How Resource Boundaries Shape Cognitive Availability

Cognitive systems are resource-limited, but "resource limitation" is often invoked without distinguishing one-shot costs (forming a representation) from sustained costs (keeping it usable under noise). We argue that the availability of internal state features for control, integration, and report is constrained by their maintainability under finite budgets. As a technical anchor we cite a companion preprint deriving an operational maintenance inequality in explicit thermodynamic control models: incremental maintenance power has a non-arbitrary lower bound tied to dynamical fragility (in its corrected form, against efficient baselines, with an unconditional entropy-production floor beneath). This motivates an operational cut: a feasibility boundary separating maintainable from unmaintainable state features. We develop an auditable bridge argument (maintainability to stability to availability), propose a neutrality-friendly principle of maintenance-feasibility bias, and show how it can be incorporated into active inference (the Free-Energy Principle) as a maintenance penalty or constraint, aligning secondarily with Global Workspace accounts of access stability. We address objections and offer falsifiable predictions for synthetic agents and neuromorphic systems, plus an explicitly exploratory psychophysics subsection framed in terms of reportability and stability rather than phenomenology. We do not propose collapse mechanisms, do not derive the Born rule, and make no claims about phenomenological consciousness.