The Stability-Compensation Paradox: A Theoretical Mechanism, Rival Explanations, and an Empirical Test Programme

This working paper proposes the stability-compensation paradox as a middle-range organisational mechanism. The starting observation is familiar: frontline actors often keep systems functioning by improvising, absorbing variability, working around failures, and contributing time or knowledge that formal plans do not contain. The proposed contribution is narrower than that observation. The mechanism is said to arise when successful local compensation does two things at once: it preserves visible output, thereby reducing the apparent urgency of structural change, and it consumes or displaces the same scarce people, time windows, knowledge, budgets, or decision capacity that would be needed to correct the generating condition. Under these conditions, the strength that rescues present performance may weaken the capacity that could make the rescue unnecessary in the future. The paper distinguishes this claim from first-order problem solving, workaround research, organisational slack, high-reliability and resilience engineering, Peter Thorvald's human centricity illusion, and Lindsay Findlay-King's key people paradox. The strongest novelty test is explicit: if pre-existing key-person concentration, workload, learning architecture, leadership, workaround ease, and general resource scarcity explain the same correction delay without a residual role for correction-relevant resource overlap, the independent causal status of the stability-compensation paradox should be reduced or abandoned. A preregistration-capable empirical programme is therefore specified for a first test in residential social-care organisations. The primary discriminating contrast compares compensation hours supplied by roles that are independently required for structural correction with otherwise comparable compensation hours supplied by a separate role pool. The main expectation is that, conditional on total compensation and pre-exposure rival factors, correction-relevant compensation should be associated with a lower hazard of structural cause correction. Causal language is explicitly conditional on longitudinal identification, positivity, common support, and consistent g-method estimates. The paper also develops a normative extension termed double dependency: a problematic form of dependence can arise when an organisation relies on a person's continuing compensation while that person cannot safely reduce the extra role without risking voice, material security, belonging, or legitimate standing. This extension is not included in the primary empirical test and is presented as a separate research question. The paper ends at an epistemic boundary: the next legitimate step is independent scientific criticism and testing rather than further authorial certainty.