Two-Key Gating: A Safety Architecture for the Timing of Psychological Interventions

Digital platforms increasingly deliver psychological and relational interventions at population scale, and they typically decide when to deliver an intervention using engagement-optimisation logic, surfacing content when a user appears most receptive or most active. This paper argues that intervention timing is not a user-experience parameter but a safety property: an intervention that is accurate in content can be harmful if it is delivered before a person has the capacity to receive it. Drawing on attachment theory, the psychotherapy side-effects literature, systems theory, and causal inference, we formalise a two-key gating architecture for the safe timing of interventions in digital relationship-readiness systems. An intervention may be surfaced only when two independent conditions are jointly satisfied, epistemic readiness (the system holds sufficiently certain evidence to act) and developmental readiness (the person holds the prerequisite capacities to receive the intervention without destabilisation), and only when no higher-precedence safety condition forbids it. We situate the model within a prerequisite capability sequence and an asymmetric account of relational change in which capacity is built slowly but can be lost rapidly under acute dysregulation. We argue that the gating logic is durable independently of advances in machine perception, because it is constrained by the timescales of human development rather than by measurement technology, while the gating thresholds are empirically tunable. The architecture is presented as a theoretical and design contribution; an empirical evaluation is specified as a pre-registered study and is the explicit next step. We close with the implication that as automated emotional perception becomes commoditised, the scarce and safety-critical capability becomes knowing when an accurate intervention is safe to deliver.