When Correction Should Enter: A Cross-Locus Framework for Timing Proportionate Intervention Under Uncertainty, Accumulating Burden, and Declining Reversibility

Systems are often corrected at one of two poorly calibrated moments: before the available evidence can support the burden of intervention, or after accumulating harm, path dependence, and loss of options have made a lighter correction ineffective. This paper develops a cross-locus framework for determining when a bounded and proportionate correction has become justified while meaningful reversibility remains. Its central construct is the correction-entry window: an action-specific interval during which the evidence is sufficient for a particular response, the expected informational value of continued waiting no longer justifies the burden and option loss produced by delay, and the proposed action remains capable of reaching the loci that sustain the problem. The framework rejects both a universal intervention threshold and a simple choice between immediate action and waiting for certainty. Observation, independent review, reversible constraint, targeted correction, broad suspension, and emergency action instead carry distinct evidentiary thresholds because they impose different burdens and preserve different degrees of reversibility. The proposed architecture jointly tracks signal sufficiency, burden accumulation, tail severity, practical reversibility, information value from waiting, intervention burden, observability, correction-authority coverage, evidence provenance, and propagation or archival persistence. It distinguishes the time at which a correction is authorized from the time at which it becomes causally effective across the relevant records, procedures, actors, and systems. It also introduces correction latency debt to describe the additional burden produced after a specified correction becomes adequately justified but before it becomes effective. A further safeguard separates antecedent evidence from evidence that has been reinterpreted or productively generated by an earlier intervention. This prevents a system from altering behavior or opportunity, treating the resulting state as independent confirmation, and then using that apparent confirmation to justify escalation. The paper develops a graduated intervention ladder, trigger families, escalation and de-escalation rules, expiry conditions, non-cases, falsifiers, cross-domain applications, a canonical twenty-field Correction-Entry Timing Ledger, and retrospective and prospective validation protocols. It further defines necessity closure, in which an earlier necessity or least-harm justification becomes ex post immunity from reassessment, and residual burden obligation, the continuing duty to monitor, mitigate, explain, compensate, restore, reassess, or reverse burdens left by an intervention, including one that was defensible when undertaken. The framework does not claim discovery of sequential decision analysis, quickest change detection, value-of-information theory, early-warning indicators, adaptive management, dynamic adaptive pathways, precaution, cumulative-impact assessment, event-triggered control, moral remainder, or contestability. Its proposed contribution is the operational integration of these distributed approaches into an action-specific, cross-locus correction-timing architecture. The closest integrated baseline identified is cumulative-effects adaptive management combined with Dynamic Adaptive Policy Pathways. The added architecture earns separate retention only where it changes intervention timing, action intensity, evidence classification, correction locus, propagation requirements, residual repair duties, or falsification conditions relative to that baseline. The framework remains conceptual and has not yet been independently validated. It is intended for research, comparative case analysis, and institutional self-audit rather than as a clinical, legal, engineering, safety, benefits, employment, or automated adverse-decision rule.