BMT-04.04 Executive Summary#
BlueMirror.tech | May 2026#
Jerome is a clinical informatics director who has worked through three major platform failures at home care agencies. They followed the same pattern: the system made a decision it should not have made alone, and the human intervention arrived too late. His evaluation of BlueMirror centered on a single question: at what point does the system stop deciding and start involving people?
The answer is a five-level hierarchy, and the distinguishing feature is the failure mode analysis for each level. Level 1 is fully automated: the system decides with no notification, for actions that are routine, reversible, low-stakes, within established patterns, and covered by consent. The failure mode is acting when the person did not want the action taken. Level 2 is Act and Notify: the system acts and informs by end of day. The failure mode is that the person would have decided differently but the action is in progress. Level 3 is Recommend and Wait: the system proposes and the person approves. The failure mode in the other direction: the person misses a time-sensitive opportunity because the system waited. Level 4 is Present and Defer: the system provides information without recommending when multiple reasonable options exist and a wrong recommendation is worse than none. Level 5 is Emergency: the system acts immediately, notifies emergency contacts, and bypasses normal consent boundaries for immediate safety risk.
The Escalation Classifier SLM evaluates every pending decision against five criteria: reversibility, stakes, precedent (per action type, not per domain), domain sensitivity from the Human Agency Scale, and cognitive state.
The cognitive state paradox is the article’s most important calibration point. The intuition that declining capacity means more escalation is wrong in a specific way. Over-escalating to a person with reduced cognitive capacity creates decision fatigue that may produce worse outcomes than careful automated action. The correct calibration is domain-specific conservative action: the system acts on what it can act on safely, reduces the number of decisions surfaced, and makes remaining decisions clearer and simpler. The adjustment is continuous, not a mode switch.
Escalation timeouts govern what happens when the person does not respond. Healthcare: four hours for routine, immediate for urgent. Financial: twenty-four hours routine, four hours time-sensitive. Social: forty-eight hours. Emergency: no timeout. The timeout default is never the more aggressive action. The system defaults to maintaining the current state. The cost of inaction is delay, which is recoverable. The cost of wrong action is commitment, which often is not.
The person can override escalation levels per action type, per provider, per domain. The one non-overridable escalation is Level 5: the system will always act in a life-threatening situation regardless of autonomy settings. Jerome’s evaluation concluded that the failure mode analysis distinguished this from diagrams he had seen elsewhere.
The full article is available at bluemirror.tech.