Signal Integrity

Signal Integrity is the first pillar of the JB-DOF™ framework. It examines how AI outputs are received, displayed, and contextualized for human decision-makers, and whether the information that reaches the human is complete, accurate, and presented in a form that supports informed decision-making.

Signal Integrity documentation captures the full information chain from AI output to human receipt:

• Output completeness. Does the human see the full AI output (including confidence intervals, qualifying factors, data limitations, and contextual information) or a reduced version? Many systems present a single score or color code that strips the nuance the underlying model was designed to communicate.
• Interface design. How does the user interface present AI outputs? Are risk scores given visual prominence that implies certainty? Are qualifying statements buried in secondary screens? Does the interface design encourage engagement or rubber-stamping?
• Information timing. When does the AI output reach the decision-maker relative to the decision point? Is it presented with adequate time for review, or delivered at a point in the workflow where meaningful engagement is impractical?
• Contextual framing. Is the AI output presented alongside relevant case information that enables informed judgment, or is it isolated in a way that makes the score or recommendation appear authoritative and self-sufficient?
• Selective presentation. Are some AI outputs shown while others are suppressed? Are confidence levels displayed for high scores but hidden for low ones? Is the presentation of information systematically shaped in ways the decision-maker may not recognize?

Signal integrity failures take many forms in justice settings:

• A risk assessment tool produces a score with a confidence interval of ±15%, but the interface displays only the point estimate , making a probabilistic output appear deterministic.
• An electronic monitoring system generates an alert with qualifying context (GPS drift, known dead zone, equipment malfunction history), but the alert displayed to the officer contains only “violation detected.”
• A case management platform recommends a supervision level but presents it as a color-coded badge (red/yellow/green) that reduces a complex assessment to a traffic light, inviting snap judgment rather than informed review.
• An AI system produces both a risk score and a set of mitigating factors, but the interface displays the score prominently and buries the mitigating factors in a secondary tab that few officers access.

In each case, the AI system may function exactly as designed. The failure is not in the algorithm. It is in the signal path between the algorithm and the human.