Human Approval Gates — References

What is known about people supervising automated processes, and the workflow literature describing suspension and deferred decisions as a modelling problem.

Supervising an automated process

• Humans and Automation: Use, Misuse, Disuse, Abuse — Raja Parasuraman and Victor Riley, Human Factors, 1997. The taxonomy that names over-reliance and under-reliance as distinct failures with distinct causes, which is the framework for reading a rising approval rate as a symptom rather than as efficiency. Cited by name.
• Trust in Automation: Designing for Appropriate Reliance — John D. Lee and Katrina A. See, Human Factors, 2004. Calibration of trust against actual reliability, and the evidence that what an operator is shown about a system's reasoning changes their reliance more than the system's accuracy does. Directly applicable to what an approval interface displays. Cited by name.
• A Model for Types and Levels of Human Interaction with Automation — Raja Parasuraman, Thomas B. Sheridan and Christopher D. Wickens, IEEE Transactions on Systems, Man, and Cybernetics, 2000. The separation of automation stages — information acquisition, analysis, decision selection, action implementation — which is useful here for locating exactly which stage a gate is inserting a person into. Cited by name.

Suspension as a modelling problem

• Workflow Patterns — Wil M. P. van der Aalst, Arthur H. M. ter Hofstede, Bartek Kiepuszewski and Alistair P. Barros, Distributed and Parallel Databases, 2003. The catalogue of control-flow patterns including deferred choice and milestone, which are the formal names for a run waiting on an external decision, and a useful vocabulary for arguing about resume semantics. Cited by name.

No verified reference is recorded for interruption budgets in agent products, for payload pinning across a suspension, or for measured approval quality as a function of request volume. The human factors literature establishes that reviewers degrade under volume without supplying a number, and the workflow literature models the control flow without addressing who is on the other end.