Framework orientation
HAOP is a work-system design framework for analyzing how humans, AI systems, and organizations jointly produce performance, control, safety, and reliability. It extends the system orientation of Human and Organizational Performance to work in which AI materially selects, transforms, routes, suppresses, frames, recommends, approves, or acts.
The organization is treated as a performer because it allocates functions and authors the staffing, evidence, interfaces, throughput, incentives, permissions, escalation routes, authority, and acceptable tradeoffs under which human and AI performers act.
The human adapts, the AI optimizes, and the organization contains and governs.
This is compressed functional language. It does not imply consciousness, intention, personhood, or moral agency in AI, and it does not make the three performers morally equivalent.
What Revision 5.0 integrates
- The three-performer architecture, including the functional performer threshold and moral asymmetry.
- True Function and Illusory Function, with the nine-question True Function Test.
- The five-hazard register: physical interaction and machine agency; psychosocial conditions; human performance and verification; recursive information degradation; and fragmented control and concentrated accountability.
- Accountability by Control, including consequence, ownership, and design accountability.
- Human-in-the-Design, Verification Requirement, Verification Capacity, Verification Overrun, Verification Gates, Verification Windows, and Workflow Profiles.
- Grounding and anchors: source, state, dynamics, and work-as-done grounding; Anchor Access; and false grounding.
- Workflow design obligations: Provision, Preservation, and Placement, including friction, slack, choke points, and protected constraints.
- The HAOP application of ARECC, translated into a ten-step work-design method.
- The human throughline, edge discipline, Imposed Offloading, pause controls, and the deposited purpose of the AI Deployment Safety Data Sheet.
Five baseline hazard classes
- Physical interaction and machine agency
- Psychosocial conditions
- Human performance and verification
- Recursive information degradation
- Fragmented control and concentrated accountability
A hazard is identified prospectively as a condition with the potential to cause harm, whether or not an event has occurred. The five classes are neither mutually exclusive nor exhaustive of AI risk.
Development status
Revision 5.0 is the controlling integrated public source. The framework remains under development and has not yet been empirically validated. Numerical VReq/VCap ratios, aggregate indices, quantitative gate thresholds, and formal operating-envelope models remain exploratory or deliberately deferred.