Abstract
Apparently agentic outcomes can support multiple causal organizations and therefore underdetermine focal causal attribution. We present a stage-resolved causal-attribution protocol that separates candidate participation, selection provenance, and realization provenance. The protocol uses inherited structural-causal, mediation, transportability, causal-abstraction, representation-intervention, action-selection, and execution machinery rather than proposing a new causal formalism or action mechanism. Its distinctive requirement is evidential composition: a causal proposition established locally may support a stronger end-to-end attribution only when its prerequisites remain valid and any source-target use has an independently justified validity basis. The protocol preserves nonfocal preparation, bypass, and joint focal-nonfocal contribution rather than converting local focal effects into claims of unrestricted autonomy. It also identifies where stronger attribution must remain unresolved when a required proposition, handoff, or provenance contrast is not established.
In plain language
Seeing an action happen does not tell us, by itself, where the decisive causal work occurred. A represented option may never have influenced selection. A selected state may have been supplied partly by processes outside the focal subsystem. A selected option may fail during implementation, or a matching outcome may occur through a bypass route.
The paper therefore separates the attribution problem into distinct stages and asks what evidence licenses a claim at each one. Evidence from one stage can contribute to a stronger end-to-end claim only when the assumptions and causal relationships needed downstream remain valid.
Why it matters
Terms such as choice, control, intention, and agency often compress several different causal questions into one label. That can make an observed outcome look more informative than it actually is. A successful action does not automatically establish that a particular represented alternative participated causally, that the focal subsystem supplied the selected state, or that it implemented the final transition.
The stage-resolved approach makes the stopping point of the evidence explicit. If candidate participation is established but selection provenance is not, the attribution stops there. If a local focal effect is established but relevant nonfocal preparation or bypass has not been excluded, that contribution is preserved without promoting it into a claim of unrestricted autonomy.
The significance is therefore methodological rather than metaphysical. The protocol is designed to make causal attribution more precise by showing exactly which part of an apparently agentic transition has been identified, which parts remain unresolved, and what additional evidence would be required to move further.
Publication record
Submitted to Systems Research and Behavioral Science. No public preprint has yet been posted. A repository record or final journal DOI will be linked here when available.