The breadth of decision latitude granted to an autonomous agent, ranging from narrow execution of pre-specified steps to discretionary action within broad goals; a graded property of a delegation arrangement rather than a discrete occurrence.
STATE (graded). The granted latitude is rated per delegation arrangement on a 1-5 ordinal scale, anchored 1 = agent may only execute explicitly enumerated steps, 5 = agent may choose means and sub-goals within a high-level objective without per-step approval. Scoring is from the configured permission set, not from observed outcomes.
Proposed measurement protocol (not yet empirically validated): Rater coding of the agent's permission configuration onto the 1-5 latitude scale; inter-rater reliability could be reported as ordinal ICC. Objective complement: mean number of distinct action types the agent may perform without seeking approval.
https://andreasehstandlicenseofclarityloc.github.io/augmanitai-periodic/#topic-agentic-ai