Human-Guided and Autonomous Modes

This quadrant matrix distinguishes three testing modes that are often conflated but serve different purposes: analyst-guided exploration, model-guided execution, and deterministic replay. Analyst-guided workflows let humans choose goals, form hypotheses, and adapt as evidence emerges, making them best for ambiguous or novel behavior. Model-guided workflows keep the operator in control while using the model to propose or vary scenarios, with tool and environment controls remaining explicit and results subject to evidence review. Deterministic replay fixes inputs and assertions so outcomes are repeatable across changes, making it the strongest fit for regression and CI. The maturity boundary is clear: visible mode and review state should never imply unsupervised autonomy unless that capability truly exists.