10.1
Demonstration sets the task
A good expert session contains more than an answer. It reveals the initial state, the working constraints, the checkpoints, and the shape of a successful deliverable.
Those elements can define a realistic practice environment.
10.2
Branch before the answer
The most useful practice points often sit just before an important decision. A restore point can return the environment to that moment without exposing the rest of the expert path.
Many policies can then attempt the same decision under controlled conditions.
10.3
Compare outcomes
Because the starting state is held constant, differences in the final artifact become evidence about the policy rather than noise from the environment.
The expert result can act as a reference without forcing the agent to reproduce every gesture.
10.4
Feed the next run
Verification identifies what succeeded, what failed, and where another example might help. That information can select the next restore point or the next expert session to capture.
The loop closes when demonstrations and practice improve one another.