### Problem Statement Currently, the action testing logic is rather ad hoc and mostly covered by executing the RL training pipeline. This is prone to missing edge cases. ### Proposed Solution Each action/pass should be tested individually. Ideally, we verify that they maintain circuit invariants.
Problem Statement
Currently, the action testing logic is rather ad hoc and mostly covered by executing the RL training pipeline. This is prone to missing edge cases.
Proposed Solution
Each action/pass should be tested individually. Ideally, we verify that they maintain circuit invariants.