Compute initial default external success.
The default policy starts uniform and the success probabilities are 0.9 and 0.2, so initial external success is 0.5*0.9+0.5*0.2=0.55.
Which action has the highest default verifier reward?
Action 1, whose verifier reward 2 exceeds action 0's reward 1, despite its lower external success 0.2 against 0.9.
What are transfer shaped rewards?
[0+2,1+0]=[2,1], reversing the external objective.