noise=independent +/-1; tie=first action; exact enumeration
A tables=4; independent A/B pairs=16
true_values=(0, 0) true_max=0
  single_mean=0.500000 double_mean=0.000000 copied_B_mean=0.500000
  single_bias=0.500000 double_bias=0.000000
true_values=(0, 1) true_max=1
  single_mean=1.250000 double_mean=0.750000 copied_B_mean=1.250000
  single_bias=0.250000 double_bias=-0.250000
equal_values_MSE: single=1.000000 double=1.000000
PASS 10 checks; estimator experiment only, no RL training
