mean reward of sampled responses (average over 8 seeds)
 step  leave-one-out          lam=0        lam=0.9       lam=0.95          lam=1
    1          0.169          0.169          0.169          0.169          0.169
   25          0.682          0.178          0.356          0.376          0.395
   50          0.884          0.208          0.793          0.847          0.876
  100          0.890          0.250          0.920          0.920          0.936
  200          0.890          0.085          0.937          0.937          0.937
  300          0.891          0.173          0.937          0.937          0.937

per-seed reward over the last 50 steps, and seeds above 0.99
 leave-one-out: [0.75  0.875 0.75  0.75  1.    1.    1.    1.   ]   4 of 8
         lam=0: [0.02  0.454 0.126 0.025 0.46  0.144 0.    0.001]   0 of 8
       lam=0.9: [0.75  0.75  0.999 1.    1.    1.    1.    1.   ]   6 of 8
      lam=0.95: [0.75 0.75 1.   1.   1.   1.   1.   1.  ]   6 of 8
         lam=1: [0.75  0.75  0.999 1.    0.999 1.    1.    1.   ]   6 of 8

critic error at the last position, RMSE of V_T against the realized reward R (mean over seeds;
explained variance is not reported: Var(R) tends to 0 once the policy becomes deterministic)
         lam=0: step 25 0.187   step 100 0.182   step 300 0.077
       lam=0.9: step 25 0.142   step 100 0.013   step 300 0.004
      lam=0.95: step 25 0.144   step 100 0.012   step 300 0.003
         lam=1: step 25 0.146   step 100 0.013   step 300 0.004
