schedule     alphabar_T   SNR_T       sigma_min  sigma_max
ddpm linear  4.0358e-05  4.0360e-05  0.0100     157.41
sd scaled    4.6601e-03  4.6819e-03  0.0292     14.61

t     chain mean  chain var   jump mean   jump var
100   +0.4248     2.5984      +0.4248     2.5969
500   +0.1260     1.1331      +0.1294     1.1407
1000  -0.0043     1.0046      +0.0063     1.0089
variance the formula predicts at t=1000: 1.0001

step   loss (a perfect guesser that ignores x scores 1.0)
500    0.2292
1000   0.2581
1500   0.2462
2000   0.2391

t     RMS(net eps - exact eps)
10    0.0769
100   0.0976
300   0.0608
600   0.0411
900   0.0380

Wasserstein-1 to the data  exact eps: 0.0344   trained net: 0.0743   two real samples: 0.0545
fraction in the right-hand mode (truth 0.700)  exact: 0.690   trained: 0.704
