Thanks. n would be a problem in the sense that real n is 20 but y is inflated by the noise and if I tried to do something just with Binomial of the form pm.Binomial(“obs”, n, p, observed=y_with_noise), I wouldn’t be able to use n=20 because sometimes observed values would be larger than 20 leading to probably -inf value in log likelihood evaluation. One way around would be to set n to be max of y but then that would likely create a bias in the model coming from an outlier y value with a large noise. n could be anywhere between 1000 to 1 basically. So I kept the original 20 in the page as it is not too small but also probably not large enough for a reasonable normal approximation.