Well, this, indeed helped. But I don’t understand why. I expected that since I specified obs = pm.Normal('obs', y_est + sig... (note the +sig), that the sig will be part of learning. Isn’t that so?
Also, what is the conceptual difference between obs = pm.Normal('obs', y_est + sig, sig, observed=y) and your suggestion obs = pm.Normal('obs', y_est , sig, observed=y) (no +sig)?