Because increasing the learn span increases the number of days used for the smoothing. If we had 10 year learn span level of smoothing all the time it would be too smooth for a 1 year learn span.
Interesting. Maybe set the smoothing window size to a fraction of “Days to simulate”, like 1/20? That way smoothing will have an effect for any value of “Days to simulate” (aside from very small values like 5 days).