I think a Taylor approximation extension can be helpful here. The estimators may include true moments and sample estimates?

Maybe the approximation error for the error bars was small enough , when you used a capital T. :)

It also looks to be the same across all the different x-values?