First, is to update the unit vector built-in type so we don’t have to put a multiplier keyword.
The second one is that you can actually use this to create a cholesky correlation matrix but doing exp(x_last) of the last value and adding jacobian += x_last to the log target for each row of the lower Cholesky factor. I don’t think it’s faster than Updated cholesky corr parameterization testing but it’s neat because of the implied priors of sech and half-cauchy before standardization. Maybe these can be used to make some interesting priors?