From what I understood, both the estimation command with analytic_derivation and the identification command compute the asymptotic Hessian of the log-likelihood (in the ML framework). Also, the estimation command stores this analytic Hessian in model_mode (hh) and the identification command stores it in idehess_point (AHess).

When I run my .mod file, these two matrices differ quite a bit, so I guess I’m missing something.

If I may add a question, which of the two matrices - analytic or asymptotic Hessian of the log-likelihood function - should be used to compute the ML covariance matrix?
And what should I conclude about identification if the analytical Hessian is full rank, while the asymptotic Hessian is not?