In a misspecified statistical model where the true distribution of data is $F$ but we assume $f(x|\theta)$, the M-estimator $\hat{\theta}$ satisfies $\sum_{i=1}^n \psi(X_i, \hat{\theta}) = 0$. What structure defines the asymptotic variance of $\sqrt{n}(\hat{\theta} - \theta^*)$ where $\theta^*$ is the pseudo-true parameter?