The sample size $n$ must be small enough to avoid overfitting.

The parameter space must be compact, and the true parameter must be in the interior of the parameter space.

The distribution of the data must be Cauchy.

The likelihood function must be strictly linear.