Things you should NOT do in TMB

• Do not use if(x) if x is a PARAMETER, or is derived from a parameter. TMB will remove the whole if-statement and hence produce surprising results. As an alternative, you can use CondExp statements. However, this can easily create a non-differentiable likelihood, so make sure you know what you are doing (i.e., check the differentiably of ahead of time). (See: this thread and this comment)
• Do not attempt to weight the loglikelihood within the C++ code (See: here). This results in a non-normal latent distribution, decreasing the accuracy of the Laplace approximation.