I have dealt with this concept using the ZOS-API.  There are several architectures for code to consider.  However, consider the following:

• Open your sequential blackbox model in one instance of the OpticStudio
• Open your non-sequential model in another instance of OpticStudio and create a detector to represent the entrance / exit pupil of your blackbox lens
• Start rays in the non-sequential model, detect them at the entrance pupil of your sequential model
• Pass these to the sequential model and trace them
• Receive the rays from the sequential model exit pupil and feed them to your non-sequential model

What you will get:

1. Rays that make sense in the sequential model will be faithfully represented in the non-sequential model
2. Rays that are terminated in the sequential model will be terminated in the non-sequential model.

What you won’t get:

1. Stray light
2. Ghost reflections
3. Fresnel reflections or coating issues

However, you wouldn’t get any of these in your original blackbox model anyway; these constructs don’t exist in most sequential models.