A software architect is typically interested in the components that exist in the application. You should be looking for the things you need in your architecture description (component model, coupling decisions, etc).
Picking the components which have a lot of communication between them might be a point of interest. As far as I can understand from this question you already seem to have an outdated architectural description. Make comparison's with it to see which points have changed, does the new code introduce high coupling? Are the layer rules being violated by the new approach?
Points of interest can include:
- The class diagram
- Component model
- Tier model
- Layer model
- Deployment model
Although these might not all be accessible by the tool you can probably work towards them if you keep these things in mind. The process of working from the bottom (source code for example) to the top is called: "Reverse Engineering".
For further reading I would like to recommend these slides.
Update
Dynamic view focuses on monitoring the process and in doing so recovers SA information. As explained by this quote:
Static analysis aims at recovering the structure of a software system, while dynamic analysis focuses on its run time behaviour.
reference
But really, aren't they working towards the same goal? While monitoring the process is different from static analysis both work towards recovering the architecture. The amount of detail such methods should recover (which is likely to be your question) is dependant on the size of your project and the amount of time that is given for this task.
Are we talking about a massive project (complete bank system) or something smaller? It is really a choice between time when it comes to this. As dynamic analysis often takes quite a long time to complete you might ask yourself if you really need all the information the tool can obtain. For a medium sized project I would be interested in the following questions:
- Which components exist within the project?
- How are methods called?
- Which components have to much communication?
With just these three questions it will be possible to recover most of the Component model. In this case you would still be missing a lot of information for your Software Architecture for which I recommend static analysis to recover the class diagram for example/perhaps detect the layer model that is being used (and its rules). Combining the two of them might be a more time efficient solution. The deployment model should not be to hard to create either.