I agree, however, I would argue that 1 needs to be an independent process, with an IPC mechanism rather than a library. This would allow for multiple systems to

I agree, however, I would argue that 1 needs to be an independent process, with an IPC mechanism rather than a library. This would allow for multiple systems to access the same research data - If my experience in research maps to the specific situation (not all research is the same), this means that several experiments can easily have a rock-solid data source, without needing to debug the library integration every time, and allowing for post-processing chains to build of each other.