Thank you for your input! I tried a different solutions and based on my limited knowledge LaFleur’s answer seemed to be the easiest and fastest solution to me.
I used the window breps as input for the analysis (382 trimmed surfaces), as output I got 791 points. So then I used the closest point component to get the index of the points that were interesting which seemed to align really well with the desired points. It was practically the offset distance that was the difference.
After that I used the LB Deconstruct matrix and matched it with the desired branch indexes. Then I got the result of every timestep for each point.
Another bonus is that I could use a flattened list as input geometry and get grafted output per point with the LB deconstruct matrix. I havn’t been able to do this before with the old components. This way it decreases the compution time drastically.
Thank you both again!