this is how your sizes are determined in XYZ direction:
be much aware that changing the Graph Mapper domain from 0:1 to anything else (I advice against that) it’s like adding a *10 multiplier hidden in plain sight…
so I’d rather go for an explicit multiplication instead:
given that the definition is based on geometries of size 1, I think the best solution would be to scale non-uniform at the very end:
this is your original size=1 geometry, you see the scaled brother in green in the background:
The following is a variation I probably prefer to the above one, where stuff is just created the right size instead of being scaled, the size controls are W for box width, H for box height, and E for extrusion Domain
there’s one important thing to say, the way you have decided to apply the checkboard pattern is a bit too much destructive 
going from a data structure where branches indicate row (or columns) to a data structure where “whatever is there” should be done for a reason or for “overwhelming convenience”:
here I think the only reason why that is done that way, is to be able to apply the checkboard 1/0 pattern with list item + Serie, taking also greater advantage of your initial number slider being set to “odd numbers only” (probably for aesthetic reasons, but consider that for even numbers that sort of checkboard pattern won’t work anymore)
{personal note: if that way of creating the checkboard pattern was indicated in the video tutorial, I would probably consider a different channel of video tutorials
}
I believe a better and very basic method to do that would be to populate a list of True/False booleans as long as your initial number of branches, this will be your start point:
then you can just graft it, in such a way you end up with a single True/False for each branch:
and as last step you merge that with the inverted self, in such a way each branch will contain either a True followed by a False, or a False followed by a True:
apply the Cull Pattern and the job is done, still 875 items, but same data structure is still there:
for your second question, this page offers very nice and insightful picture with awesome details: Serpentine Pavilion 2016 by BIG such as:
where you can notice that L shapes in the inside of each box are as long as the box extrusion (a few mm less)
while outside connection elements shaped like a + are as long as the overlapping edges of neighbor boxes (same few mm less)
another important axpect is that the initial grid size we have set (400x500) would produce a result where boxes have common edges:
while you want the box size to be respected despite having a gap in-between which is equal to the thickness of the + connection elements:
the easiest way to do that is to create a new parameter that sets the thickness of the + connection element, and sum it up to the size of the cells:
this means that each cell will need to be offset inwards by half connection element thickness before being extruded, or if you want to see that from a less geometrical and more parametrical point of view, it means that the grid of planes on which construction planes are created should take into account that thickness, while the rectangles size should not (by doing this we avoid 875 -useless- offset operations using just two spaghetti wires
):
serpentine pavilion 2016, BIG_inno.gh (17.8 KB)
there’s a very nice thing from this page https://parametrichouse.com/serpentine-pavilion-2/ that you might want to take into account at a later stage, where each box thickness is determined in function of the amount of stress applied to it (anyway this does not interfere with the external size of each box because to get thickness they get extruded inwards):
what I believe is probably the most interesting part of this whole study is to understed how to calculate the length of each + connection element, and I believe if you go geometrical for that (measuring the length of the extruded box edges) then you are not leveraging at all the real beauty of data handling (hahaaa!!) but that’s for another episode because I really do need to play some LoL now 