Seems like the shearing displaced the coordinates horizontally almost exactly the size of your population square, hence there's only a little row of instances present. I wonder if a lower lean factor in example 1 would cause a proportional shift of the population or whether it would 'snap' into the boundaries suddenly, like it does in example 3 when it's after a compute node.
If there's snapping behavior then perhaps there's a bug somewhere, but for now I tend to say you're doing a whole bunch of complex stuff before even computing anything and it's simply going haywire.
In a simple setup there are no issues with the T&S shader versus sampling populators, I just checked that.
All I can think of is that sometimes, when using large terrains with large altitude differences, the 3D preview draws a rectangle in the UI while the coordinates are actually somewhere else. So you think it's in the correct place, but it actually isn't.
The screenshots are impossible to interpret as there's so much going on.
Several surface layers as holders for displacement functions with certain displacement factors we don't know, neither displacement type, slope/altitude key etc. etc.
Not to mention the clutter of nodes which are also typical for my own works as well
but very hard to understand for someone without a similar brain