Philippe Rivière

And now forceWalls appeared in https://observablehq.com/d/21d2053b3bc85bce ; I think we're getting a bit closer with each step (as suits a force simulation).

> Not sure if it was intentional yes it was to show the cushion's "impact" ( can a cushion have an impact? :) )—but it's ugly.

here's a new iteration of forceWalls, with a simpler API : https://observablehq.com/d/66b419d51abee995

I've taken another route completely and came up with a solution using a Sliced Optimal Transport approach. Test forceTransport at https://observablehq.com/d/21d2053b3bc85bce The gist of the algorithm is: 1) sort all...

I've now found a way to properly transport to a disk 

You can definitely play with the algorithm and see what it does (and doesn't). I don't think it's ready at all for being a module.

To get a better intuition of the sliced OT algorithm, suppose we are on a unique dimension *x*. The points are ordered according to their *x*: `0-1----------2-3-4----5--6-------7` now we want...

In https://github.com/d3/d3-force/pull/167 forceRadial has an accessor for the center, which can help clustering around various central nodes. I'm not sure it totally covers the use cases you had in mind,...

Related to #128

PS: It would be useful to share the problematic graph as a block or observable notebook.