I can pin any element to anchor it at any point in the graph (say, in one corner) to keep it steady—and to keep its associated links in tow—while the rest moves around. This enables me to isolate one cluster of related ideas and keep it easy to access while I explore other parts of the graph. It would be fine if the whole thing dissolves and comes unpinned after I close the graph view, as long as I can keep that cluster in view while I’m using the graph.
It’s fun to drag nodes around in graph view, but since the graph snaps back into its spherical shape, it’s hard to extract much information from doing so. What I’d really love is to be able to manually move nodes around in the graph-space, and pull apart strands in a way that stays visible. Doing this one could, for example, end up with a tree-like shape for a particular cluster of notes.
It could also be nice to have a feature that I called “starching”, in a previous feature request concerning graph relaxation/arrangement, inspired from a property of the After Effects puppet tool in name only. In graph view, I would appreciate the ability to rearrange but know that too much freedom comes with a mess and loss of speed.
To combat that, I envision a mode you turn on and it allows you some freedom to rearrange while respecting the tension of the graph. It would understand you mean business when you persistently and repeatedly move and release and move and release the same node towards a desired position relative to others.
At the point that you get some semblance of the arrangement you desire, a weight could be dialed in for the important anchors. The values could only be important in their amount relative to others. And all of this would be done with the understanding that when you exit this mode, your nodes would automatically shift a bit the way we are already accustomed to. I don’t suspect you would be able to or even want to be multi tasking in this mode, since it would not respond quite as nicely with spacing and interaction as the graph view currently does. It would be optimized, so as to be effective for its purpose.
After leaving arrangement mode, we would continue to build out and connect structures, the positions of the weighted nodes relative to each other would be respected on a sliding scale. Although I am sure this sounds complicated, the general idea would be to have a balance between the graph allowing manual arrangement and it doing the automatic nitty gritty spacing and readability arranging for you.
Of course, a version of this is already in place somewhat, but could be further developed, especially if the weighted anchor nodes had a dedicated linking system (independent, invisible, and unrelated to standard Obsidian note links/connections) with options for both forward and inverse kinematics to provide flexibility and great hierarchical control in deciding how you want your graph arranged, for the most part. I am sure I will be back to edit this for clarity as this has been a good bit of rambling off the top.
I don’t actually think all of these specific behaviors and sub-features could or should all necessarily be implemented. I just wanted to throw some ideas out there in case they maybe inspire ideas or discussion that increase likelihood of custom arrangements. Also, sorry for the length. Thank you.
As time has passed since I posted this previous idea, I now have a revised thought of a new twist on how the graph might work.
First, take the current behavior of the graph and apply it to all notes constrained to existing within the area of the path of a circle (nothing allowed inside or outside.
Give it a bit of time to work itself out knowing that it will stay quite tangled and bunched. The initial scattering of nodes along circle path could be quite randomized starting with the most connected. This would help keep the web wrapped around circle and prevent collapsing (think pelt mapping if you know 3ds Max).
Anyways, I am thinking that the process and strategy of slowly letting the nodes leave the circle path could nudge some insight out of your vault. Likely the most well connected should stay put the longest (kind of the opposite of current behavior). But there could be options and variables for how it goes, so this and many other effectors could be reversed and dialed.
These kind of experimental graph manipulation features in coordination with pinning nodes could be really fun and helpful for engineering (or reverse engineering) practical and meaningful layouts.