kinda wonder what would happen if a spring graph/hyperassociative map were given particular land boundary constraints.

i abused @freemo 's algorithm by adding rules like instead of some nodes having enforced lengths they have partial penalty springs (it must be 'at least' 50 units from another node but greater distance is not constricted) and that can be kinda fun sometimes.

i suppose if one wanted to constrain all the bouncy shit to, say, the border of a municipality, you'd do some signed distance field nonsense and have the nodes get pressured away from the edges like this but then otherwise let them be.

the point of that is mostly weird shit like if you wanted to simulate zoning a whole bunch of houses and shops with certain dirty industry (ex. shove the tanner and blacksmith away they smell) you could do up a graph and let the clusters shove each other out.

i dunno. this probably wouldn't help all that much tbh. trying to ponder how one might use this to compute loose layouts for where a whole network of people would live (ex. doing up clusters of housing with zones/nodes for pubs which are constrained to walking distance of the houses they belong to)

there's other fun stuff i haven't tried yet. like using the graph to set up a progression of villages for some hypothetical JRPG and letting the graph bounce around to put prospective landmarks down but then have them 'roll downhill' to route around terrain and make the shape more irregular but still be able to drag around stuff. and then you can use wavefunction collapse to actually fill in the small details around the big ones.

anyway yeah nerd shit

@icedquinn This sounds like you might be hitting on the original intent of the algorithm. The algorithm was intended originally to distribute a process graph across a cluster of computers while producing minimal "bounce" across server boundaries. So unless I'm misunderstanding you it should work nicely.