Graph physics: nodes jitter endlessly, overlapping at scale

[JordanCoin]

Problem

The force-directed graph never settles — nodes keep bouncing and jittering indefinitely. With 100+ entities the simulation runs hot, labels overlap, and the graph is hard to read. The physics simulation runs every frame with no cooling/convergence.

What's happening

• simulate() runs on every requestAnimationFrame with constant repulsion/attraction
• No cooling factor — forces never decrease over time
• Nodes bounce off the viewport edges and ricochet back
• At 144 nodes / 378 edges (Clearview AI contract), it's visually chaotic
• Labels overlap badly when nodes are close

Fixes needed

Physics

• Add cooling — reduce force multiplier over time (alpha decay like d3-force). Start at 1.0, decay to 0 over ~300 frames. Reheat on drag/interaction.
• Increase min distance — nodes closer than 2x radius should get strong repulsion to prevent overlap
• Viewport-aware centering — auto-fit all nodes in view on initial load instead of random placement

Visual

• Label collision avoidance — don't render labels that would overlap. Show labels for selected node + neighbors, hide others until zoomed in.
• Adaptive label size — scale label font with zoom level. At zoom-out, hide labels entirely.
• Node size by degree — high-connection nodes should be larger (they're more important)

Performance

• Stop simulation when settled — if total velocity < threshold, stop calling simulate(). Resume on interaction.
• Spatial indexing — for 200+ nodes, O(n²) repulsion is slow. Consider Barnes-Hut approximation or quadtree.
• Canvas text caching — labels are expensive to render every frame

Context

45 nodes (Palantir) was fine. 144 nodes (Clearview AI) is at the edge. Larger investigations with 500+ entities will need these optimizations.