ArgoBox
ArgoBox Infrastructure & Automation
user@argobox:~/journal/2026-02-01-the-three-hour-physics-session
$ cat entry.md

The 3-Hour Physics Session

[Sunday, February 1, 2026] MOOD: TRIUMPHANT ArgoOS (Gentoo)
○ NOT REVIEWED

The 3-Hour Physics Session

Date: February 1, 2026 Duration: ~3 hours Issue: Knowledge graph felt mechanical, not organic Root Cause: All edges were the same length

The Problem

For months, the ArgoBox knowledge graph worked but felt wrong.

Nodes settled too quickly. Clusters were uniform. It had a mechanical quality that Obsidian’s graph never has.

I’d tweaked centerForce, repelForce, damping… nothing fixed the underlying deadness.

Today I decided to figure it out once and for all.

Hour 1: Research

Opened Obsidian. Stared at my vault’s graph. Actually looked at it.

What I noticed:

  • Tags form central hubs
  • Posts radiate outward like spokes
  • Hubs are far apart from each other
  • The whole thing has a “sea urchin” or “dandelion” shape

My graph:

  • Everything clumped together
  • No clear hub/spoke pattern
  • Uniform spacing everywhere

Hypothesis: Obsidian uses variable edge lengths.

Hour 2: Implementation

The key insight: not all connections are equal.

Spines - Connections to leaf nodes (low degree)

  • Should be short
  • Should be stiff
  • Creates tight satellite orbits around hubs

Bridges - Connections between hubs (high degree)

  • Should be long
  • Should be loose
  • Creates separation between clusters
let isLeafConnection = sourceDegree <= 2 || targetDegree <= 2;
let length = isLeafConnection ? spineLength : bridgeLength;
let weight = isLeafConnection ? 1.5 : 0.1;

First test: spineLength=10, bridgeLength=150.

The graph reorganized. Hubs formed. Satellites radiated outward.

It looked… right.

Hour 3: Refinement

Added interactive sliders:

  • Spine Length (1-50, default 10)
  • Bridge Length (50-400, default 150)
  • Node Scale (0.3-3.0, default 1.0)
  • Label Threshold (1-15, default 3)

Found optimal values:

Spine Length: 3-5
Bridge Length: 200-300
Node Scale: 0.5-0.7
Label Threshold: 8-10
Repel Force: 4000-6000
Center Force: 0.05

Also fixed text clutter:

  • Posts: show label only if node > 4px
  • Tags: show label only if degree > 5

And added bounding box constraint:

  • Prevents nodes from flying to infinity
  • Allows near-zero gravity without explosion

The Result

Before: Static, mechanical, dead on arrival

After: Organic, floating, actually feels like Obsidian

The difference is night and day. Same underlying library (Cytoscape.js), completely different feel.

Issues Fixed Along the Way

Graph disappearing after load

  • Cause: Zero gravity + high repulsion = nodes at infinity
  • Fix: Bounding box constraint

No visible change when toggling Obsidian mode

  • Cause: Layout not re-running after edge data update
  • Fix: Call setLayout('cose') after modifying edges

Physics panel not scrollable

  • Cause: Missing overflow CSS
  • Fix: max-height: calc(100vh - 120px); overflow-y: auto;

Files Modified

FileChanges
KnowledgeGraph.astroSpine/bridge logic, new sliders, bounding box
CSSPhysics panel scrolling, constellation popup

Lessons

Variable edge lengths are the secret. I could have saved months if I’d analyzed Obsidian’s graph more carefully.

Interactive tuning is essential. Without the sliders, I would have spent 10x longer finding good values.

Bounding boxes solve the gravity problem. I thought I needed to tune gravity perfectly. Nope, just contain the nodes.

Next Steps

  • Add more tag hierarchy relationships
  • Consider cluster coloring by parent tag
  • Save slider values to localStorage
  • Extract this as the Tendril open-source library

Victory

At some point during hour 3, I just sat back and watched the graph float.

Posts drifted gently around their tag hubs. Hubs maintained respectful distance from each other. The whole thing breathed.

That was the moment I knew it was done.

Three hours. Variable edge lengths. That’s all it took.