Loading…
gathering yarn balls…
Sources: dream graph (623 sources) Generated: 2026-10-16T02:38:38Z

DREAM CONSOLIDATION

-- nodes -- edges -- density
◀
HUB EKG
Interval: --s Rate: --/min Pulses: 0
DREAM::GRAPH -- Help ×
Graph Navigation
Drag Rotate the camera around the 3D graph. The graph is rendered in a 3D space using Three.js with a perspective camera. Dragging left and right rotates the view horizontally, while dragging up and down tilts the view.
Scroll / Pinch Zoom in and out. The camera moves closer to or further from the center point. Use scroll wheel on desktop or pinch-to-zoom on touch devices.
Right-Drag / Two-Finger Pan the camera. This moves the orbit target point, allowing you to recenter the view on different parts of the graph.
Node Interaction
Hover Shows a tooltip with node summary including label, category, importance, phase, timestamp, and description. The tooltip follows the cursor and disappears when the cursor leaves the node.
Double-Click Selects a node, zooms the camera to it, opens the property panel on the left side, and triggers a pulse trace that cascades through all connected nodes and edges. The pulse propagates outward from the selected node, illuminating edges and nodes in sequence, then repeats after a short delay.
Click empty space Deselects the current node, closes the property panel, and stops the pulse trace. This returns the graph to its default state with no selection highlight.
Property Panel The left panel shows detailed information about the selected node including all metadata fields and a list of connections with edge weights. Use the collapse button on the right edge of the panel to hide it without deselecting the node.
Fly Navigation
W Move camera forward in the direction it is facing. The camera moves along its local forward axis.
S Move camera backward, opposite of forward direction.
A Strafe camera left, perpendicular to the forward direction.
D Strafe camera right, perpendicular to the forward direction.
Z Move camera straight up along the world Y axis. Useful for getting a top-down perspective.
X Move camera straight down along the world Y axis.
Q Yaw orbit left. Rotates the camera around the current orbit target point, keeping the same distance.
E Yaw orbit right. Rotates the camera around the current orbit target point in the opposite direction.
Shift + WASD Fast movement mode. When Shift is held, the camera accelerates to a maximum speed of 12 units per frame, much faster than normal movement.
Shift + Z/X Fast vertical movement. Moves the camera up or down at maximum speed when Shift is held.
Normal movement (without Shift) starts at a base speed of 2.4 and accelerates to 4.8 units per frame. The acceleration is gradual, giving smooth speed ramp-up.
Fly controls are disabled when the help modal is open, when typing in input fields, or when the Graph view is not active.
Camera Controls (top-left buttons)
+ Zoom in. Moves the camera 20% closer to the orbit target.
− Zoom out. Moves the camera 25% further from the orbit target.
⟳ Reset camera to default position at (0, 0, 400) looking at the origin.
◎ Top-down view. Positions the camera above the graph looking straight down.
▷ Side view. Positions the camera to the side of the graph at the same height as the origin.
Simulation Controls
❚❚ Pause / Resume the force-directed simulation. When paused, nodes stop moving but the graph remains interactive.
♻ Reset nodes to initial spherical positions and restart self-organization. This redistributes all nodes on a sphere and lets the force simulation reorganize them.
⇆ Cycle through close, wide, and cosmic node spacing layouts. Close mode packs nodes tightly with stronger springs. Wide mode spreads nodes apart with weaker centering. Cosmic mode uses a constellation-style force model with anti-gravity, dark-energy expansion, and hub dark matter to keep clusters separated.
The simulation arranges nodes like a cosmic web: mass-scaled anti-gravity pushes all nodes apart (more strongly for high-degree hubs), edge springs pull connected nodes together, dark energy drives Hubble-like outward expansion, and hub dark matter applies an outward force proportional to a node's net edge attraction to prevent collapse into a dense ball. The simulation automatically pauses when average kinetic energy drops below a threshold.
Views
Graph 3D interactive knowledge graph. This is the primary view showing nodes as colored spheres and edges as billboard lines connecting them. The graph uses a force-directed layout that self-organizes in real time.
Nodes Tabular view of all nodes with ID, label, category, importance score, phase, and timestamp. Importance is color-coded: green for high (>=0.7), yellow for medium (>=0.4), red for low.
Edges Tabular view of all edges with source node ID, target node ID, weight, and relation type. Edge weights influence spring force strength in the simulation.
Raw CSV Decoded base64 CSV data. Shows the raw merged CSV content that was used to generate the graph. This is the same data that is base64-encoded in the page source.
Fullscreen
Fullscreen Button Toggle fullscreen mode for the graph container. In fullscreen, the graph fills the entire viewport. Press Escape to exit fullscreen. The fullscreen preference is saved to localStorage.
Data Format
The dream graph is generated from CSV files containing node and edge definitions. Multiple CSV files from different dream sessions are merged with global ID remapping to ensure unique node identifiers across sources.
Node fields: id, label, category, importance (0-1), description, phase, timestamp.
Edge fields: source_id, target_id, weight (0-1), relation_type.
Node Colors by Category
Architecture -- Structural design decisions, system architecture, component relationships
Root Cause -- Identified root causes of bugs, issues, or behavioral problems
Lessons Learned -- Insights gained from debugging, development, or analysis sessions
Procedure -- Step-by-step processes, workflows, and operational procedures
Observation -- Noted behaviors, patterns, or phenomena observed during development
Edge Colors
Default edge color (cyan-blue). Edges glow brighter when traversed by the pulse trace.
Pulse-traversed edge color. When a pulse wave travels along an edge, it transitions from the base color to bright cyan.
Node Sizing
Node size is determined by the importance field. Base size is 3.0 units, scaled up by up to 4.0 additional units based on importance (0-1). Higher importance nodes appear larger in the graph.
1 / 4
Pulse Trace Mode

Trace mode iterates through every eligible node (degree ≥ 2) in the graph in random order, recording detailed pulse propagation telemetry for each. Nodes with 0–1 edges are fragment memories and are skipped because they cannot sustain pulse cascades. The simulation uses a biological neuron-inspired model with action potentials, refractory periods, charge accumulation, edge resistance, and per-hop voltage decay.

Each node is traced for 5 minutes at 60 FPS fixed-timestep sampling (18,000 frames per node) with deterministic default settings. A 5-second cooldown separates each node. Nodes with 0–1 edges are labeled fragment memory and skipped.

Telemetry uses a fixed 1/60s timestep — data is identical regardless of machine FPS. Slow machines take longer in wall-clock time but produce the same trace.

The collected data includes:

  • FIRE — node fires action potential (voltage, fire rate, interval, degree, hop depth)
  • RECV — node receives pulse (voltage, source, accumulated charge, charge efficiency)
  • EDGE — edge traversal (wave voltage, resistance, traversal count, hop depth)
  • EKG — pacemaker fire intervals (vector embedding of node rhythm)
  • PARAM — all hyper-p settings snapshot at trace start (for recursive tuning)
  • HOP — cascade depth statistics per pulse (max depth, total waves, active nodes)
  • POSITION — 3D transform positions of active nodes at each event (source/target for edge distance & topology analysis)

On completion, data is exported as a ZIP file with one CSV per node, a matching node_X_positions.csv sidecar with 3D node positions for each event, plus _meta.txt with trace order and _params.csv with all simulation parameters. This data feeds back into hyper-p tuning for recursive improvement of the biological neuron model.

TOTAL NODES --
ELIGIBLE NODES --
FRAGMENT MEMORIES --
ESTIMATED TIME --
DISK SPACE ~1.5 GB – 6 GB per 2000 nodes
RAM USAGE ~750 MB peak (telemetry buffer + per-node chunks + position sidecars)
Trace Complete