One fact. Six perspectives. Infinite depth. It moves.
A single-file, offline-first, holographic 3D knowledge system. No server. No API. No dependencies. Drop the HTML in a browser and watch your knowledge grow, cluster, fission, and reorder itself in real time.
Data Cube is a personal knowledge tool built around one idea: a fact alone is incomplete. Every piece of knowledge has a counter-argument, a cultural lens, a speculative edge, a historical frame, and an unresolved question. Data Cube makes these visible — literally — by mapping them onto the six faces of a rotating 3D cube.
Each cube is a topic. Each face is a perspective lens. Rotate to change your viewpoint.
But the system doesn't stop at one cube. As you add knowledge, cubes attract each other based on keyword similarity, drift together in real time, form clusters with merged vocabularies, and split (fission) when they grow too large or diverse. The space reorganises itself continuously. You watch it learn.
| Face | Lens | Colour | Purpose |
|---|---|---|---|
| Front | ◈ FACT | Cyan | The prime verifiable statement that seeds the cube |
| Back | ⊘ COUNTER | Red | The refutation, opposing argument, or challenge |
| Left | ◎ OPINION | Purple | A personal, cultural, or subjective perspective |
| Right | ◇ FICTION | Amber | A speculative, narrative, or imagined take |
| Top | ⊡ CONTEXT | Green | Historical, scientific, or wider framing |
| Bottom | ? UNKNOWN | Grey | What remains unresolved, uncertain, or unexplained |
A fully filled cube is a complete knowledge unit — six ways of seeing the same thing, held in the same object.
This is a living knowledge organism. The space reorganises itself continuously.
A single perspective. Six per cube. The atomic unit.
A complete topic. One fact seen through all six lenses.
When related cubes share enough keywords, they drift together and lock into a cluster. Black Holes, Light Speed, and Quantum Mechanics form a PHYSICS cluster. The cluster has its own merged vocabulary and visual label.
Clusters group into domains. PHYSICS + BIOLOGY + CHEMISTRY converge into a SCIENCE domain. The space becomes a galaxy — dense cores with connective tissue between them.
Every new entry triggers a re-score:
- Cubes with growing keyword overlap drift toward each other in real time
- Orphan cubes with enough shared vocabulary get pulled into existing clusters
- Clusters that grow too large fission into two sub-clusters that drift apart, with explosion animations
- Edges glow between related cubes, fading with distance
Nothing is permanently placed. The system has its own opinion about where things belong, and it updates that opinion as it learns.
- Zoomed out — clusters appear as glowing formations with topic labels
- Zoomed in — clusters dissolve into individual cubes
- Zoomed in further — face text becomes readable
Three levels of resolution. The same continuous 3D space.
Type into the input bar, hit Enter or +. If no strong match exists, your entry becomes the FACT face of a new cube.
Add related entries. The rule-based classifier suggests the right lens:
- "however", "wrong", "contrary" → COUNTER
- "think", "believe", "perspective" → OPINION
- "imagine", "story", "what if" → FICTION
- "history", "research", "origin" → CONTEXT
- "unknown", "mystery", "unresolved" → UNKNOWN
Confident matches auto-route. Ambiguous ones open the Lens Picker.
- Filled face → expand overlay showing full text, timestamp, confidence score, and delete option
- Empty face → Lens Picker to add directly to that slot
- Drag to orbit the scene
- Scroll (desktop) or pinch (mobile) to zoom
- Pip row under each cube shows filled lenses, colour-coded by type
Use the ↓ import button to drop .txt, .md, .csv, or .json files, or paste lines directly. Each line seeds a new FACT cube.
Ctrl+F (or Cmd+F) opens the search panel. Filter by lens, search keywords, and jump directly to cubes in 3D. Results are sorted by confidence and matching text is highlighted.
Ctrl+L (or Cmd+L) opens the layer panel. View all entries for a specific lens across every cube, sorted by confidence score. Click any card to jump to the cube in 3D.
Click the clock icon to open timeline view. Scrub through history to see how your knowledge evolved. Cubes appear and faces update as you move through time.
Click the export icon to download a JSON snapshot of your entire graph — topics, entries, timestamps, and top keywords — for use in other tools or as a backup.
Click the moon/sun icon to switch between dark and light themes. Your preference is saved.
From any expanded entry, click DELETE CUBE to remove it permanently (with confirmation).
Every entry is scored the moment it is committed — no AI, no API, pure signal analysis.
Each face entry receives a confidence score based on three components:
- Lens keyword density (50pts) — how many of that lens's signal words are present. FACT scores on absence of hedge language instead
- Entry substance (30pts) — word count plus unique vocabulary density, rewards rich detailed entries
- Specificity bonus (20pts) — bigram count rewards multi-word concrete concepts over vague single words
A composite of completeness and quality: (filled faces / 6) × 40% + (average face confidence) × 60%. A half-filled cube of strong entries scores higher than a fully-filled cube of weak ones.
| Score | Grade |
|---|---|
| 90–100 | CRYSTALLINE |
| 65–89 | COHERENT |
| 35–64 | FORMING |
| 0–34 | SPARSE |
- Each face — coloured confidence bar along the bottom edge, percentage on hover
- Cube label — integrity badge top-right, colour-shifts from grey → cyan → white → glowing as it climbs
- Expand overlay — animated confidence bar with grade label
- Search panel — mini confidence bar under every result
- Layer panel — cards sorted highest confidence first, with grade label
The classifier is entirely rule-based and offline — no AI API, no internet, no external calls.
- Tokenisation — strips stopwords, extracts keywords and bigrams
- Similarity scoring — compares entry tokens against each cube's vocabulary (bigrams count double)
- Stem matching — partial root comparison catches plurals and conjugations
- Lens detection — language cue lists per lens suggest the right face
- Three-tier routing:
- Score ≥ 42% → auto-routes silently
- Score 12–41% → opens Lens Picker
- Score < 12% → new cube spawned automatically
Every entry feeds the cube's keyword vocabulary. Every cluster merger combines those vocabularies. The system gets smarter the more you use it — entirely without AI, entirely from the structure of what you have already put in.
Memory persists across sessions via localStorage.
- Attraction — keyword similarity pulls cubes together continuously
- Repulsion — overcrowding pushes cubes apart (inverse square law)
- Edges — lines connect related cubes, opacity scales with similarity and proximity
- Cluster detection — connected components of ≥ 3 cubes get a floating label and merged vocabulary
- Fission — when cluster variance > 0.6 or size > 12, it splits via k-means with explosion animation
datacube.html ← entire application, single file
│
├── CSS ← holographic aesthetic, light/dark themes
│ ├── CSS 3D transforms
│ ├── Per-lens colour variables
│ ├── Animated star field (canvas)
│ ├── Perspective grid floor
│ └── Theme system with color-mix()
│
├── Classifier ← pure JS, zero dependencies
│ ├── Tokeniser + stopword filter
│ ├── Bigram extractor
│ ├── Similarity scorer (TF-IDF style)
│ ├── Stem-lite root matcher
│ └── Lens cue detector
│
├── Confidence Engine ← quality scoring layer
│ ├── Face confidence scorer (lens density + substance + specificity)
│ ├── Cube integrity composite (completeness × quality)
│ ├── Grade classifier (SPARSE → CRYSTALLINE)
│ └── Visual output (face bars, cube badge, expand overlay, panels)
│
├── Gravity Engine ← continuous spring-force simulation
│ ├── Attraction (keyword similarity)
│ ├── Repulsion (inverse square law)
│ ├── Velocity damping and capping
│ └── Centre gravity to contain drift
│
├── Cluster Engine ← emergent organisation
│ ├── Graph-based cluster detection
│ ├── Merged vocabulary per cluster
│ ├── Variance calculation (cosine similarity)
│ ├── Fission detection + k-means split
│ └── Visual cluster labels
│
├── Visual Edges ← relationship rendering
│ ├── Dynamic line creation
│ ├── Opacity based on similarity and distance
│ └── 3D rotation between cubes
│
├── Timeline View ← temporal exploration
│ ├── Time-range detection
│ ├── Canvas visualisation of cluster events
│ ├── Slider scrubbing through history
│ └── Face filtering by timestamp
│
├── Data Model
│ ├── nodes[] ← cube objects
│ ├── node.entries[] ← [{lens, text, addedAt, confidence}]
│ ├── node.memory ← Map<token, frequency>
│ ├── node.velocity ← {x,y,z} for gravity simulation
│ ├── clusters[] ← grouped node formations
│ └── cluster.memory ← merged vocabulary of all member nodes
│
├── Persistence ← localStorage, key: datacube_v3
│
└── UI
├── 3D orbit (mouse, touch, pinch-zoom)
├── Lens picker modal
├── Expand overlay (full text + confidence + delete)
├── Import modal (drag/drop, paste)
├── Search panel (lens filters, keyword highlight, confidence sort)
├── Layer view (per-lens card grid, confidence sorted)
├── Timeline controls
├── Export button (JSON download)
├── Theme toggle (dark/light, persisted)
├── Toast notifications
└── Keyboard shortcuts (Ctrl+F, Ctrl+L, Esc)
- Six-lens cubes with colour-coded faces
- Rule-based classifier with auto-routing
- Confidence scoring (face + cube integrity)
- Full gravity simulation (attraction/repulsion)
- Cluster detection with visual labels
- Fission (automatic splitting of large clusters)
- Visual edges between related cubes
- Timeline view (scrub through history)
- Light/dark theme toggle
- Export to JSON
- Cube deletion
- Search with lens filters and highlighting
- Layer view by lens type sorted by confidence
- Import from file or paste
- Full persistence (localStorage)
- Mobile touch support
- Keyboard shortcuts
- Level of Detail (LOD) — clusters render as solid formations at far zoom, dissolve into cubes as you approach
- Collaborative mode — shared cube spaces where multiple people fill different lenses
- Teaching and debate tools — assign specific lenses to students or participants
- Research assistant — track open questions systematically via UNKNOWN faces
- No build step. Open the
.htmlfile directly in any modern browser. - No external dependencies beyond two Google Fonts (Orbitron, Share Tech Mono) — works fully offline if fonts are cached.
- localStorage is used for persistence. Export provides a full JSON backup.
color-mix()requires Chrome 111+, Firefox 113+, or Safari 16.2+.- Performance tested to ~500 cubes with stable 60fps gravity simulation.
Giblets Creations — I wanted it. So I forged it. Now forge yours.
Special thanks to parallel processing, pasta fagioli, and mum. 🍝
Single file. Zero server. Infinite depth. It moves. The more you feed it, the more it knows where everything belongs.
