Last Updated: April 11, 2026 Status: Living documentation (updated as architecture evolves)
This directory contains the canonical architectural specifications for Knowledge3D. Each specification defines a critical component of the system, providing detailed design, implementation guidance, and integration contracts.
Reading Order for New Contributors:
- Start with SUPERHUMAN_GENERAL_INTELLIGENCE_SPECIFICATION.md 🌟 Ultimate Goal (March 5, 2026)
- Read SPATIAL_GENERAL_INTELLIGENCE_SPECIFICATION.md (foundational prerequisite for SHGI)
- Read THREE_BRAIN_SYSTEM_SPECIFICATION.md (foundational architecture)
- Read HYPER_MODULAR_ARCHITECTURE.md (structure paradigm — how knowledge is organized)
- Read HYPER_PARALLEL_PROCESSING.md 🆕 (function paradigm — how knowledge is processed)
- Read KNOWLEDGEVERSE_SPECIFICATION.md (runtime memory substrate)
- Read DUAL_CLIENT_CONTRACT_SPECIFICATION.md (human + AI duality)
- Read MEANING_CENTRIC_STAR_SCHEMA_SPECIFICATION.md 🆕 (atomic unit of knowledge + semantic gravity)
- Read CANONICAL_REGISTRY_SPECIFICATION.md 🆕 (canonical IDs, Qdrant registry, symlink discipline)
- Then explore domain-specific specs as needed
Superhuman General Intelligence (SHGI) — The Ultimate Goal of K3D
SHGI is the emergent collective intelligence that arises when multiple K3D Tiny Recursive Models (TRMs) — each created by different makers with different architectures — collaborate with humans inside a shared Galaxy Universe substrate. This is NOT futuristic; it is the natural consequence of K3D adoption and becomes achievable the moment the AI industry embraces K3D technology.
Key Principles:
- Distributed Collective Intelligence: SHGI emerges from collaboration (billions of TRMs + humans), not monolithic scaling (single trillion-parameter model)
- Transparent Execution: All TRM reasoning visible in Galaxy Universe (humans inspect, guide, steer collective intelligence)
- Sovereign Decentralization: No central authority controls SHGI (peer-to-peer network, W3C open specification)
- Procedural Composition: TRMs compose outputs procedurally (RPN programs combine seamlessly, not lossy text interfaces)
- Multi-Modal Unity: ALL modalities (visual, language, math, physics) converge in ONE spatial substrate (emergent cross-domain insights)
The Progression:
- AGI (Artificial General Intelligence) → Isolated, opaque AI (2020s goal)
- SGI (Spatial General Intelligence) → Transparent human-AI collaboration (March 4, 2026)
- SHGI (Superhuman General Intelligence) → Distributed collective intelligence (March 5, 2026)
Historical Principle: Every human breakthrough = Genius idea + TEAM
- Einstein + physics community = Relativity
- NASA + 400K engineers = Moon landing
- Linus Torvalds + millions of developers = Linux
- K3D + Multiple TRMs + humans = SHGI
Timeline: Achievable in 12-18 months (2027) with K3D adoption, not decades
Date Defined: March 5, 2026 (third paradigm defined by PM-KR Community Group)
Spatial General Intelligence (SGI) — Foundational Prerequisite for SHGI
SGI is the paradigm where intelligence—both human and artificial—operates within a shared, navigable 3D spatial environment. Unlike AGI (Artificial General Intelligence), which pursues intelligence in abstract, opaque forms, SGI grounds intelligence in spatial reality where humans and AI cohabit the same cognitive workspace.
Key Principles:
- Spatial Grounding: Intelligence operates in navigable 3D (Galaxy Universe)
- Dual-Client Transparency: Humans + AI share identical data at identical (x, y, z) coordinates
- Procedural Composition: 7M param navigator + procedural programs (not 100B+ monoliths)
- Sovereign Execution: PTX kernels, deterministic traces, zero hidden state
- Multi-Modal Unity: ALL modalities (visual, language, audio, physics) in ONE substrate
AGI vs SGI:
- AGI: Abstract reasoning, opaque systems, separate realities, Gt-scale carbon
- SGI: Spatial grounding, transparent workspaces, shared reality, sustainable (12 Gt CO₂ savings)
Date Defined: March 4, 2026 (from NotebookLM press kit podcast insight)
Quote (Press Kit Podcast):
"For decades, we've interacted with data through flat, two-dimensional windows on a screen. This new paradigm treats software as a three-dimensional, navigable place. A cognitive habitat where we, and AI, can finally explore knowledge together, in a shared home."
Unified Sovereign Memory Architecture (v5.0)
The Knowledgeverse is the runtime memory substrate where all active galaxies, house context, TRM weights, and sovereign reasoning assets coexist in one persistent CUDA/PTX execution domain.
Companions:
- KNOWLEDGEVERSE_TERM_DEFINITION.md 🆕 — Simple term definition for emails/press materials (March 5, 2026)
- KNOWLEDGEVERSE_MVP_ROADMAP.md — Grounded implementation plan separating MVP from research
Grounded Implementation Plan (MVP vs Future)
Analysis of all partner contributions (Grok, Qwen, Kimi, DeepSeek, GLM 4.7) against v5.0 architecture. Separates:
- Already Implemented (in v5.0 spec)
- MVP Additions (sovereignty firewall, compressed audit, self-healing, temporal metadata)
- Post-MVP (adaptive governor, temporal engine, uncertainty TRM, HCMF)
- Research Track (consciousness, scaling laws, quantum cognition, oneiric engine, volition)
Key Finding: 80%+ of partner ideas already in v5.0 or are natural extensions. 20% are MVP-critical hardening.
Production Status: ✅ Implemented and validated (MVP Phase 1 complete, 28/28 tests passing)
Key Concepts:
- 7 Memory Regions (Kernels, Galaxy, House, World, TRM, Audit, Ingestion)
- ONE persistent PTX context (eliminates CUDA switching conflicts)
- Shadow Copy learning (continuous inference-time enhancement)
- SleepTime two-phase commit (knowledge + logic consolidation)
- Ingestion Stargate (raw data → RPN transmutation)
- Router Cartographer (topology learning)
- Hyper-Context Paging (intent-based predictive loading)
- Cross-modal synesthesia (audio ↔ visual ↔ text)
Production Status: ✅ Validated (46.7% ARC-AGI, Sovereign TRM v7)
Integration: Core infrastructure for all K3D operations
Cranium + Galaxy + House Architecture
Defines the three-layer memory hierarchy that enables sovereign reasoning with persistent knowledge.
Key Concepts:
- Cranium (Execution): PTX kernels, RPN VM, sovereign-only hot path
- Galaxy Universe (Active Memory): Multi-modal workspace, embeddings, active reasoning state
- House (Persistent Storage): glTF objects, procedural RPN programs, galaxy boxes
- Shadow Copy learning mechanism (inference-time continuous learning)
- SleepTime consolidation protocol (two-stage: knowledge + logic)
Production Status: ✅ Validated (foundational architecture)
Integration: Referenced by Knowledgeverse, Sovereign Training, Reality Enabler
Paradigm: Cross-Domain Procedural Composition (Structure)
Defines the Hyper-Modular Architecture paradigm where procedural RPN programs compose across ALL modalities (visual, mathematical, physical, auditory), ALL client types (human, AI, robot), and ALL scales (atomic → cosmic).
Term Coined: February 20, 2026 by Daniel Ramos
Key Concepts:
- Cross-domain composition: Math + Drawing + Reality + Audio in ONE RPN program (not separate systems)
- N-client reality: SAME Galaxy entry renders for human (readable) + AI (executable) + robot (actionable)
- Symlink deduplication: Content-based references (zero code duplication)
- Infinite procedural spawning: Programs spawn sub-programs dynamically
- VRAM-resident workspace: Galaxy Universe (not database-backed)
- Procedural sovereignty: PTX + Galaxy only in hot path (zero external dependencies)
- Shadow copy learning: Architecture learns from successful compositions
Companion: HYPER_PARALLEL_PROCESSING.md (how knowledge is processed)
Production Status: ✅ Foundational paradigm (publicly timestamped, defensible)
Integration: Organizing principle for ALL K3D specifications (cross-cutting concern)
Paradigm: Concurrent Specialized Procedural Cognition (Function)
Defines the Hyper-Parallel Processing paradigm where multiple specialized procedural cores operate simultaneously on the same problem, each carrying domain-specific learned weights (LoRA-like specialist adapters) and cross-referencing other cores' intermediate results via shared stack registers, converging to one unified answer.
Term Coined: March 16, 2026 by Daniel Ramos
Key Concepts:
- Specialist parallelism: Each core carries different domain-specific LoRA-like weights (not identical copies)
- Cross-core register communication: STORE/RECALL registers span across cores during execution
- One-mind convergence: All specialists produce ONE unified answer (not a vote or ensemble)
- TRM-spawnable specialist lifecycle: Avatar autonomously creates, activates, and prunes specialists
- RPN as native parallelization substrate: Stack machines map directly to CUDA/ternary hardware cores
- Ternary logic as hardware imperative: Balanced ternary (−1/0/+1) gives 58.5% more info per unit; "uncertain" as first-class value
- Persistent brain model: No cold starts — versioned living brain with rollback on drift
Companion: HYPER_MODULAR_ARCHITECTURE.md (how knowledge is organized)
Together: Hyper-modular = how knowledge is stored. Hyper-parallel = how knowledge is thought.
Production Status: ✅ Foundational paradigm (formally defined with W3C conformance levels A/B/C)
Integration: Processing paradigm for ALL K3D reasoning (cross-cutting concern, companion to Hyper-Modular)
Human + AI Shared Data Contract
Specifies how the same data serves both human (aesthetic) and AI (semantic) needs without duplication.
Key Concepts:
- Dual-texture: UV Map 0 (human visual) + UV Map 1 (AI semantic embeddings)
- Procedural foundation: RPN programs (form + meaning unified)
- Save Information Principle: References/symlinks, not duplication (~70% reduction)
- Galaxy composition: Drawing → Character → Word → Grammar → TRM
- Inspectability: Humans can see AI's working memory (Galaxy View)
- Form→Meaning Evolution: 40,000 years of human knowledge (cave paintings → meta-cognition)
Production Status: ✅ Validated (Drawing/Character/Math galaxies)
Integration: Foundational for all K3D objects (House, Galaxy)
Semantic Gravity & Ternary Force — The Atomic Unit of Knowledge
Defines the Meaning-Centric Star — the canonical unit of knowledge in K3D. A star represents a CONCEPT, not a word. "Cat" is cat in every language; the meaning is the center, language-specific surface forms are references. Stars live in the House (permanent, intentionally organized by the TRM like a librarian) and are reasoned about in the Galaxy where "semantic gravity cohered by meaning" (Christoph Dorn) — a ternary force where meaning replaces mass — organizes loaded concepts during active reasoning.
Term Defined: March 16, 2026 by Daniel Ramos
Key Concepts:
- Meaning at the center: One concept = one star. All languages, visuals, sounds, behaviors are references
- Semantic gravity: Stars attract (affinity, +1), repel (contradiction, −1), or float (unknown, 0)
- Meaning mass: Richly-connected stars are heavy (gravitational centers → room nuclei)
- Emergent rooms: House rooms form from gravitational clustering, not manual placement
- Ternary force operator: TCOMP opcode computes semantic force between any two stars
- Content-addressed identity: star_id = hash(meaning_rpn) — same concept = same ID everywhere
- All 4 layers unified: Form (surface_forms) → Meaning (meaning_rpn) → Rules (grammar_refs) → Meta (meta_refs)
- House ↔ Galaxy lifecycle: Stars LIVE in House, are LOADED into Galaxy for reasoning, WRITE BACK during sleep
Production Status: 📐 Architecture specification (Phase H primary deliverable)
Integration: Extends Reality Enabler dual-program stars to ALL knowledge. Implements Foundational Knowledge Layer 2 as canonical center. Integrates Hyper-Parallel ternary logic as spatial force operator.
Canonical ID Registry & Qdrant Overlay
Defines the k3d_canonical registry: the deterministic ID functions, file-backed Qdrant credential resolution, strict CanonicalLookup overlay, bidirectional symlink helper contract, and bootstrap seed script that keep K3D ingestion aligned on one canonical naming surface.
Key Concepts:
k3d_canonicalas the ingestion-side librarian for stable IDs- deterministic ID helpers (
canonical_slug,canonical_*_id, UUIDv5 point IDs) - strict
CanonicalLookupsemantics (canonical_lookup_miss:{kind}:{key}) - file-backed secret resolution for Qdrant access
- bidirectional symlink writes through
append_ref()/link() - canonical seed set (69 entries after Phase 7.A.1 mathematical-role and nine-language grammar-template registration)
Production Status: ✅ Landed (Phase 7.0 / 7.A first slice)
Integration: Meaning-Centric Star schema, Foundational Knowledge ingestion, vocabulary Qdrant indexing, future letter/math/grammar/word canonical shelves
Physical Robot Integration Architecture
Documents how K3D's House-centric architecture enables physical robots to use the same spatial navigation, procedural knowledge, and semantic understanding as human or AI agents without robot-specific training.
Key Concepts:
- Avatar abstraction is hardware-agnostic (human VR, AI agent, or physical robot)
- Embodiment is built-in (spatial memory, navigation, semantic understanding)
- Form→Meaning bridges perception to action (sensors → Galaxy → RPN → actuators)
- Zero additional training (robots query same Galaxy, execute same RPN programs)
- Actuator mapping layer (abstract commands → motor commands)
- SLAM → House Universe (physical space becomes navigable K3D House)
- Shadow copy enhancement (robots learn from success, no offline retraining)
Production Status: 🔬 Architectural specification (implementation future work)
Integration: Extends avatar abstraction from House Universe, demonstrates K3D's embodied-first design
Sovereign Neural-Symbolic Interface
Defines the PTX-based execution layer that enables zero-dependency inference.
Key Concepts:
- PTX-only hot path (no numpy, cupy, scipy, torch in inference)
- RPN VM (stack-based procedural execution)
- Kernel library (45+ operations: embeddings, matryoshka, grammar, physics)
- Deterministic execution (same inputs → same outputs)
- Fail-fast sovereignty gates (no silent fallbacks)
Production Status: ✅ Validated (Sovereign TRM v7)
Integration: Core execution substrate for Cranium, Knowledgeverse
Ternary-Gated Adaptive Computation Governance
Defines the Adaptive Reasoning Budget (ARB) — a ternary-signal-driven computation governance mechanism that wraps the TRM game loop. The budget scales exponentially with knowledge uncertainty, decomposes into recursive sub-tasks via HTN patterns, and manages parallel-to-serial overflow when processing limits are reached.
Key Concepts:
- Ternary knowledge signal: 5 sources aggregated via Kleene conjunction (+1/0/−1)
- Budget formula: B(q) = B_base × 2^(1−σ(q)) — exponential scaling by signal
- Minimum budget enforcement: Aspiration levels prevent premature halting
- Recursive sub-task decomposition: HTN-style task splitting with dependency DAGs
- Parallel saturation → serial prioritization: Work-stealing then priority queue
- Knowledge persistence: ALL intermediate results as MeaningCentricStars
- Memory watermark governance: Budget/depth/worker reduction under VRAM pressure
X3D Companion: ../w3c/x3d/PM_KR_X3D_ADAPTIVE_REASONING_COMPONENT.md
Production Status: 📐 Architecture specification (grounded in ACT, PonderNet, Metareasoning, HTN, SOAR research)
Integration: Wraps TRM game loop (Three Brain System §3.2), extends Halting Gate (Sovereign NSI §9.2), respects memory watermarks (Knowledgeverse §3.3), budget as Class B RPN recipes (RPN Opcode Registry §6)
Mathematical Reasoning System
Defines the 3-tier math core (symbolic, numeric, geometric) and scaling patterns.
Key Concepts:
- 3-tier allocation: Tier 1 (school/olympiad), Tier 2 (undergrad), Tier 3 (research)
- Symbolic reasoning: LaTeX → RPN programs
- Math Galaxy: Symbols with procedural templates (\frac, \binom, etc.)
- Procedural physics: 9 systems across 18 compute cores
Integration: Math Galaxy (Knowledgeverse Region 2), Reality Galaxy
Physics Simulation Framework
Defines procedural physics systems for the Reality Galaxy.
Key Concepts:
- 9 physics systems (mechanics, EM, thermo, fluids, etc.)
- Procedural RPN implementation (no PhysX, Blender, external sims)
- 18 compute cores allocation (2 cores per system)
- Integration with Math Galaxy (cross-domain reasoning)
Integration: Reality Galaxy (Knowledgeverse Region 2)
Visual Reasoning System (v1.1)
Defines procedural visual primitives, Drawing Galaxy, and 3D technique fusion.
Key Concepts:
- Drawing primitives: LINE, CIRCLE, RECT (as RPN programs)
- VectorDotMap: Universal audio ↔ visual codec
- ARC-AGI visual reasoning (46.7% validation)
- Procedural font rendering (Character Galaxy)
- 3D Technique Fusion: CSG, mesh, L-system, sculpting, parametric, physics, voxel, NURBS as composable RPN (v1.1)
- 2D-to-3D Fusion: Drawing Galaxy entries usable as both 2D textures and 3D materials (v1.1)
- Tool-Nodes: Techniques stored as reusable Galaxy knowledge entries (v1.1)
Integration: Drawing Galaxy, Character Galaxy, Reality Galaxy (Knowledgeverse Region 2)
Audio + Visual + Video Unification (v1.1)
Defines cross-modal signal processing, temporal video architecture, and signal tool-nodes.
Key Concepts:
- Spectrograms as universal interface (audio → visual)
- VectorDotMap codec (shared between Audio/Visual galaxies)
- Sonification (visual → audio)
- Procedural generation (both modalities)
- Five-Layer Temporal Video Contract: Scene/Dynamics/Camera/Render/Audio separation (v1.1)
- Scene-Time Video: Video as scene program evaluated over time, NOT frame lists (v1.1)
- Signal Tool-Nodes: Waveform synthesis, FFT, filter chains as Galaxy knowledge (v1.1)
- K3D Video Benchmarks: Deterministic rebuild, variant edit cost, symlink reuse ratio (v1.1)
Integration: Audio Galaxy, Drawing Galaxy, Reality Galaxy (Cross-Modal Bridge)
Sovereign Training Protocol
Defines training methodology for TRM and specialists.
Key Concepts:
- Shadow Copy learning (inference-time continuous enhancement)
- SleepTime consolidation (two-phase commit)
- Multi-curriculum training (ARC-AGI, math, physics, language)
- LoRA-style specialist adapters
- Deterministic validation (reproducible metrics)
Integration: TRM Weight Manager (Knowledgeverse Region 5)
Knowledge Consolidation Protocol
Defines the two-stage consolidation process (knowledge + logic).
Key Concepts:
- Stage A: Galaxy → House (export knowledge as procedural RPN)
- Stage B: Shadow Copy → TRM (refine specialist adapters)
- Two-phase commit (rollback guarantees)
- Trigger strategies (time-based, buffer-based, manual)
Integration: Knowledgeverse SleepTime, TRM Weight Manager
Base Knowledge Corpus
Defines foundational knowledge to pre-populate galaxies.
Key Concepts:
- Character Galaxy: Procedural fonts (Latin, Greek, mathematical symbols)
- Word Galaxy: Character sequences (symlink references)
- Grammar Galaxy: Transformation rules
- Math Galaxy: LaTeX templates, proofs
- Reality Galaxy: Physics laws, simulations
Integration: Galaxy Manager, Ingestion Stargate
Procedural Compression System
Defines adaptive compression using RPN programs.
Key Concepts:
- Procedural generation > cached snapshots (storage reduction)
- RPN programs as primary source (form + meaning)
- Matryoshka embeddings (64/128/512/2048D multi-resolution)
- Zstd compression for cached snapshots (fallback)
Integration: Galaxy boxes, House objects, Knowledgeverse Region 3
3D User Interface Design
Defines the spatial UI for human interaction with K3D.
Key Concepts:
- Galaxy View (inspect AI working memory)
- House View (navigate persistent storage)
- World View (network collaboration)
- FOV/LOD system (semantic + spatial proximity)
- Doors protocol (network streaming)
Integration: Viewer (TypeScript), Knowledgeverse Region 4
Primary Interface Object (Post-MVP)
Defines the Memory Tablet — K3D's primary interactive surface for humans and AI to navigate, manipulate, and create knowledge in spatial environments (House and Galaxy).
Key Concepts:
- Physical 3D object (not 2D overlay) — persistent canvas in spatial environment
- Dual-client perception — humans see visual UI, AI sees semantic graph (same object)
- Procedural UI rendering — all UI elements = Galaxy references (70% compression)
- Projection screens — virtual monitors inside K3D (controlled via tablet)
- Game menu system — entry sequence (House vs Galaxy selection)
- Simultaneous environments — House + Galaxy run concurrently, tablet bridges both
- Multi-modal interaction — visual, audio, Braille, haptic (native accessibility)
- Legacy content bridge — file system access, VM output integration
- Collaborative workspace — human and AI share same tablet (different perceptions)
Status: 🔬 Architectural design (Phase P implementation, Q1 2027)
Integration: Spatial UI layer (top of stack), Procedural Codecs (VectorDotMap, Procedural Fonts), Galaxy Universe (browse/query), House Universe (private workspace)
Multi-Sensory Accessibility
Defines accessibility features for diverse users.
Key Concepts:
- Visual: High contrast, color blindness modes
- Audio: Screen reader integration, sonification
- Motor: Keyboard navigation, voice control
- Cognitive: Simplified modes, progressive disclosure
Integration: Viewer UI, Spatial UI
glTF Node Extensions
Defines K3D-specific glTF extensions for House objects.
Key Concepts:
extras.k3d_type: Object classificationextras.k3d_dual_client: Human + AI contentextras.k3d_ai_data: Galaxy boxes (procedural RPN)extras.k3d_metadata: RDF semantic links- Dual-texture extensions (TEXCOORD_0 + TEXCOORD_1)
Integration: House objects, glTF loader
RPN Instruction Set & Opcode Admission Pipeline
Canonical registry of all RPN opcodes across domains, with formal capability classification and promotion pipeline.
Key Concepts:
- Core operations (PUSH, POP, SWAP, DUP, etc.)
- Math domain (ADD, MUL, POW, SIN, etc.)
- Visual domain (LINE, CIRCLE, RECT, etc.)
- Galaxy domain (QUERY, COMPOSE, CREATE, etc.)
- Physics domain (VELOCITY, FORCE, INTEGRATE, etc.)
- Three Capability Classes (A/B/C): Executable Now, Representable Now/Kernel Later, Research (v0.2)
- Four-Stage Admission Pipeline: Galaxy Recipe -> Macro -> Opcode Candidate -> PTX Kernel (v0.2)
- Multimodal Target Opcodes: 3D, Signal, Image, Temporal domains at Stage 0-1 (v0.2)
Integration: Cranium RPN VM, Sovereign PTX kernels
HYPER_MODULAR_ARCHITECTURE (Structure Paradigm — Cross-Cutting)
│ └── Applies to ALL regions (cross-domain composition, N-client reality, symlink deduplication)
│
HYPER_PARALLEL_PROCESSING (Function Paradigm — Cross-Cutting)
│ └── Applies to ALL reasoning (specialist swarm, ternary-ready registers, persistent brain model)
│ └── Companion to Hyper-Modular: structure + function = complete cognitive architecture
│
ADAPTIVE_REASONING_BUDGET (Computation Governance — Cross-Cutting)
│ └── Wraps TRM game loop: ternary-gated budget → recursive decomposition → parallel/serial overflow
│ └── Integrates: Halting Gate (NSI §9.2), Memory Watermarks (Knowledgeverse §3.3), RPN recipes (Opcode Registry §6)
│
Knowledgeverse (Runtime Substrate)
├── THREE_BRAIN_SYSTEM (Cranium + Galaxy + House)
│ ├── SOVEREIGN_NSI (PTX execution)
│ ├── DUAL_CLIENT_CONTRACT (Human + AI duality)
│ ├── ROBOTIC_EMBODIMENT (Hardware-agnostic avatar)
│ └── SLEEPTIME_PROTOCOL (Consolidation)
│
├── Region 1: KERNELS
│ └── RPN_DOMAIN_OPCODE_REGISTRY
│
├── Region 2: GALAXY_UNIVERSE
│ ├── MATH_CORE (Math Galaxy)
│ ├── REALITY_ENABLER (Reality Galaxy)
│ ├── PROCEDURAL_VISUAL (Drawing Galaxy)
│ ├── UNIFIED_SIGNAL (Audio Galaxy)
│ └── FOUNDATIONAL_KNOWLEDGE (Base corpus)
│
├── Region 3: HOUSE_CONTEXT
│ ├── K3D_NODE_SPECIFICATION (glTF extensions)
│ └── ADAPTIVE_PROCEDURAL_COMPRESSION (Galaxy boxes)
│
├── Region 4: WORLD_VIEW
│ └── SPATIAL_UI_ARCHITECTURE (Viewer integration)
│
├── Region 5: TRM_WEIGHTS
│ └── SOVEREIGN_TRAINING (Shadow Copy, specialists)
│
├── Region 6: AUDIT_JOURNAL
│ └── SLEEPTIME_PROTOCOL (Event logging)
│
└── Region 7: INGESTION_STARGATE
└── FOUNDATIONAL_KNOWLEDGE (Raw data → RPN)
Each specification includes:
- Version: Semantic versioning (major.minor.patch)
- Last Updated: Date of last significant change
- Status: Draft / Validated / Production
- Dependencies: Other specs it depends on
- Integration: Where it's used in the system
When updating a spec:
- Increment version appropriately
- Update "Last Updated" date
- Add changelog entry at bottom of spec
- Update this README if new concepts added
These rules apply to ALL agents (Claude, Codex, Gemini, specialists) working on K3D:
-
No Python orchestration in hot path. ALL reasoning, rule selection, strategy, and decision-making happens on GPU via PTX kernels, Galaxy (VRAM) star navigation, and RPN programs. Python is boot + I/O only (~200 lines target). See THREE_BRAIN_SYSTEM_SPECIFICATION.md §6b.4.
-
No stubs, fakes, or placeholders. Every implementation must be real, functional code that leverages the K3D architecture. Test mocks are acceptable in test files only — never in production code paths. Galaxy stars must contain real knowledge, not placeholder data.
-
No simulated results. Benchmark scores, learning signals, and route selections must come from actual GPU execution through the sovereign pipeline. Never synthesize or hardcode results.
-
No Python fallbacks. EVER. If the GPU path fails, the system fails — and we fix the GPU path. No
except: return default, noif gpu_failed: use_python(). See SOVEREIGN_NSI_SPECIFICATION.md. -
Ground all work in these specifications. Every architectural decision, kernel composition, and Galaxy entry must trace back to a specification in this directory. If a spec doesn't cover a new capability, write the spec first, then implement.
-
Real knowledge leverages real architecture. Galaxy entries must use the 4-layer structure (Form → Meaning → Rules → Meta-Rules) from FOUNDATIONAL_KNOWLEDGE_SPECIFICATION.md. RPN programs must use registered opcodes from RPN_DOMAIN_OPCODE_REGISTRY.md. Route families must match SOVEREIGN_NSI_SPECIFICATION.md.
When adding a new specification:
- Naming: Use descriptive uppercase names (e.g.,
NEW_COMPONENT_SPECIFICATION.md) - Structure: Follow existing spec templates (see KNOWLEDGEVERSE_SPECIFICATION.md)
- Sections: Include Overview, Key Concepts, Integration, Code Examples, Testing
- Add to README: Update this file with summary and integration map
- Cross-Reference: Link to/from related specs
- Architecture questions → Read KNOWLEDGEVERSE_SPECIFICATION.md
- Implementation questions → Read SOVEREIGN_NSI_SPECIFICATION.md
- Integration questions → Read THREE_BRAIN_SYSTEM_SPECIFICATION.md
- For briefings and high-level overview → See ../briefings/
Maintained by: Claude (Architecture Partner) + Codex (Implementation) + Gemini (Integration) + Community License: See repository LICENSE file Last Full Review: February 28, 2026