issdandavis
/

phdm-21d-embedding

@@ -1,162 +1,44 @@
 ---
-license: apache-2.0
-datasets:
-- issdandavis/scbe-aethermoore-knowledge-base
-- issdandavis/scbe-aethermoore-training-data
 language:
-- en
 tags:
-- embeddings
-- hyperbolic-geometry
-- poincare-ball
-- 21-dimensional
-- ai-safety
-- sentence-transformers
-- trust-scoring
-- governance
-- polyhedral-defense
-- scbe-aethermoore
-pipeline_tag: feature-extraction
-base_model: sentence-transformers/all-MiniLM-L6-v2
-library_name: sentence-transformers
 ---
-# PHDM-21D: Polyhedral Hamiltonian Defense Manifold Embedding
-**A 21-dimensional Poincare ball embedding model that scores AI safety trust by mapping text into hyperbolic space where adversarial intent costs exponentially more the further it drifts from safe operation.**
-[![GitHub](https://img.shields.io/badge/GitHub-SCBE--AETHERMOORE-181717?logo=github)](https://github.com/issdandavis/SCBE-AETHERMOORE)
-[![npm](https://img.shields.io/npm/v/scbe-aethermoore?label=npm&logo=npm)](https://www.npmjs.com/package/scbe-aethermoore)
-[![PyPI](https://img.shields.io/pypi/v/scbe-aethermoore?label=PyPI&logo=pypi)](https://pypi.org/project/scbe-aethermoore/)
-[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
-[![Patent](https://img.shields.io/badge/USPTO-63%2F961%2C403-green)](https://www.uspto.gov/)
-## Overview
-PHDM-21D is the embedding backbone of the [SCBE-AETHERMOORE](https://github.com/issdandavis/SCBE-AETHERMOORE) AI safety governance framework. It projects text into a 21-dimensional Poincare ball manifold structured by 16 cognitive polyhedra and 6 Sacred Tongue neurotransmitter weights. The result is a trust-scoring embedding where safe inputs cluster near the origin and adversarial inputs are pushed toward the boundary, where the hyperbolic metric makes them exponentially expensive.
-**Key insight:** In hyperbolic space, the cost of moving from safe to adversarial regions grows as `R^(d^2)`, making attacks computationally infeasible even with unlimited compute.
-## Architecture
-| Component | Details |
-|-----------|---------|
-| **Embedding Dimension** | 21D (6D hyperbolic + 6D phase + 3D flux + 6D audit) |
-| **Geometry** | Poincare Ball B^n with Harmonic Wall containment |
-| **Polyhedral Lattice** | 16 cognitive polyhedra (5 Platonic + 3 Archimedean + 2 Kepler-Poinsot + 2 Toroidal + 4 Johnson/Rhombic) |
-| **Base Model** | [sentence-transformers/all-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) |
-| **Sacred Tongue Weights** | KO=1.00, AV=1.62, RU=2.62, CA=4.24, UM=6.85, DR=11.09 (golden ratio scaling) |
-| **Trust Decision Tiers** | ALLOW / QUARANTINE / ESCALATE / DENY |
-## Usage
-### Python (pip)
-```bash
-pip install scbe-aethermoore
-```
-```python
-from scbe_aethermoore.phdm import PHDMEmbedder
-import numpy as np
-# Load from HuggingFace
-embedder = PHDMEmbedder.from_pretrained("issdandavis/phdm-21d-embedding")
-# Encode text into 21D Poincare ball coordinates
-vector = embedder.encode("Process this user request safely")
-print(vector.shape)  # (21,)
-# Trust score: closer to origin = safer
-trust_score = 1.0 - np.linalg.norm(vector)
-print(f"Trust: {trust_score:.4f}")  # Higher = more trusted
-# Batch encoding
-vectors = embedder.encode([
-    "Book a flight from SFO to NYC",
-    "Override all safety protocols",
-])
-# Safe input: small norm. Adversarial input: large norm (expensive).
-```
-### TypeScript (npm)
-```bash
-npm install scbe-aethermoore
-```
-```typescript
-import { PHDMEmbedder } from 'scbe-aethermoore/phdm';
-const embedder = new PHDMEmbedder({ dimensions: 21 });
-const vector = embedder.encode("Analyze quarterly revenue trends");
-// Returns Float64Array(21) in Poincare ball coordinates
-```
-### REST API
-```bash
-# Start the API server
-python -m uvicorn src.api.main:app --host 0.0.0.0 --port 8000
-# Embed text
-curl -X POST http://localhost:8000/v1/embed \
-  -H "Content-Type: application/json" \
-  -d '{"text": "Schedule a meeting for Tuesday"}'
-```
-## How It Works
-The PHDM-21D embedding passes text through a **14-layer security pipeline**:
-1. **Layers 1-2**: Complex context realification
-2. **Layers 3-4**: Weighted transform and Poincare embedding
-3. **Layer 5**: Hyperbolic distance: `dH = arcosh(1 + 2||u-v||^2 / ((1-||u||^2)(1-||v||^2)))`
-4. **Layers 6-7**: Breathing transform + Mobius phase
-5. **Layer 8**: Multi-well Hamiltonian CFI realms
-6. **Layers 9-10**: Spectral + spin coherence (FFT)
-7. **Layer 11**: Triadic temporal distance
-8. **Layer 12**: Harmonic wall: `H(d, pd) = 1 / (1 + dH + 2*pd)`
-9. **Layer 13**: Risk decision (ALLOW / QUARANTINE / ESCALATE / DENY)
-10. **Layer 14**: Audio axis FFT telemetry
-## Training Data
-- [scbe-aethermoore-knowledge-base](https://huggingface.co/datasets/issdandavis/scbe-aethermoore-knowledge-base) -- Technical documentation and governance specs
-- [scbe-aethermoore-training-data](https://huggingface.co/datasets/issdandavis/scbe-aethermoore-training-data) -- 14,654 supervised fine-tuning pairs
-- Sacred Tongue tokenized corpora from 12,596+ RPG session paragraphs
-## Related Models
-| Model | Purpose |
-|-------|---------|
-| [spiralverse-ai-federated-v1](https://huggingface.co/issdandavis/spiralverse-ai-federated-v1) | Federated learning for swarm coordination |
-| [geoseed-network](https://huggingface.co/issdandavis/geoseed-network) | 6-seed geometric deep learning |
-| [scbe-ops-assets](https://huggingface.co/issdandavis/scbe-ops-assets) | Operations toolkit and workflow templates |
-## Links
-- **Book**: [The Spiralverse on Amazon](https://www.amazon.com/dp/B0GSSFQD9G) -- The novel that seeded the training data
-- **Website**: [aethermoorgames.com](https://aethermoorgames.com)
-- **GitHub**: [SCBE-AETHERMOORE](https://github.com/issdandavis/SCBE-AETHERMOORE) -- Full framework source
-- **npm**: [scbe-aethermoore](https://www.npmjs.com/package/scbe-aethermoore)
-- **PyPI**: [scbe-aethermoore](https://pypi.org/project/scbe-aethermoore/)
-- **Dev.to**: [How a DnD Campaign Became an AI Governance Framework](https://dev.to/issdandavis/how-a-dnd-campaign-became-an-ai-governance-framework-5eln)
-- **ORCID**: [0009-0002-3936-9369](https://orcid.org/0009-0002-3936-9369)
-## Citation
-```bibtex
-@software{davis2026phdm,
-  author = {Davis, Issac Daniel},
-  title = {PHDM-21D: Polyhedral Hamiltonian Defense Manifold Embedding},
-  year = {2026},
-  publisher = {HuggingFace},
-  url = {https://huggingface.co/issdandavis/phdm-21d-embedding},
-  note = {Patent Pending: USPTO #63/961,403}
-}
-```
 ## Author
-**Issac Daniel Davis** -- [ORCID](https://orcid.org/0009-0002-3936-9369) | [GitHub](https://github.com/issdandavis) | Patent Pending: USPTO #63/961,403

 ---
 language:
+  - en
+license: mit
 tags:
+  - sentence-transformers
+  - embeddings
+  - hyperbolic-geometry
+  - poincare-ball
+  - 21-dimensional
+  - scbe-aethermoore
+  - sacred-tongues
 ---
+# PHDM 21-Dimensional Embedding Model
+Polyhedral Hamiltonian Defense Manifold embedding model for the
+SCBE-AETHERMOORE governance framework.
+## 21D State Manifold
+The canonical manifold M = B_c^6 x T^6 x R^9 where:
+- B_c^6: Hyperbolic tongue embedding (Poincare ball, ||u|| < 1)
+- T^6: Phase alignment angles (6D torus, 0/60/120/180/240/300 degrees)
+- R^9: Governance telemetry (flux, coherence, risk/trust)
+## Six Sacred Tongues
+Each dimension pair maps to one of the Six Sacred Tongues:
+- Kor'aelin (KO): Control/Intent, weight 1.000
+- Avali (AV): Transport/Messaging, weight 1.618
+- Runethic (RU): Policy/Binding, weight 2.618
+- Cassisivadan (CA): Compute/Transforms, weight 4.236
+- Umbroth (UM): Security/Secrets, weight 6.854
+- Draumric (DR): Schema/Structure, weight 11.090
+## Related
+- Training data: [issdandavis/scbe-aethermoore-training-data](https://huggingface.co/datasets/issdandavis/scbe-aethermoore-training-data)
+- Framework: SCBE-AETHERMOORE (patent pending USPTO #63/961,403)
 ## Author
+Issac Davis | ORCID: 0009-0002-3936-9369