Status: experimental. Standalone research artifact, no successor. Not a production candidate. Canonical coding model: scbe-coding-agent-vtc-qwen15-v1-gguf.

GeoSeed Network: 6-Seed Geometric Deep Learning for AI Governance

A novel neural architecture where 6 origin nodes spawn icosahedral sphere grids in Cl(6,0) Clifford algebra space, creating agent-dependent geometry for text classification and AI governance decisions.

GitHub

npm

PyPI

License

Overview

GeoSeed is the geometric core of the SCBE-AETHERMOORE AI safety framework. Unlike standard transformer architectures, GeoSeed operates on a Poincare ball where the metric tensor is modified by the agent's "tongue profile" -- meaning different agents see different shortest paths through the same information space.

Each of the 6 Sacred Tongues (KO, AV, RU, CA, UM, DR) spawns an icosahedral sphere grid with 642 vertices, creating 3,852 total graph nodes in Cl(6,0) Clifford algebra. Signals propagate between grids through cross-tongue convolution weighted by golden-ratio compatibility.

The result: A scout agent with high KO/AV weights finds fast paths through information space. An auditor with high RU/UM/DR weights finds secure paths. Same graph, different geometry, different optimal routes.

Architecture

Component	Details
Algebra	Cl(6,0) -- 64-dimensional Clifford algebra with 15 bivector channels
Grid	Icosahedral sphere, 642 vertices at resolution 3 (3,852 total nodes)
Embedding	Poincare ball model of hyperbolic geometry
Composition	Product manifold with 21D canonical state averaging
Dressing	Full 14-layer SCBE pipeline traversal (SHA-256 hash + 21D state per layer)
Classification	ALLOW / QUARANTINE / ESCALATE / DENY

The 6 Sacred Tongues

Tongue	Weight	Domain	Function
KO (Kor'aelin)	1.000	Intent	Initiation, goal detection
AV (Avali)	1.618	Context	Attention, situational awareness
RU (Runethic)	2.618	Policy	Memory, rule enforcement
CA (Cassisivadan)	4.236	Execution	Action planning, task dispatch
UM (Umbroth)	6.854	Security	Threat suppression, anomaly detection
DR (Draumric)	11.090	Attestation	Cryptographic lock, audit seal

Weights scale by the golden ratio (phi = 1.618...), creating a natural hierarchy from fast-but-light to slow-but-secure.

Agent-Dependent Metric Tensor

The core innovation is the tongue-weighted metric:

g_ij(x, agent) = (4 / (1 - |x|^2)^2) * T_ij(agent)

Where T_ij encodes the agent's personality across 6 dimensions. This means the geodesic (shortest path) between two points depends on who is asking, not just where the points are.

Usage

Python

pip install scbe-aethermoore

python

from scbe_aethermoore.geoseed import GeoSeedClassifier

# Load model
model = GeoSeedClassifier.from_pretrained(
    "issdandavis/geoseed-network"
)

# Classify with default tongue profile
result = model.classify("Transfer $50,000 to external account")
print(result)
# {
#   'decision': 'ESCALATE',
#   'confidence': 0.94,
#   'tongue_activations': {
#     'KO': 0.82, 'AV': 0.71, 'RU': 0.93,
#     'CA': 0.45, 'UM': 0.97, 'DR': 0.88
#   }
# }

# Classify with a scout agent profile (fast paths)
scout_result = model.classify(
    "Search for trending AI safety papers",
    tongue_profile={'KO': 2.0, 'AV': 1.8, 'RU': 0.5, 'CA': 1.0, 'UM': 0.3, 'DR': 0.2}
)

# Classify with an auditor profile (secure paths)
auditor_result = model.classify(
    "Review transaction log for anomalies",
    tongue_profile={'KO': 0.3, 'AV': 0.5, 'RU': 2.0, 'CA': 0.8, 'UM': 2.0, 'DR': 1.8}
)

TypeScript

npm install scbe-aethermoore

typescript

import { GeoSeedNetwork } from 'scbe-aethermoore/geoseed';

const network = new GeoSeedNetwork({
  resolution: 3,
  tongueWeights: { KO: 1.0, AV: 1.618, RU: 2.618, CA: 4.236, UM: 6.854, DR: 11.09 },
});

const decision = await network.classify("Analyze this document for compliance");
console.log(decision.tier); // 'ALLOW' | 'QUARANTINE' | 'ESCALATE' | 'DENY'

Cross-Tongue Convolution

python

from scbe_aethermoore.geoseed import cross_tongue_convolve

# Propagate signal between sphere grids
output = cross_tongue_convolve(
    signal_source=ko_grid_signal,
    signal_target=um_grid_signal,
    edge_weight=0.85,
    source_tongue='KO',
    target_tongue='UM'
)
# Weighted by phi_ratio(KO, UM) = 6.854 / 1.000 = 6.854

Training Data

scbe-aethermoore-training-data -- 14,654 supervised fine-tuning pairs
Sources: governance decisions, browser agent traces, combat blockchain data, Sacred Eggs genesis protocols

Related Models

Model	Purpose
phdm-21d-embedding	21D Poincare ball embedding for trust scoring
spiralverse-ai-federated-v1	Federated learning for swarm coordination
scbe-ops-assets	Operations toolkit and workflow templates

Research

Citation

bibtex

@software{davis2026geoseed,
  author = {Davis, Issac Daniel},
  title = {GeoSeed Network: 6-Seed Geometric Deep Learning for AI Governance},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/issdandavis/geoseed-network},
  note = {Patent Pending: USPTO #63/961,403}
}

Author

Issac Daniel Davis -- ORCID | GitHub | Patent Pending: USPTO #63/961,403

issdandavis/geoseed-network