AI Governance · Bounded AI in Governed Loops · Port Angeles, WA
The pattern across everything I build: bounded AI inside governed loops
Three domains. One pattern.
| Domain | Project | Role |
|---|---|---|
| AI Governance | SCBE-AETHERMOORE | The control plane — hyperbolic geometry makes adversarial behavior geometrically expensive |
| Inference Orientation | Polly Pump | Inference-time orientation layer — separates orientation from expression so models stay aligned without retraining |
| Agent Orchestration | Flock Shepherd (inside SCBE) | Swarm coordinator — bounded teammate AI in flock/assistant lanes |
Patent pending: USPTO #63/961,403
Quantum-resistant. 14-layer hyperbolic pipeline. EU AI Act ready. From $49/mo.
What it does: Every AI agent action passes through 14 mathematical layers in a Poincaré ball model. Safe behavior clusters near the center. Adversarial drift hits an exponential cost wall. The framework can't be fooled by prompt injection because it doesn't classify intent — it measures geometric distance from trusted-behavior space.
95.3% adversarial detection · <8ms latency · 950+ tests passing ML-KEM-768 + ML-DSA-65 post-quantum crypto · signed audit trailInstall:
npm install scbe-aethermoore pip install scbe-aethermooreQuick score:
import { harmonicScale } from 'scbe-aethermoore/harmonic'; const score = harmonicScale(2.5, 0.1); // 0.238 = elevated risk // 1.0 = trusted center · 0.0 = boundary
Six constructed languages — original IP by Issac Davis — each with 256 tokens and phi-weighted scaling:
| Tongue | Weight | Domain |
|---|---|---|
| Kor'aelin (KO) | 1.00 | Intent |
| Avali (AV) | 1.62 | Metadata |
| Runethic (RU) | 2.62 | Binding |
| Cassisivadan (CA) | 4.24 | Compute |
| Umbroth (UM) | 6.85 | Security |
| Draumric (DR) | 11.09 | Structure |
Cassisivadan = "Issac" reversed + "Davis" reversed.
Null space tongue signatures detect attacks by absence patterns — what a message does not say is as diagnostic as what it does.
A 233K multi-view SFT dataset on HuggingFace (issdandavis/scbe-aethermoore-training-data) built on a four-level stack:
| Level | Function |
|---|---|
| L0 | Raw bytes |
| L1 | Tongue tokens (Sacred Tongues bijective tokenization) |
| L2 | Pump orientation (Polly Pump inference-time alignment) |
| L3 | Expression (final output) |
Govern AI agent swarms as a collective. Spawn, monitor, and rebalance fleets using balanced ternary consensus. Agents that degrade get their tasks redistributed automatically — no central coordinator required.
from scbe_aethermoore.flock_shepherd import FlockShepherd, SheepRole
shepherd = FlockShepherd(max_flock_size=50)
agent_id = shepherd.spawn_agent(role=SheepRole.EXECUTOR, training_track="code_review")
decision = shepherd.propose_action("deploy_update") # ALLOW / QUARANTINE / DENY| Repo | What it is |
|---|---|
| six-tongues-geoseal | Six Tongues tokenizer + GeoSeal CLI (MIT) |
| spiralverse-protocol | Spiralverse Protocol (MIT) |
| phdm-21d-embedding | PHDM 21D embedding experiments |
| naksha-studio | Naksha Studio |
| oh-my-codex | Oh My Codex |
| arckit-codex | Arckit Codex |
| issdandavis.github.io | aethermoorgames.com landing page source |
The Spiral of Pollyoneth — a Spiralverse novel. How a DnD campaign became an AI governance framework.
SCBE-AETHERMOORE is dual-licensed: MIT core (non-commercial free) + commercial license for production use.
| Tier | Price | Agents |
|---|---|---|
| Starter | $49/mo | Up to 8 |
| Pro | $149/mo | Up to 40 |
| Enterprise | $499/mo | Up to 250 |
→ Full pricing → GitHub Sponsors → Ko-fi
Issac Daniel Davis 📧 issdandavis@gmail.com 🐦 @davisissac 📝 Substack 🌐 aethermoorgames.com 🔬 ORCID: 0009-0002-3936-9369 🏛️ Patent pending: USPTO #63/961,403