AetherMoore
Geometric AI governance — 31% training improvement, zero extra compute

Your AI already has a geometry. We give it the right one before training begins.

AetherMoore builds structured geometric priors into AI training using six Sacred Tongue channels, hyperbolic cost scaling, and multi-view supervision. The result: 31% better code training and 14% better chat training at matched compute, with crossover at step 5. The model learns what to process AND what to skip — constraint-aware reasoning from the first gradient.

Patent pending (USPTO #63/961,403). SAM.gov registered (UEI: J4NXHM6N5F59). DARPA-verified entity. Solo build by Issac Davis. Ask Polly anything — she's the trained AI in the corner.

See the Results Try the Demos Explore the Toolkit ($29) Contact for Licensing

Buyers: starter pack with templates and manual. Investors: measurable results, live demos, patent-pending geometric moat. Researchers: 14-layer pipeline with 5 quantum axioms mapped to production code.

Proof before purchase
31% better code training Sacred Tongue triangulation (KO/CA/DR channels) beat single-view baseline by 31% on T4 GPU. The geometric scaffold works from the first gradient — crossover at step 5.
14% better chat training Four-layer supervision (L0 substrate + L1 tokens + L2 orientation + L3 expression) beat expression-only training. Same model, same data, same GPU budget.
91 / 91 attacks blocked Hyperbolic cost scaling + multi-factor omega gate blocked the full attack corpus. 85.7% detection, 0% false positives — attacks become computationally infeasible.
Canonical wall: Hwall(d*,R) = R(φ·d*)2 Super-exponential cost barrier. At d*=1 the cost is 13.7×. At d*=2 it's 35,341×. At d*=3 it's 1.6×109. Attacks don't just cost more — they become computationally impossible.
Try AI Governance — Live Real SCBE math

Type any AI prompt or task below. The 14-layer pipeline evaluates intent, calculates hyperbolic risk, and returns a governance decision — the same math that blocked 91/91 attacks.

Uses the canonical harmonic wall: Hwall(d*,R) = R(φ·d*)2. Same formula, same math, running in your browser.

Full demo hub →
31% code improvement 14% chat improvement 7.9% mixed L0+vault improvement 3/3 dataset types beaten Patent pending (USPTO #63/961,403)
NPM
PyPI
HuggingFace
DOCKER
GITHUB
Three ways in

Buy the toolkit, test the live demos, or verify the math. Your call.

The starter kit gets you governing AI workflows in one afternoon. The demos run your inputs through the actual 14-layer pipeline. The research stack has every formula, benchmark, and test result — nothing hidden.

Buy the starter pack

Templates, decision records, threshold worksheets, and a manual. One checkout, no subscription. You'll have a governed AI workflow running before dinner.

Open the offers page

Try the live demos

Run text through the hyperbolic risk calculator, encode with Sacred Tongues, test the Hydra swarm, or chat with Polly — our fine-tuned Qwen model trained on the SCBE architecture.

Open the demo hub

Inspect the proof

Every claim on this page has a corresponding benchmark, formula, or test result. The research index separates what's proven from what's experimental. Verify before you trust.

Open research and proof

Dual-Layer Training

The story is part of the training stack, not decorative lore.

AetherMoore teaches the same system in two forms. Buyers can work from the toolkit and manuals directly, or use the novel layer to absorb the same invariants through narrative, dialogue, and repeated pressure-tested scenarios.

Surface layer

The Six Tongues Protocol reads like a 41-chapter isekai epic: Marcus Chen, betrayal, cosmic stakes, the Crystal Archive, and Polly guiding readers through a world that runs on governed intent.

Submerged layer

Every major beat carries actual SCBE mechanics: 14-layer governance, Sacred Tongue tokenization, drift pressure, hyperbolic distance, and the same four-band public gate the live demos expose in public. You should finish the story with the architecture already in your head.

AI bonus

Most governance systems hand you a dry PDF. This one gives you the technical manual and the memory hook. Feed the chapters to a model and the same safety invariants land through narrative context, dialogue, and repetition instead of raw papers alone.

Read Chapter 1 Free → Train Your Brain Open Toolkit + Novel Paths Open the Training Vault
Who this is for

Built for people wiring AI to real tools, not collecting whitepapers.

If you're connecting LLMs to tools, running browser automations, or deploying agent fleets — you need a deterministic control layer, not more prompt engineering.

AI agent builders

Use this if you are building tool-using agents with OpenAI, Claude, Hugging Face, LangChain, MCP, or your own local lane and need an explicit gate instead of ad hoc prompt rules.

Automation operators

Use this if you run workflow automations, cross-app connectors, or browser/CLI routines and want a first pilot surface for thresholding, review, and recovery before a larger rollout.

Security-minded teams

Use this if prompt injection, unsafe tool calls, or retrieval misuse are already on your threat list and you want a proof-backed starter surface rather than a vague “AI policy” document.

Concrete first pilot

Example: stop an AI agent from executing unsafe tool calls when drift, intent accumulation, or context mismatch pushes the session outside a safe boundary. The toolkit gives you the decision record, threshold worksheet, and manual path; the demo hub shows the geometry and gate behavior in public.

How the surfaces connect

The demo hub is the validation lane. The toolkit is the package that turns those same surfaces into a manualized first rollout. The research stack exists so you can inspect the math and benchmark history before trusting the product claim.

Open the demo hub

Buyer expectations

Know exactly what you get and how to recover if anything breaks.

The manual, delivery path, and support route are all public before checkout. No mystery boxes.

One-time package

This is a starter pack, not a subscription or a consulting mystery box. Buy once, receive the package, follow the manual.

Manual before checkout

The toolkit manual and delivery page are public on purpose so you can inspect the operating surface before you commit.

Open the manual

Visible support route

If something arrives broken or incomplete, the support page and contact path are part of the product, not an afterthought.

Open support

What you get

Open the box, fill one worksheet, govern one workflow. Done.

Fifteen minutes to your first governed decision record. The toolkit is templates and a manual, not a framework you need to study for a week.

Inside the package

  • Decision record template (ALLOW / QUARANTINE / ESCALATE / DENY)
  • Threshold worksheet (starting boundaries + assumptions)
  • Pilot checklist (so you do not over-scope the first rollout)
  • Review notes format (what surprised you, what to tighten next)
  • Package manual (first-run path + success checks)
  • Delivery + recovery instructions if anything breaks

First win (15 minutes)

  • You pick one real workflow to govern first.
  • You fill one threshold worksheet.
  • You complete one decision record on a real example.
  • You write one review note about what to tighten next.

If you can do those four steps, the toolkit is working.

What this helps you do

  • Start a governed workflow from a fixed template instead of a blank page
  • Make thresholds and review steps explicit (so “policy” isn’t just vibes)
  • Evaluate whether SCBE fits your environment before going deeper
  • Use a manual-first package instead of guessing what matters in the repo
The Narrative

The Six Tongues Protocol is the narrative layer of the system.

Governance is not just a spreadsheet and the novel is not just flavor text. Follow Marcus Chen into Aethermoor and watch the same SCBE mechanics from the demos show up as plot pressure, dialogue, and world rules you can actually remember.

Why an isekai?

The complete novel is a second training surface for the same architecture: a story of survival, engineering, intent, and governance pressure that teaches the 14-layer stack without forcing readers through a sterile whitepaper first. It is a training artifact disguised as a story on purpose.

  • 41 chapters of tactile magic and system pressure
  • SCBE mechanics embedded into plot, dialogue, and stakes
  • Built to train human memory and model context at the same time
Aethermoor: The World Tree
Polly: The Archive Keeper

Polly: The Archive Keeper

"CAW. You are not lost. You are just early." Meet Polivara (Polly), the Fifth Circle Keeper. She is the guide through the Crystal Archive, where system theory becomes physical place.

Archive Guide 5th Circle Ink & Logic
For builders

Train your own AI security model.

The SCBE AI Security Training Vault gives you the clean training path we actually validated: a raw synthetic corpus generator, a train-ready SFT lane, projector weights, benchmarks, and the Colab workflow to fine-tune a governed model on a free T4 GPU.

If you want the narrative layer too, pair the vault with The Six Tongues Protocol. The vault gives you the structured lane; the novel gives you the same invariants expressed as memorable scenes, dialogue, and semantic pressure instead of plain documentation.

The vault follows a Cadet → Role School → Squad training doctrine: shared foundation first, role specialization second, squad integration third. See the full training pipeline.

What's inside

  • Raw Spiralverse corpus generator for semantic conversation data
  • Train-ready SFT dataset lane (Hugging Face)
  • Semantic projector weights (385x6 ridge regression)
  • Benchmark suite for evaluation
  • Colab fine-tuning workflow (free T4 GPU)
  • Quickstart conversion docs: raw generator -> SFT -> training

AI Security Training Vault

Everything you need to train a governed model from scratch.

$29 one time
Buy the Training Vault
Watch

See the system in action.

Video walkthroughs, chapter recaps from the novel, and live demos of the governance pipeline.

Story Series

The Six Tongues Protocol — narrated chapter-by-chapter. Follow Marcus Chen through Aethermoor as he uncovers the governance system that holds the world together.

  • 41 chapters, narrated with AI voice
  • Based on the published novel
  • Each episode explores a layer of the SCBE system
Field Notes

Latest from the lab.

Research results, architecture discoveries, and training pipeline updates from active development.

April 5, 2026

Canonical Formula Lock

All 14 layers locked. Harmonic wall finalized as R(φ·d*)2 after four iterations. Toroidal resonant cavity gives 176-bit equivalent security from geometry alone. 6008 tests passing.

See the evolution →
April 3, 2026

HQNN + Polyhedral Light-Path Router

The Scattered Attention Sphere is 80-90% of a Holographic Quantum Neural Network. Dodecahedral routing in 12D with A5 symmetry gives O(1) path selection. Confirmed novel.

Read the research →
April 3, 2026

Holographic Bit-Matrix Architecture

Binary substrate + ternary tongue modulation + holographic scatter field. 10^52 unique engraved shapes from a single dodecahedron. Infinite training data pairs.

See the architecture →
Formula Evolution

Four iterations. One winner. Here's the math and why it matters.

The harmonic wall is the core of the system — it decides how expensive adversarial behavior becomes. We didn't get it right on the first try. Here's the honest path from broken to canonical.

V1: Raw Exponential (January 2026)

Retired
H(d) = Rd²

Problem: Numerical collapse at small distances. The bare exponent d² produced AUC of only 0.054 — essentially random. The formula couldn't distinguish safe from dangerous at close range.

Lesson: Raw exponential without golden ratio scaling has no structure — it's either too flat or too steep.

V2: Bounded Tanh (February 2026)

Retired
H(d*) = 1 + α · tanh(β · d*)

Problem: Bounded output — maximum amplification was only 6.76×. An attacker could eat the full cost and still operate. The ceiling made attacks expensive but not infeasible.

Lesson: Any bounded function (tanh, sigmoid, etc.) has a ceiling. Security walls need unbounded cost growth.

V3: Additive Linear (March 2026)

Retired
H(d, pd) = 1 / (1 + d + 2 · pd)

Problem: Linear scaling meant geometry contributed only 8.5% of the final risk decision. The formula was stable but toothless — it scored, it didn't wall. Attacks at d=3 faced only 6× higher cost.

Lesson: Additive formulas can't create walls. Security needs multiplication, and exponentiation beats addition.

V4: Super-Exponential Harmonic Wall (April 2026)

Canonical
Hwall(d*, R) = R(φ · d*)2   |   Hscore = 1 / Hwall
13.7×
Cost at d*=1
35,341×
Cost at d*=2
1.6×109
Cost at d*=3
1053
Full cavity at d*=1

Why this wins: Super-exponential growth via φ² ≈ 2.618 in the exponent. No ceiling — cost grows without bound. The golden ratio creates self-similar spacing (Fibonacci cascade). Tunable base R controls steepness. d* clamped at 10 prevents overflow.

How φ works here: φ² = φ + 1 means adjacent harmonic walls couple through the Fibonacci recurrence. Six tongue walls in orthogonal planes create a toroidal resonant cavity: R(122.99 · d*²) — equivalent to 176-bit cryptographic security from geometry alone.

Patent pending (USPTO #63/961,403). Full formula registry, all 14 layers locked, and cross-layer invariant proofs available in the research index.

See the full formula registry
Fit

Good for builders who read manuals. Not for people expecting magic.

This is a starter pack for people who wire things themselves. If you want a fully managed enterprise deployment for $29, this isn't it.

Good fit

  • Independent builders and operators who want a structured first step
  • Teams exploring policy, thresholds, decision records, or review gates
  • Buyers who want a compact package before touching the full SCBE system

Not for

  • People expecting a fully managed enterprise rollout for $29
  • Buyers looking for the entire repo, all research history, and every experiment at once
  • Anyone who does not want to read a short manual and configure their own workflow
Proof and trust

Inspect everything before you spend a dollar.

The manual is public. The demos are live. The benchmark data is downloadable. Judge the system on evidence, not promises.

Delivery

Checkout, receipt, manual, first governed workflow. Four steps.

Stripe handles payment. You get the package instantly. The manual tells you what to do first. Support exists if anything breaks.

Public contact

If delivery fails or a file is broken, the public support route is ai@aethermoore.com.

FAQ

Four questions, straight answers.

Is this a subscription?

No. This offer is a one-time purchase.

Do I need the whole repo to use it?

No. The package is paired with a buyer manual so you can work from the usable surface first.

What if I buy it and something is missing?

Use the delivery page and support route immediately. The buyer path is part of the product, not an afterthought.

Can I talk to the AI before buying?

Yes. Polly (bottom-right corner) is our fine-tuned Qwen model trained on the SCBE architecture. Add a free HuggingFace token in her settings panel and ask anything about the system.

Open Source — Coming Soon

Free small business tools. AI-powered. No subscription fees.

Payment processing, employee scheduling, and business management tools that don't charge you $50/month for basic operations. Open source, with a personal AI assistant that grows smarter over time through governed training updates.

Fast Payment Setup

Accept payments, send invoices, and track revenue without the processing bloat of enterprise platforms. Connect to Stripe, Ko-fi, or your own gateway. Zero monthly fees from us.

Employee Management

Scheduling, time tracking, basic HR workflows. Built for shops and small teams who don't need (or want to pay for) a full HCM suite. Local-first, your data stays yours.

Personal AI Assistant

Every installation gets an AI assistant that learns your business patterns over time. Governed by the same SCBE pipeline above — your data trains YOUR model, not a cloud vendor's. Updates on a schedule you control.

Built by someone who works at Wendy's and knows what it's like when tools cost more than they help. These will be free and open source — forever.

Follow on GitHub for launch Support on Ko-fi
AI Governance Benchmark

Train a model to predict ALLOW, QUARANTINE, or DENY.

We're releasing the governance decision dataset as a public benchmark. Train your own models to predict risk scores and governance decisions from the same data our pipeline produces. Launching on Kaggle.

What you'll train on

  • Real governance decisions from the 14-layer pipeline
  • Risk scores with hyperbolic distance + harmonic wall values
  • Actor types, resource classifications, intent signals
  • Adversarial drift patterns and benign baselines
🎯

Coming Soon

The benchmark dataset and Kaggle competition page are in preparation. Try the live demo above to see what the training data looks like.

Try the demo first
Get started

Ready? Pick your lane.

Toolkit for builders. Training Vault for model trainers. Free tools for small businesses. Or ask Polly first — she knows the system inside out.

Buy the Toolkit — $29 Buy the Training Vault — $29 Read the manual first