GeoHorizon Protocol
Confidential · V1
GeoHorizon is building the hedge fund that everyone can use — institutional-grade capital protection and AI-powered hedging, without the $10M minimums. A four-layer pre-trade risk gate and hard portfolio constraints enforce discipline before any capital is committed. Options-based downside protection activates in crash regimes. A closed learning loop sharpens every risk parameter after every cycle.
This is not a signal service or a model portfolio. It is a risk architecture — the kind institutional funds use to protect capital across regimes. Regime-strategy compatibility, volatility surface gating, concentration limits, and stress-tested sizing enforced in sequence, before every position, across all five strategy types under a single World Model.
Why GeoHorizon Is Different
The AI-driven markets space is attracting significant capital for good reason — hundreds of billions are at stake in portfolios that remain exposed during regime shifts and correlation breakdowns. Most projects focus primarily on signal generation and alpha within traditional equity markets.
GeoHorizon is built around a more critical and less addressed problem: protecting capital when markets break, while still competing on intelligence and alpha generation.
Hard Portfolio Constraints Before Execution
Capital is deployed with discipline from day one — and the constraints governing that deployment improve in precision as the system accumulates portfolio outcomes.
Regime-Aware Defensive Positioning
Defensive posture is adopted earlier and with greater accuracy over time, because regime detection sharpens continuously across all five strategy types.
Explicit Failure Memory (Anti-TIMG)
The same conditions that caused a costly error will trigger a warning the next time. Protection improves and so do returns.
Unified Cross-Asset Intelligence Loop
All five strategy types feed one closed learning system. Every portfolio outcome improves regime detection, strategy selection, and sizing.
Superior Compounding Intelligence
Decision quality compounds faster than siloed alpha systems because the architecture learns across all strategies simultaneously.
In short: GeoHorizon solves the hardest part of portfolio management — surviving regime shifts and correlation breakdowns — while building one of the most sophisticated cross-asset intelligence systems in the space.
Live Platform Metrics
Refreshes every 60s
700+
Simulation Cycles
283 crypto + 424 markets swarm
87.0%
Directional Accuracy
Live backtested forecasts
48K+
Agent Memories
Cross-asset TIMG trajectories
5
Strategy Types Active
Leverage · Scalp · Options · Equity · Spot/On-Chain
Hedging Architecture
Five Strategy Types. One Capital Protection Mandate.
Crypto Leverage
Funding-rate-aware long/short perpetuals, delta-neutral basis trades. Active in Risk_On / Neutral regimes. Funding costs charged proportionally every 15 minutes.
Scalping (15–30 min)
Microstructure entries via RSI, Fibonacci confluence, and MACD crossover. Regime-gated to Risk_On / Neutral only. ATR dynamic stops 1.5–2.5× by vol regime.
Equity / ETF Leverage
Directional SPY, QQQ, NVDA, sector ETF leverage positions. Half-Kelly sizing × drawdown tier. 4h + 24h forecast agreement required.
Options Strategies
8-strategy library: iron condors, credit spreads, straddles, tail hedges, covered calls, protective puts. Each leg priced from live per-strike IV. CRASH_FEAR blocks premium-selling.
Spot + On-Chain Hedging
Spot BTC/ETH/SOL positions with on-chain hedge overlays via DeFi protocols. Hedge overlays activate in CRASH_FEAR and HIGH_VOL as a protective layer over existing exposure. Liquidity-weighted sizing accounts for on-chain execution costs and slippage. Supported with increasing depth as on-chain execution routes expand.
All five types share one mandatory pre-trade gate: (1) Regime-Strategy Performance Matrix + Thompson Sampling bandit — underperformers auto down-weighted; (2) vol surface regime check — CRASH_FEAR surface vetoes premium-selling regardless of IV rank; (3) Portfolio Construction Agent hard constraints — C1–C4; (4) pre-trade stress overlay — position size tested against 2008, 2020, 2022, and 2025 shock scenarios. No strategy bypasses this sequence.
Defensibility
Four Differentiators That Compound Over Time
Portfolio Construction — Not Signal Generation
Four hard pre-trade constraints enforce discipline before any position opens across all five strategy types — including spot and on-chain exposure: single-asset concentration (≤40%), correlated group exposure (≤60%), strategy family weight (≤50%), and net portfolio beta (≤1.5×). Five continuous macro signals inject independently of regime classification. This is institutional pre-trade risk architecture — not a signal feed with no guardrails.
Regime + Vol Surface Awareness
An 8-state Bayesian Hidden Markov Model classifies the macro regime every cycle. Every strategy is checked against both the regime gate and a live volatility surface check before execution. CRASH_FEAR surface vetoes premium-selling regardless of IV rank. Per-strike implied vol prices every options leg — not a single ATM estimate. These are independent blocking gates, not soft adjustments.
Learning from Failure — Anti-TIMG
The system captures high-confidence wrong predictions (probability ≥60%, direction incorrect) as Anti-TIMG warning trajectories. Before the next similar forecast on the same asset class and regime, these failures inject explicitly into the Forecaster context. The closed feedback loop learns from overconfident mistakes — not just from successes. 48K+ trajectories and compounding every 8 hours.
Automatic Risk Reallocation — No Manual Overrides
A multi-armed bandit (Thompson Sampling) continuously calibrates which strategy types are working in the current regime. Strategies that underperform in CRASH_FEAR are automatically down-weighted before they accumulate losses — no manual rule changes, no human latency. Every (regime, strategy family) pair maintains a live risk-adjusted win rate, updated after every scored outcome. Capital is automatically reallocated away from what is failing and toward what the current regime rewards.
GEO Token
Value Tied to Risk-Adjusted Performance
The GEO token is not an access key or speculative instrument. Its value is anchored to demonstrated capital protection and trading performance. Primary utility is revenue participation. Secondary utility is protocol and treasury governance — capital allocation, fee structure within hard-coded ceilings, and protocol upgrades. Live risk parameters remain under the World Model, hard-coded safety rails, and multi-sig oversight.
Primary — Performance-Driven Revenue
60% of all protocol revenue distributed to GEO stakers in USDC every 7 days. Revenue is earned when the system demonstrably protects capital — performance fees scale with risk-adjusted returns, not trading volume. Stakers earn real yield because the protocol earns it first. No inflationary emissions.
Protocol & Treasury Governance
GEO holders vote on treasury allocation (R&D, security audits, liquidity, insurance reserves, marketing, buybacks), fee structure changes within hard upper limits governance cannot raise, prioritisation of new markets and strategy families, insurance fund parameters and claims process, major upgrades and new module proposals, and emergency pause and recovery mechanisms. Live risk parameters — strategy gates, position sizing, drawdown halts, concentration limits — stay under the World Model, hard-coded safety rails, and multi-sig oversight. Quadratic voting prevents whale capture.
Fee Reduction + Priority Access
1,000 GEO: 50% fee reduction across all five strategy types. 10,000 GEO: full fee waiver plus premium tier — complete regime history, per-agent performance attribution, and strategy breakdown by market condition.
Simulation → Live Execution
The current simulation book demonstrates the full trading and risk infrastructure at scale — the same engine that will earn protocol fees once live capital is deployed. Revenue potential scales meaningfully in Phase 3: performance fees activate on managed capital, execution fees compound with volume, and subscription revenue grows with the user base. The infrastructure that would generate those fees is already operational and Brier-scored against real outcomes.
Current Epoch
#84
Next Distribution
6d 08h
Every Monday · USDC on-chain
1B
Total Supply
Fixed cap — no minting post-launch
60%
Revenue to Stakers
Performance + execution fees + subscriptions
7 days
Epoch Duration
Paid in USDC, on-chain, every Monday
Current Status & Roadmap
Phase 1 & 2 Live. Phase 3 on Deck.
Phase 1 & 2
Live- Dual-swarm architecture: crypto + markets running 24/7 under shared World Model
- Five active strategy types: crypto leverage, scalping, equity/ETF leverage, options, and spot plus on-chain hedging (on-chain depth increasing)
- 700+ live simulation cycles at 87.0% directional accuracy
- 48K+ TIMG trajectories; Anti-TIMG failure learning active
- Regime-Strategy Performance Matrix + Thompson Sampling bandit
- Strategy Lab: autonomous shadow testing, walk-forward validation, self-promotion
- Position-level protective stops (−8% alert, −12% close-only, −20% hard close)
- Task-aware ModelRouter across six models and three providers: Haiku 4.5 (high-frequency calls), Sonnet 4.6 (macro synthesis), Opus 5 (crisis-regime inference), Fable 5 (narrative synthesis), Gemini Flash (on-chain + prediction markets), Grok (real-time social signal)
- Haiku sanity check on every World Model synthesis; Reflection Ensemble gate (≥90% agreement boosts conviction, <70% forces neutral posture)
Phase 3
Upcoming- GEO token launch + staking contracts (quadratic governance)
- On-chain revenue distribution (7-day USDC epochs)
- Smart contract execution layer (1-click hedge approval)
- MEV protection via private mempool routing
- Cross-chain: Base, Arbitrum, Solana + broker integration (Alpaca live execution)
Phase 4
Upcoming- Institutional white-label API for family offices and hedge funds
- Hyperliquid stock perps — on-chain equity exposure without a broker
- Multi-asset expansion: FX, commodities, interest rates
- Federated learning across anonymised user portfolios
- Q-learning calibration for RL-based agent parameter tuning
Get Started
View the Live Dashboard or Read the Full Technical Whitepaper
All simulation results are live, Brier-scored against real price outcomes, and verifiable. No cherry-picked backtests, no synthetic data.
This document is for informational purposes only and does not constitute an offer to sell or a solicitation to buy any securities or tokens. All performance figures are from paper/simulation trading. Past simulated performance is not indicative of future results. GEO token has not launched. See the full whitepaper for risk disclosures.