🐺 Sniper Armada v21.0 Omni-Interface

Sovereign HFT Trading System — Pure Rust Hive with Omni-Router and in-process AI inference.

---

⚡ Key Features

• Omni-Router (Dual-Exchange) — Native Binance and Bitfinex support running concurrently in L0 hot paths.

• The War Room Dashboard — 10Hz HTML5 Canvas SSE live data stream displaying OBI tachometers and VWS (Volume Weighted Spread).

• 1088B Zero-Copy MMap Architecture (ArmadaStateV2) — Ultra-fast IPC telemetry memory mapping avoiding kernel overhead.

• Sub-millisecond execution — Zero-allocation Rust hot paths with fixed-point arithmetic (i64 × 1e8)

• 5-bot fleet — Hydra (MM), Moonshot (flash dip), Grid (grid maker), Trigon (tri-arb), Nexus (cross-venue arb)

• Dual AI inference — In-process Candle Logit Sniping (<20ms) + Gemini 3.1 Pro strategic oracle via ZeroClaw

• Hexagonal Architecture — VenueAdapter trait enables zero-effort exchange integration (Bitfinex, Binance, ...)

• Cognitive Engineering — Few-Shot anchored L1 prompt + Sovereign Constitution L2 with Chain-of-Thought forcing

• Brain Transplant Protocol — Safe hot-swap GGUF models with shadow mode validation, zero downtime

• Autonomous operation — Sovereign Boot Protocol (v20.0 Strategy Pattern) with 5 RiskClass Pillars (HEDGE, ARBITRAGE, MARKET_MAKER, STAT_ARB, POSITIONAL), quota watchdog, Sentinel guardian

• Lock-free IPC — mmap with SeqLock atomics, 64-byte cache-line aligned structs via /dev/shm/beroun/

🔧 V21.0 Omni-Interface Architecture Details

• The War Room (V2 Omni-Interface Dashboard): A modern HTML5 Canvas-based tactical view streaming live telemetry at 10Hz via Server-Sent Events (SSE). Replaces static port-based views with zero-lag OBI (Order Book Imbalance) tachometers and Volume Weighted Spread depth-charts.
• ArmadaStateV2 (1088B Zero-Copy MMap): Highly optimized memory mapping protocol for sub-millisecond data sharing. Layout eliminates kernel syscall bottlenecks, enabling lock-free state synchronization between Python orchestrators and Rust HFT threads.
• Cross-Exchange Arbitrage (VWS): Omni-Router natively integrates Bitfinex and Binance under a Volume Weighted Spread mechanism. Nexus analyzes true depth parity before executing latency-padded atomic cross-venue fills.
• Autonomous ML Injector & Hot-Swap: Safe, lock-free Candle GGUF tensor weight deployments, allowing zero-downtime model swaps dynamically informed by L2 regime context.

🏗️ Architecture

┌─────────────────────┐     ┌─────────────────────────┐     ┌──────────────────────────┐
│   L0: MUSCLE         │     │   L1: REFLEXES           │     │   L2: GENERAL             │
│   (Rust Execution)   │     │   (Candle + Cortex)      │     │   (ZeroClaw + Gemini)     │
│                     │     │                         │     │                          │
│  Hydra  ──┐         │     │  Candle L1 Brain        │     │  Sovereign Constitution  │
│  Grid   ──┤ mmap ◀──┼─────┤  Logit Sniping (<20ms)  │     │  Gemini 3.1 Pro Preview  │
│  Moonshot─┤         │     │  4-token: BID/ASK/      │     │  <thinking> CoT forcing  │
│  Trigon ──┤         │     │    HOLD/KILL             │     │  RAG memory (SQLite)     │
│  Nexus  ──┘         │     │  CPU mode (GTX 1060)    │     │  Telegram alerts         │
│                     │     │                         │     │  Quota Watchdog (5min)   │
│  VenueAdapter trait  │     │                         │     │                          │
│  Bitfinex + Binance  │     │                         │     │  Hard Guardrails:        │
│                     │     │  ChatTemplate Resolver   │     │  FAIL-FAST / MAX EXPO /  │
│                     │     │  (Phi3/Llama3/Qwen2/    │     │  ANTI-JITTER             │
│                     │     │   Gemma2/Mistral)       │     │                          │
└─────────────────────┘     └─────────────────────────┘     └──────────────────────────┘
        <1ms                        <20ms                          ~5min cycle

🚀 Quick Start

# Build all bots + AI
cargo build --release

# Deploy fleet
./deploy_armada.sh --build

# Start ZeroClaw L2 Oracle
export GEMINI_API_KEY=your_key
zeroclaw daemon

# Hot restart via systemd
sudo systemctl restart sniper-armada

# Status
zeroclaw status

📦 Repository Structure

sniper/
├── hydra/              # L0 Market Maker (Rust — Bitfinex WS)
├── moonshot/           # L0 Flash Crash Dip Buyer (Rust — multi-pair)
├── grid/               # L0 Dynamic Grid Maker (Rust — L2 warped)
├── trigon/             # L0 Triangular Arbitrage (Rust — 14 triangles)
├── nexus/              # L0 Cross-Exchange Arb (Rust — Binance↔Bitfinex)
├── shared/             # sniper-shared crate (types, IPC, framework, venues)
│   └── src/exchange/   #   VenueAdapter trait + BitfinexVenue + BinanceVenue
├── candle-brain/       # 🧠 In-process L1 AI inference (Candle + GGUF)
│   ├── src/brain.rs    #   CandleL1Brain — GGUF model loading + inference
│   ├── src/logit_sniper.rs  # Logit Sniping — 4-token classifier
│   ├── src/chat_template.rs # Model-agnostic template resolver
│   └── src/transplant.rs    # SBP v3.1 Brain Transplant Protocol
├── architect/          # L1 Cortex (Rust) + L2 Oracle + Commander
│   ├── cortex/         #   Sovereign Cortex daemon (Sentinel, GPU, UDS)
│   ├── l2_oracle.py    #   Legacy L2 brain (replaced by ZeroClaw)
│   └── tg_commander.py #   Telegram interface + bot management
├── zeroclaw/           # 🐝 ZeroClaw L2 Orchestrator
│   ├── config.toml     #   Gemini 3.1 Pro, SQLite RAG, Telegram
│   ├── system_prompt.md#   Sovereign Constitution (CRO persona)
│   └── tools/          #   mmap tools (read_engine_state, write_risk_param, quota_watchdog)
├── infra/              # systemd service, install/stop scripts
├── state/              # Persistent pre-crash state (armada_state.json)
├── deploy_armada.sh    # Sovereign Boot Script
├── watchdog.sh         # Cron-based process monitor
├── ARCHITECTURE.md     # Full system documentation
└── README.md           # This file

🧠 AI Layers

L1 — Logit Sniping (Candle, In-Process)

ROLE: Ultra-low latency HFT Reflex Core.
TASK: Classify L2 microstructure to predict next 100ms price tick.
OUTPUT: EXACTLY ONE TOKEN FROM [BID, ASK, HOLD, KILL].

• No text generation — reads raw logits for 4 tokens in <20ms
• Model-agnostic — ChatTemplate auto-detects Phi-3.5, Llama-3, Qwen2, Gemma2, Mistral
• Few-Shot anchored — 3 in-context examples force deterministic classification
• KILL action — immediate order cancellation on toxic sweep detection

L2 — Sovereign Constitution (ZeroClaw + Gemini 3.1 Pro)

• Chain-of-Thought forcing — <thinking> block required before any Tool Call
• Hard Guardrails — FAIL-FAST, MAX EXPOSURE (0.05 BTC), ANTI-JITTER
• RAG Memory — SQLite vector DB recalls past decisions and their outcomes
• Brain Transplant — safe hot-swap of GGUF models with shadow validation

🛡️ Safety Guarantees

🔧 Production Setup

# Prerequisites
sudo apt install -y nvidia-cuda-toolkit   # For future GPU acceleration

# Install systemd service + kernel tuning
sudo ./infra/install_service.sh

# Configure API keys
cp .env.example .env
# Edit: BITFINEX_API_KEY, BINANCE_API_KEY, TELEGRAM_BOT_TOKEN, GEMINI_API_KEY

# ZeroClaw onboarding
zeroclaw onboard --quick --provider gemini --model gemini-2.5-pro

# Verify
sudo systemctl status sniper-armada
zeroclaw status

📚 Guides

• Add a New Bot — Step-by-step bot creation with SovereignEngine
• Add a New Exchange — VenueAdapter implementation guide

📄 License

Private. All rights reserved.