JudyAI WaveRider — AI Trading Agent

Hackathon submission for LabLab.ai "AI Trading Agents with ERC-8004"

Overview

AI-powered trading agent achieving 82.2% out-of-sample win rate across 8 pair-direction combos, validated through Walk-Forward Optimization (8 windows, IS=90d, OOS=30d, 366 OOS trades). Unlike typical backtest-only agents, every parameter is proven on unseen data — ensuring live performance matches validation results, not curve-fit parameters.

The agent combines three complementary strategies (WaveRider, BB Squeeze, MACD Divergence) with regime-adaptive routing, automated risk management, and ERC-8004 on-chain identity for verifiable reputation.

Built for the LabLab.ai AI Trading Agents with ERC-8004 hackathon. Agent identity registered on Sepolia following the ERC-8004 standard (Agent ID: 2328).

Strategies

Strategy	Description	Signal Type
WaveRider	EMA trend + RSI momentum + Volume confirmation	Primary (trend-following)
BB Squeeze	Bollinger Band compression → breakout detection	Secondary (mean-reversion)
MACD Divergence	Swing-point price-MACD divergence for reversal signals	Secondary (reversal)

Regime-Adaptive Routing

The agent automatically detects market conditions using ADX + Bollinger Band width:

Regime	Action	Position Sizing
Trending Up	Full long strategies	100%
Trending Down	Block longs, use shorts as exits	100% shorts
Ranging	Longs allowed, wider SL (1.5x ATR)	75%
High Volatility	Longs allowed, wider SL (1.8x ATR)	50%
EMA Convergence	Block all trading	0%

Risk Management

• Max 5% portfolio per trade (risk-based position sizing)
• 3% daily loss → automatic stop
• 10% total drawdown → emergency close all
• 5 consecutive losses → pause trading
• 3 consecutive losses → reduce position size to 50%
• Trailing stop: TP1 → breakeven, TP2 → tighten, TP3 → close

Architecture

┌─────────────────────────────────────────────┐
│ KrakenTradingAgent (main loop)              │
│  4H signal scan + 5min position monitor     │
├─────────────────────────────────────────────┤
│ KrakenDataAdapter                           │
│  OHLC + ticker via kraken CLI               │
├─────────────────────────────────────────────┤
│ StrategyEngine                              │
│  WaveRider + BB Squeeze + MACD Divergence   │
│  + Regime Detection (ADX/BB/EMA)            │
├─────────────────────────────────────────────┤
│ KrakenExecutor                              │
│  Paper buy/sell + SL/TP monitoring          │
├─────────────────────────────────────────────┤
│ RiskManager                                 │
│  Position sizing + daily/total limits       │
├─────────────────────────────────────────────┤
│ ERC-8004 Identity                           │
│  Agent Card + Sepolia registration          │
└─────────────────────────────────────────────┘

Prerequisites

• Python 3.11+
• Kraken CLI — kraken command must be available in PATH (install guide)
• numpy, pandas — pip install -r requirements.txt
• For ERC-8004 on-chain registration (optional):
  • cp .env.example .env and fill in your Sepolia wallet keys
  • Get Sepolia test ETH from Google Cloud Faucet or Chainlink Faucet
  • pip install web3 or install Foundry for cast CLI

Quick Start

# Install dependencies
make install          # or: pip install -r requirements.txt

# Run tests (63 tests)
make test             # or: pytest tests/ -v --tb=short

# Check status
make status           # or: python3 agent.py --status

# Run single scan (test)
make scan             # or: python3 agent.py --single-scan

# Dry run — see signals without executing
make dry-run          # or: python3 agent.py --dry-run

# Start main loop (4H scan + 5min monitor)
make run              # or: python3 agent.py

# Reset paper balance
make reset            # or: python3 agent.py --reset

# Generate ERC-8004 Agent Card
make card             # or: python3 erc8004.py --generate-card

# Register on Sepolia
make register         # or: python3 erc8004.py --register

Sample Output

$ python3 agent.py --single-scan
2026-03-28 22:30:05 | INFO    | Agent initialized | Balance: $100000.00 | Pairs: 6 | Scan interval: 4h
2026-03-28 22:30:05 | INFO    | === Full Scan #1 ===
2026-03-28 22:30:05 | INFO    | Portfolio value: $100000.00
2026-03-28 22:30:12 | INFO    | [BTC/USDT] Regime: TRENDING_UP | ADX: 28.5
2026-03-28 22:30:14 | INFO    | [ETH/USDT] Regime: RANGING | ADX: 18.2
2026-03-28 22:30:16 | INFO    | [SOL/USDT] WaveRider LONG | Entry: $186.42 | SL: $180.15 | TP1: $193.28
2026-03-28 22:30:18 | INFO    | Found 3 signals (2 long, 1 short)
2026-03-28 22:30:18 | INFO    | Executed 2/2 long signals

$ python3 agent.py --status
============================================================
  JudyAI WaveRider Trading Agent
============================================================

Paper Account:
  Starting Balance: $100,000.00
  Current Value:    $100,014.65
  Unrealized PnL:   +$14.65 (+0.01%)
  Total Trades:     3

Risk: 0/5 positions | Daily PnL: $0.00 | Scale: 100%
Positions: (none open)

$ python3 erc8004.py --show
{
  "name": "JudyAI WaveRider",
  "registrations": [{"agentId": 2328, ...}],
  "performance": {"oosWinRate": 82.2, "strategies": 3},
  "livePerformance": {"currentValue": 100014.65, "realizedPnl": 12.40}
}

Trading Pairs

Pair	Long	Short	OOS Win Rate	Notes
BTC/USDT	Blacklisted	WaveRider	75.4%	Long WFO OOS 40% — short only
ETH/USDT	WaveRider	WaveRider	93.3% / 97.8%
SOL/USDT	WaveRider	WaveRider	72.5% / 75.9%
XRP/USDT	WaveRider	WaveRider	78.7% / 75.5%
DOGE/USDT	Blacklisted	WaveRider	87.8%	Long WR 30.3% — degraded
ADA/USDT	Observe	Observe	N/A	No OOS data — monitoring only

Note: Kraken paper trading is spot-only. Short signals are used as exit indicators for open long positions rather than executed directly. Long positions are still opened during ranging/volatile regimes with reduced sizing and wider stop-losses to capture opportunities.

Validation

All strategy parameters are validated via Walk-Forward Optimization — no cherry-picked parameters:

• 360-day WFO period, 8 rolling windows (IS=90d, OOS=30d)
• Pass criteria: OOS Win Rate >= 65%, IS-OOS gap < 15%
• 366 out-of-sample trades across all combos
• Blacklisted pairs automatically excluded when OOS performance degrades

Risk-Adjusted Metrics (OOS data)

Metric	Value
Expectancy per trade	+0.61R
Profit Factor	3.79
Max OOS Drawdown	-8.7%
Recovery Factor	3.1

See docs/wfo_summary.md for full WFO methodology and per-pair results.

Files

File	Lines	Purpose
agent.py	463	Main agent loop (4H scan + 5min monitor, crash recovery)
strategies.py	607	Multi-strategy engine (WaveRider, BB Squeeze, MACD swing-point divergence)
executor.py	360	Kraken paper execution (immutable Position, trailing stop via replace())
risk_manager.py	182	Portfolio risk controls
kraken_data.py	152	Market data adapter
kraken_cli.py	36	Shared Kraken CLI wrapper (DRY)
erc8004.py	65	ERC-8004 CLI entry point + backward-compatible re-exports
erc8004_card.py	123	Agent Card generation, save/load, live performance
erc8004_chain.py	337	On-chain registration + reputation updates (web3.py / cast)
erc8004_abi.py	80	ABI fragments for Identity & Reputation contracts
config.py	93	Configuration (pairs, intervals, WFO parameters)
tests/test_core.py	576	48 unit tests: risk, indicators, strategy, SL/TP, ERC-8004, MACD swing
tests/test_integration.py	252	15 integration tests: data adapter, executor, position immutability
docs/wfo_summary.md	86	Walk-Forward Optimization results + risk-adjusted metrics
Dockerfile	17	Container image (Python 3.11-slim)
Makefile	34	Build/test/run targets
.env.example	—	Environment variable template

Technology

• Language: Python 3.11+
• Data: Kraken CLI (kraken ohlc/ticker)
• Execution: Kraken paper trading (kraken paper buy/sell)
• Indicators: numpy + pandas (no external TA library)
• Identity: ERC-8004 on Sepolia testnet (web3.py or Foundry cast)
• Container: Docker (python:3.11-slim)
• Tests: 63 tests (pytest) — 48 unit + 15 integration
• Source: Ported from judy-crypto (private, 12K+ lines)

Demo

# 1. Install + run tests
make install && make test

# 2. Check agent status and paper balance
make status

# 3. Run a single scan — shows signals and risk decisions
make scan

# 4. Generate ERC-8004 Agent Card
make card && make show

# 5. Register on Sepolia (needs test ETH in wallet)
make register

# 6. Start continuous loop (4H scans + 5min monitoring)
make run

ERC-8004 Integration

The agent registers an on-chain identity following the ERC-8004 standard and posts realized PnL on-chain after every trade closure:

• Identity Registry (0x8004A818...): ERC-721 NFT per agent — portable, transferable
• Reputation Registry (0x8004B663...): tradingYield feedback posted after each trade close (realized PnL, not unrealized)
• Agent Card: JSON metadata synced on every state save — live performance, risk controls, capabilities
• Runtime integration: agent.py imports erc8004.py and calls update_reputation() after SL/TP hits, short exits, and emergency closes

Live Registration

Field	Value
Agent ID	2328
Identifier	eip155:11155111:0x8004A818BFB912233c491871b3d84c89A494BD9e:2328
Chain	Sepolia (11155111)
Wallet	0x95C8B49C2A6124C436EA1a3f378991313f6f1c0A
Registration TX	0x348195e99e900014b2c48db4840dfed8b6693854444141b00f94c7c737fe93ac

After registration, the agent posts PnL snapshots on-chain as reputation updates, building a verifiable trading track record.

License

Proprietary — JudyAI Lab 2026