This is a walkthrough of building, testing, and deploying a momentum scalping strategy on Hyperliquid using COD3X. Not a hypothetical — a specific configuration with specific parameters and results.
The Strategy#
Momentum scalping on 15-minute candles. The thesis: when a market is trending strongly (high ADX), pullbacks to the trend's moving average are high-probability entries. Enter on the pullback, ride the continuation, exit at a predefined target.
Entry Conditions#
All four must be true:
- ADX above 25 — Confirms a strong trend exists. Below 25, the market is choppy and momentum entries fail.
- RSI pulling back — RSI between 40-55 (in an uptrend) or 45-60 (in a downtrend). Not oversold/overbought — just pulling back within the trend.
- Price at EMA — Price touches or comes within 0.3% of the 21-period EMA. This is the "pullback to the mean" that gives a tight entry.
- Volume confirmation — Current volume above the 20-period volume moving average. No entries on thin volume.
Exit Conditions#
- Take-profit: 2:1 reward-to-risk ratio. If stop-loss is 1.5%, take-profit is 3%.
- Stop-loss: Below the most recent swing low (for longs) or above the most recent swing high (for shorts). Typically 1-2% from entry.
- Time stop: Close any position open longer than 4 hours. Momentum scalps should resolve quickly — if they haven't, the thesis was wrong.
The Configuration#
The Goal Prompt#
Written in plain English — no code:
Scan BTC-PERP, ETH-PERP, and SOL-PERP on the 15-minute timeframe. Look for momentum pullback setups where ADX is above 25, RSI is pulling back to the 40-60 zone, and price is touching the 21 EMA. Volume must be above the 20-period average. Enter in the direction of the trend with 3x leverage, 5% position size. Set stop-loss below the recent swing low (longs) or above the recent swing high (shorts). Take-profit at 2:1 reward-to-risk. Close any position open longer than 4 hours regardless of P&L.
That's it. The AI model interprets this goal, evaluates market conditions using Chain-of-Thought reasoning, and decides whether a valid setup exists.
Why Chain-of-Thought#
CoT was chosen over the other reasoning modes for a specific reason:
- Direct mode would execute without analyzing the setup — too trigger-happy for a strategy that requires multiple confirmations
- Graph-of-Thought would evaluate multiple strategies in parallel — unnecessary here since we have one strategy with clear rules
- ReAct would loop through observe-reason-act cycles — overkill for a strategy that can be fully evaluated in one pass
CoT gives a single, thorough analysis: here's what ADX reads, here's where RSI is, here's the EMA distance, here's the volume profile, here's my decision. One pass, fully explained, moderate credit cost.
Backtest Results#
Three-month backtest on historical Hyperliquid data (September–November 2025):
Breakdown by Market#
BTC had the cleanest momentum setups — likely because it trends more consistently on the 15m timeframe. SOL had more false signals during volatile meme-coin-correlated moves.
Lessons Learned#
ADX is the filter. Without the ADX > 25 requirement, the strategy fires in choppy markets and the win rate drops to ~48%. The ADX filter alone improved profit factor from 1.1 to 1.8.
The time stop matters. Positions that didn't hit take-profit within 4 hours had a 35% win rate — meaning the initial thesis was wrong. Cutting them early preserved capital for the next setup.
Three markets is enough. Adding more pairs diluted the strategy's effectiveness. BTC, ETH, and SOL have enough momentum setups to keep the agent busy without forcing trades in illiquid markets.
15m is the sweet spot. 5m had too much noise (45% win rate). 1h had too few setups (only 40 trades in three months). 15m balanced signal quality with trade frequency.
From Backtest to Live#
The same goal, same parameters, same AI model from the backtest are deployed live. No translation layer. The agent runs the same logic it was tested on, with the same risk parameters.
After the first month live: 8 trades, 5 winners, consistent with the backtest's 58% win rate. The strategy is performing within expected parameters.
A momentum scalping agent: ADX filter, EMA pullback entry, 2:1 reward-to-risk, 58% win rate across three months. Built, tested, and deployed in under an hour.
