Crypto strategy backtester

AI backtesting in plain English, with the code in view

Say what you want to trade, on which market and timeframe. The builder answers with the rules it understood, the numbers you can tune, and the TypeScript that will run. If a rule reads wrong, you see it before a single candle is fetched, and you fix it by saying so.

import { RSI } from "technicalindicators";

export function onCandle(candles, portfolio, parameters) {
  const { rsiPeriod, oversold, exitLevel } = parameters;
  const closes = candles.map((c) => c.close);
  const price = closes[closes.length - 1];
  const rsi = RSI.calculate({ period: rsiPeriod, values: closes });
  const [prev, now] = rsi.slice(-2);
  const holding = portfolio.positions.length > 0;

  let action = "HOLD";
  if (!holding && prev >= oversold && now < oversold) action = "BUY";
  if (holding && prev <= exitLevel && now > exitLevel) action = "SELL";

  return {
    action, size: 1, price, orderType: "MARKET",
    stopLoss: price * (1 - parameters.stopLossPercent / 100),
    takeProfit: 0, confidence: 0.6,
    reasoning: `RSI ${now.toFixed(1)}`,
  };
}

Results you can check, trade by trade

Every run streams its log as it goes, then opens on one plain sentence about what happened. Behind it: return, drawdown, win rate against the rate needed to break even, profit factor, Sharpe, Sortino and Calmar, a grade out of 100, and every trade with the reason it closed. Move a slider, run again, and the new result is compared with the last.

Fees, slippage and stops, counted on every fill

A backtest is only worth what its fills are worth. The engine charges for every trade, and where it can't know what happened inside a candle it assumes the worse fill.

Fees and slippage on every fill

Each side of a trade costs 0.085%: 0.05% slippage plus a 0.035% taker fee. A strategy that only works for free shows up as one that doesn't.

Stops, gaps and liquidation

A stop or target fills at its level, or at the open when a candle gaps past it. Above 1x, positions are isolated margin and close at their liquidation price.

A drawdown kill switch

Set a maximum drawdown. Equity is checked at each candle's low, and when it falls through the limit every position closes and the run stops trading.

The run sizes the trades

The strategy decides when to enter and exit. Each entry is sized at the strategy's position size as margin, times the run's leverage. A strategy that sets its own size is told so in the log.

Conservative on purpose

Stops and targets are tightened by 15% and numeric parameters vary by ±1.5% per candle, so an edge has to survive small changes. Anything still open on the last candle closes there, costs included.

Checked code, fresh candles

Strategy code is validated, then run in an isolated sandbox with a time limit on every candle and on the whole run. Candles are read from GoldRush HyperCore on your key each run.

What traders test with it

• Mean reversion on SOL

  Buy oversold dips on SOL-PERP and sell the bounce, then slide the RSI levels and the stop to see how fragile the edge is.

• Trend following on BTC and ETH

  Moving-average crosses and breakouts on 4h or daily candles, with up to two years of history on the daily chart.

• Port a TradingView strategy

  Paste the Pine Script you already trust and test it on Hyperliquid prices, with Hyperliquid-like costs, before you trade it there.

• Ask why it worked

  Ask the analyst about a finished run, or tell the builder what to change and run the new version against the last one.

Start free

Strategy Backtest reads candles on your own GoldRush API key, the same key that serves the Hyperliquid API and 100+ other chains. Every account starts with a free trial.