Strategy Tester · beta

Test a trading idea in seconds, without fooling yourself

Write the rules, or ask your AI for ten ideas and paste its answer. Each one is backtested on your own bars in a few seconds, with honest fills and costs, and judged by the same six checks as our published tests.

Runs in your browser; your data never leaves itEvery idea you try is counted, and the bar rises with itThe last 30% of your data stays locked until you chooseFree while in beta

1. Price data

Bars as CSV: MetaTrader 5 (View → Symbols → Bars → choose the symbol and timeframe → Request → Export Bars), TradingView (chart menu → Export chart data), a Binance archive file, or any CSV with time, open, high, low and close. Your broker's own bars are the best test of how a strategy trades there.

Clock of the times in the file MT5 writes broker time. Files with Unix or ISO times carry their own zone and skip this.

Choose a bar file

2. The idea

Start from an example Cost per trade, round trip, in price pointsSpread + commission + slippage. Nasdaq CFD: often 1–2 points; EUR/USD: about 0.0001 plus commission. …plus a percentage of the priceExchange fees: Binance spot is 0.1% a side, so 0.2 round trip.

Rules (JSON)

Let AI write the ideas

With your own AI (ChatGPT, Claude, Gemini…), free and without a limit:

Copy the prompt. It explains the format and lists the ideas already tested on this data, so the AI does not repeat them.
Paste it into your AI and ask it to go. It answers with a block of JSON.
Paste the whole answer here and test them all. Then copy the prompt again for the next round.

Your AI's answer

The rule language

Why this tester says “no” more often than others

Every idea counts. Try 40 ideas and the best one looks good by luck alone. The tester counts every different set of rules you run on the same data and raises the multiple-testing bar with it. Renaming an idea does not reset the count.
A vault. The last 30% of your bars are locked. All tests use the first 70%. When you have one idea you believe in, open the vault for it: a single test on data none of your choices touched. Opening it again for another idea is allowed, but the report says the vault is no longer clean.
Fills that do not flatter. Signals use the bar that has closed and trade in the next one. If a bar touches both the stop and the target, the stop counts. A limit order needs the price to trade through it. A gap through the stop fills at the open. Costs come off every trade.
The same six checks as the Backtest Health Check and our own methodology: the bar for the number of ideas, double costs, year by year, drawdown, a prop-firm challenge, and the last quarter of trades.

Most ideas fail. That is the point: finding out in ten seconds is cheaper than finding out in a funded challenge. A pass here is a reason to forward-test, not to trade.