What if your ideas
could’ve tested
themselves

TextToQuant turns a trading idea, written the same way you’d describe it to a friend, into rules, runs those rules against years of real market data, and hands back a graded report: return, drawdown, win rate, expectancy, and whether the edge holds up under stress.

It was built by an actual hedge fund. This is the infrastructure they use to test and execute their own trades. It was not built to be sold; it was built to be used, and then opened up.

No. The AI does one job: it reads your sentence and turns it into a structured rule set. You see that rule set before anything runs. Every entry, exit, stop, size and filter is shown back to you as a parameter you can check and edit. If the parse got something wrong, you fix it before spending a run.

The backtest itself is not AI. It is a deterministic engine running your rules bar by bar on historical candles, simulating a trader sitting at the chart, checking every condition on every candle, and taking the trade only when all of them line up. With 100% accuracy, every time, for ten years of candles, without getting bored. Same rules, same data, same result. Nothing is invented along the way.

P.S. If you already live inside Claude, ChatGPT or any AI agent, add TextToQuant through MCP and give it this trader. The agent can open the terminal, read the report, run the test, tweak the rules and run it again, all through the same engine, with zero hallucination. It types the idea, the engine does the testing.

Every backtest gets a score out of 100 and a letter. The score comes from four parts: profit (did it make money after costs), risk (how deep and how long the drawdowns were), consistency (was the return spread across time or did one month carry it), and edge (expectancy per trade, profit factor, payoff).

The grade is not a promise. It is a summary of what the history showed. Fewer than 30 trades and the grade is flagged: not enough evidence to judge. A high grade on a thin sample means little, and the report says so.

Every number is shown with its working. Click any metric in the performance header and the inputs and outputs behind it are plotted on the chart: which trades, which candles, which formula. Nothing is a black box. Check it yourself.

Fees and slippage are line items in every report. If a run has no fees, no stop, or too few trades, an honesty flag says so on the result, not in the small print.

A good backtest on its own does not mean it works live. That is what the robustness suite on Power and above is for: a parameter sweep shows whether the edge survives small changes to your settings, walk-forward tests it on data it never saw, Monte Carlo (from Pro) reshuffles the trades to show the range of outcomes, and the overfit verdict tells you if the result looks curve-fit. If a strategy passes those, you have a reason to trust it. If it does not, you found out for the price of a backtest instead of a live account.

Yes. Your strategies and results are private by default. Nobody sees them unless you share a link or publish to the gallery yourself.

Because the tool is not the edge. The idea is. TextToQuant does not generate strategies; it tests yours. It’s a test bench. Think of it like a sword: hand it to two different warriors and you get very different results.

We do trade with it, every day. Selling it costs us nothing, because our edge was never in the engine; it’s in what we run through it. We’re cool people, so instead of locking it away we opened it up. It’s a second income for the fund, and it’s the thing we wish someone had handed us years ago. Infrastructure like this is expensive, slow and ugly to build, and that’s exactly what keeps funds and prop firms a level above everyone else. Consider that level yours now. You’re welcome.

Yes. If your version of backtesting has been scrolling back on a chart and marking up where the trade would have gone, this is the first time you’ll see the real number. Type the idea the way you would explain it to a friend, and the tool does the rest.

A new account gets five backtests free, no card. They are a one time trial for accounts that have never had a paid plan, not something that comes back after a paid plan ends. Most people learn more from their first result than from the last three courses they bought.

You will understand it. The report leads with the grade and four plain words: profit, risk, consistency, edge. The equity curve is a line going up or down. Every metric has a one-line explanation next to it, and the honesty flags are written in English, not in Greek letters.

The deeper numbers (Sharpe, Sortino, Kelly, omega) are there for the people who want them. You do not need them to read the verdict. And if you want to learn them, hover any metric and it explains itself, with the inputs and outputs of the calculation shown live. There’s a full tutorial too.

No code needed. Plain English in, report out.

If you do code, you can bring it. Custom indicators can be saved in Pine on every plan, and in Python or JavaScript from Pro, and used inside your strategies by name, in whatever way fits you.

You can set up alerts and get a Telegram message or an email when a trade fires, on any asset in the crypto universe. Automated execution is coming soon.

Yes. Every result is a shareable page with the full report, the grade breakdown and the honesty flags. On Power and above it also carries the full robustness suite you ran (walk-forward, parameter sweep, Monte Carlo, overfit verdict), and you can export a PDF tearsheet.

What a desk respects is not the return number, it is the validation behind it. With the full suite on the page, you can walk them through your strategy and explain any metric the way a fund would expect to hear it. That says more about how you think than a screenshot of a P&L.

Crypto on every plan. Stocks, forex and metals, US and international, from Power. Timeframes from 1 minute to 1 month: Free and Pro run 1 hour to 1 week, Power runs 1 minute to 1 week, and Quant and Enterprise run 1 minute to 1 month.

Crypto has the full history of the pair on every timeframe: Bitcoin runs back to the day it listed. Stocks reach back a fixed number of bars, so the window shrinks with the timeframe: about fourteen years on daily bars, seven on 4-hour, two on 1-hour, and nine trading days on 1-minute. Forex and metals trade around the clock five days a week, so the same number of bars reaches less on intraday charts: about seven months on 1 hour bars and two trading days on 1 minute bars. A run that asks for more is told the earliest date before it starts, so nothing is charged.

Pro and above put no cap on the years; Free runs the last three. The only other cost is time: the bigger the window, the longer the load.

MCP is a standard that lets AI agents (Claude, ChatGPT, Grok and others) use outside tools. TextToQuant runs as an MCP server, so you hand your agent the keys to the terminal. It follows your instructions, runs any command you could, tests strategies in bulk, reads the reports and tweaks the rules. This is the only way to get AI strategy testing with zero hallucination in execution: the agent types, the engine does the work.

You do not need it. The website does everything. Your plan limits apply the same way in both places.