Build your trading strategy on evidence. 542,100 backtests behind every answer.
Tell it your market and style. It looks up which rules did best in historical backtests, shows how they did after costs and their worst drop, and gives you a strategy draft as rules or code to test yourself. No coding needed.
Or try it right here: 3 answers free, no account. Research statistics from historical tests, not investment advice; past results do not predict future returns.
Previous chats
Hi! I can show you which strategies did best in our backtests for your market, check an idea you have, or tell you if your backtest is just luck. What would you like to do?
Most popular trading ideas lost to simply holding the market. Only 1 in 3 beat it in our tests.
We built hundreds of popular trading ideas on data up to 2014 and judged them on 2015 to 2026, after trading costs. Two in three lost to doing nothing, so where you start matters.
beat simply holding the same market. The rest did worse than doing nothing.
Pick your market and style. See which rules did best in our backtests.
Three taps, and the chat above drafts a strategy from the historical backtests for that market, for you to test.
A strategy you can test today, with the numbers behind it.
- Exact entry, exit and filter, as they ran in the backtests
- On how many markets it worked, and whether it beat buy-and-hold
- Its worst drop, so you know what holding it feels like
- Ready code for TradingView, Python or MetaTrader
Already have an idea? Check it before you risk a cent.
Tap one and the answer appears in the chat above.
Does it work?
Ask about any indicator, market or timeframe and see how ideas like it did after costs.
Ask thisIs it better than holding?
See how often a popular rule beat simply buying and holding the same market.
Ask thisIs my backtest just luck?
See what luck alone produces after the number of variants you tried.
Ask thisWhat should I avoid?
Skip the kinds of ideas that failed again and again, before you spend time on them.
Ask thisGot a strategy from an AI? See if it beats luck before you trade it.
AI strategy tools try many settings and keep the best one. The more tries, the better a result looks by pure chance. Enter four numbers from your backtest and see what luck alone produces after that many tries, and how often winners like it stayed ahead on data they were not built on.
Free, no account. The answer appears in the chat above.
Why not just ask ChatGPT or YouTube? Because neither knows what happened next.
Each alternative misses one thing: how ideas like yours did on data they were not built on, after costs.
It tells you when nothing held up. So you do not spend weeks on dead ends.
- Costs included: every result is after trading costs, with the worst drop
- Every try counted: it says how many variants a strategy was picked from
- Luck check: it tells you when a result is within what chance produces
- Same answer for everyone: research, never a buy or sell call for you
No rule held up on EURUSD in our tests. Of 4,377 rule variants on daily currency charts, 3 had a positive typical result after costs.
The closest market with a rule that held up in our tests is US stocks. Want that one instead?
Start free. Upgrade when you want more answers and the large models.
Free account
sign up with your email
- About 10 answers a month in the chat
- Strategies as rules or as code
- No card, no password
- For your own AI: 20 checks a day by MCP
Pro
then $190 a year · cancel any time
Founding offer for the first 100 members
- 900 credits a month, about 1,500 answers
- The large models: Claude Sonnet and Opus, GPT Sol, Gemini Argon
- Every test behind each number
- For your own AI: 1,000 checks a day by MCP
Scale
billed $490 yearly
- 3,000 credits a month, about 5,000 answers
- The large models
- For your own AI: 10,000 checks a day, five URLs
- Everything in Pro
What people ask before they start.
Built by Justin Jaeger, who has spent years building and testing systematic strategies. How we test is in the methodology.
How does it build a strategy?
It does not just copy one backtest. It combines several findings: the rule type that was profitable most broadly in your market, the version of it that held up on the most symbols, and the market conditions (trend, volatility) in which that rule type did better or worse on 2023–2026 data kept untouched during research. It tells you which parts were tested together and which were not, as a starting point to test, not as a forecast.
Do I need to know how to code or backtest?
No. You ask in plain words and get the strategy as clear rules. If you want code, it writes it for TradingView, Python or MetaTrader, ready to paste.
How is this different from asking ChatGPT?
ChatGPT answers from what is written on the internet. This chat may only answer with our test results: how ideas did on data they were not built on, after costs. When nothing held up, it says so.
Will the strategy make money?
Nobody can promise that. Backtests look better than live trading, especially when a rule was picked from many, and the chat says so. That is why every strategy comes with a plan to test it on your own data before you use money.
Which markets does it cover?
US stocks, sector and commodity ETFs, the major currency pairs and the large cryptocurrencies, on 15-minute to daily charts. Index futures such as the DAX are not in the data; it then uses the closest market and says so.
Is this investment advice?
No. The chat shows how kinds of trading rules did in historical tests, after costs. It does not know your situation, never tells you to buy or sell, and never promises returns.
Does everyone get the same answers?
Yes. The same question gets the same research from the same tests, whoever asks. The chat does not ask about or consider your savings, goals or risk, so nothing is tailored to you personally. Strategies from the library are historical backtests: test them on new data before you trust them.
What does it cost, and can I cancel?
A free account gives about ten answers a month, without a card. Pro has 900 credits a month and the large models; the founding offer is $99 for the first year, after that $190 a year or $19 a month. Cancel any time, effective at the end of the period.
What happens to my questions?
We store your questions and the answers, without your IP address, for 12 months, to learn what people ask, how good the answers are and which tests to run next. We never sell data. Details are in the privacy policy.
I use Claude or ChatGPT. Can my own AI use this?
Yes. Connect BacktestAtlas as an MCP server and your AI checks the same evidence before it backtests: 20 checks a day free. Setup takes one minute on the start page.
Your next strategy, from evidence. Not from guesswork.
Start free with your email. No card, no password.