How to choose a trading bot: the checklist to demand from any seller

Give me an afternoon, historical data and no scruples and I'll hand you an equity curve that climbs in a near-straight line for seven years. It will be worthless, and it will look better than anything honest you have ever seen. That is the problem with buying a trading system: the marketing is cheap to fake and the product is invisible.

So you don't evaluate the pitch. You evaluate what the seller is willing to put in writing and what you can check on your own machine. Here is the list.

Start from "it doesn't work"

The default assumption for any automated system, from anybody, should be that it does not work until specific evidence moves you. That isn't cynicism, it's arithmetic: thousands of these are sold, a small number are built on anything real, and the ones that aren't look exactly like the ones that are from the outside. If you start at "prove it," bad products fail fast and good ones have no trouble.

The other half is knowing what you're not entitled to. You are not entitled to a guarantee, and any seller offering one has just told you something about themselves. What you are entitled to is enough information to decide for yourself.

The checklist

1. A backtest you can rerun yourself. Not a screenshot, not a video, not a link to a third-party dashboard you can't audit. The strategy file, a stated period, a stated instrument, and a standard tool (the Strategy Analyzer in NinjaTrader 8, for instance) so you can press Run and compare. If the only proof is an image, it isn't proof.
2. The losing periods shown, not just the winning ones. A month-by-month table with red rows in it is worth more than any headline number, because it tells you what a normal bad patch looks like before you're inside one. A seller who only shows the good months is selling a highlight reel.
3. A defined instrument and a defined trading window. "It trades the markets" is not an answer. Which contract, on which exchange, at what times. A system that claims to run 24 hours across every asset class is either several systems wearing a trench coat or none at all. The mechanism is written out in how it works.
4. Real stop and target orders, sent to the market. Ask directly: are the stop and target working orders resting at the exchange, or does the software hold them internally and act when the price is touched? The difference matters the day your internet drops mid-trade. Anything worth buying places both in the market at the instant of entry.
5. An answer to "how will I know if it stops working." Every system eventually degrades. The seller should be able to describe what abnormal looks like: how many consecutive losses still fall inside historical behavior, what a drawdown past the backtest's worst would mean, what to do about it. If nobody has thought about that, you're the risk management and nobody told you.
6. A refund policy you can read before you pay. Software that can be copied is hard to refund and most sellers say so. That's fine, as long as they say it up front and in plain language. What isn't fine is a policy that appears only after the charge goes through, or one written to be impossible to satisfy.
7. A support channel that leaves a record. A ticket with a reference number and a written history beats any chat window. If the only way to reach a seller is a messaging app where the conversation can be deleted, you have no record of what you were promised. Here, the contact form is the only door in, and every message becomes a ticket.

Answers that should end the conversation

The full catalog of tricks is in how to spot a trading bot scam, and the specific case of systems marketed as artificial intelligence is in AI trading: what is real.

Now run the list on us

The awkward part: that record is a backtest, not an audited live account. It is a simulation over historical data, prepared with the benefit of hindsight, with no real money at risk. It cannot fully capture execution, slippage or liquidity on a real fill, and past performance, real or simulated, does not guarantee future results. If your checklist requires a third-party-audited multi-year live track record, we do not clear that bar, and you should apply that consistently to every competitor who doesn't either.

Judge a seller by what they show you when it makes them look worse.

Two things the checklist can't tell you

Second, whether you can leave it alone. The most common way a working system loses money is the owner switching it off in a valley or adjusting parameters mid-flight, and no checklist protects against that. Start with what a trading bot actually is.

Frequently asked questions

Is a live track record better than a backtest?

An independently audited live track record over several years is stronger evidence, and it is also rare, expensive to produce and easy to fake in its unaudited form. A screenshot of a live account proves nothing at all, since demo platforms produce identical images. A long, reproducible backtest with costs subtracted and drawdowns published beats an unverifiable live claim, and both should be read as history rather than forecast.

Should I test a system on a demo account first?

Yes, and for longer than feels necessary. A demo run confirms the mechanics: that it connects, places orders at the times it should, and places the stop and target where it claims. It will not validate the strategy, because a few weeks of trades is far too small a sample to tell skill from luck. Treat it as a plumbing check, not a verdict.

What if the seller refuses to answer these questions?

That is an answer, and it's a complete one. A seller with a documented system finds these questions ordinary and boring, because they've written the material already. Evasion, urgency and offense are all the same signal wearing different clothes. Close the tab and keep your money.

Educational content, not investment advice. Trading futures involves substantial risk of loss and is not suitable for every investor. Any reference to Algomyth results is hypothetical, simulated over historical data, and shown in full on the performance page; past performance does not guarantee future results.