The uncomfortable starting point
We build and run automated trading ourselves — real strategies, executed automatically on a real exchange via API. So this is not an anti-bot article. It’s the opposite: because we automate with our own money, we know exactly how much marketing survives honest testing. The answer: very little. Most of our own strategy candidates failed verification and were killed before deployment.
If you’re evaluating any auto-trading program — ours included — here is the checklist that separates a tested system from a story.
1. Demand a backtest that includes costs
A strategy that trades often pays fees and slippage on every single order. We’ve measured configurations where the raw signal looked fine but four thousand trades of costs turned the result deeply negative. If a bot’s performance page doesn’t state that fees and slippage are included, assume they aren’t — and assume the real result is worse.
2. Don’t buy win rate — buy the full distribution
Win rate is the most abused number in bot marketing. We published a case study where the best configuration of a famous indicator had a 56.8% win rate and still lost about 40% over its July 2026 test window (the live grid re-measures — see the linked post for current cells): it won small and lost big. Ask for average win vs average loss, max drawdown, and profit factor. A bot that leads with win rate alone is hiding the other half of the math.
3. Beware strategies that can’t show you a losing cell
Any honest grid of results has red cells. We publish full parameter grids for our strategies — including the ones where every cell loses — because a strategy that only works in the one configuration shown to you doesn’t work. If every screenshot is green, you’re looking at selection, not evidence.
4. Understand the failure mode, not just the equity curve
Every strategy family has a characteristic way it dies. Grid and martingale bots die in sustained trends (we wrote up why). Trend strategies die in chop. If the seller can’t tell you how their bot loses, they either don’t know or won’t say — both are disqualifying.
5. Check the sample size
A dozen trades proves nothing. Ask how many trades the verification covers and over what period. Results built on thin samples are noise wearing a suit.
6. Defined risk beats predicted returns
The single biggest thing we learned from our own research: what survives isn’t a smarter entry signal, it’s risk control — position sizing, stop discipline, exposure caps, surviving drawdowns. Any bot pitch that talks return without talking maximum loss has the priorities backwards.
7. Verify it yourself — for free
You don’t have to take anyone’s word, including ours. Run the strategy in our free simulator — costs included, parameters open, losing results shown. If you want to see how we run automation against a live exchange, the setup is documented on the autotrading page. Verify first. Deploy second. That order is the whole point.