Why the bot you bought does not work

Someone sells a bot. The page shows an equity curve rising smoothly to the right. You buy it, run it, and it loses money. This happens often enough to be a pattern rather than bad luck, and the mechanism is worth understanding — it applies to bots you build yourself too.

The curve you saw was selected, not produced

Take one mediocre strategy with six parameters, each with ten plausible values. That is a million configurations. Run them all against history and some will look spectacular purely by chance — with a million draws, the best draw is always impressive.

The seller does not have to be dishonest. They ran a parameter sweep, kept what worked, and published it. The curve is real; the process that produced it guarantees the curve overstates what happens next.

You are seeing the survivors

For every bot on sale with a good track record, some number were built, performed badly and were quietly withdrawn. You never see those. The visible population is filtered by exactly the outcome you are trying to evaluate.

The same applies within a single seller's catalogue. Ten strategies launched, three did well, the page shows three.

Everyone runs the same one

This is the part specific to bought bots and it is structural. If a thousand people buy the same bot with the same defaults, a thousand accounts place similar orders at similar moments.

• Entries push price against the later entrants. The first hundred get the level; the rest get slippage.
• Stops cluster at the same place, so a move into that zone triggers all of them and makes the move worse.
• Whatever edge existed is being consumed by the crowd that bought access to it.

A strategy sold to many people is, by construction, a strategy whose edge is shared. This is the same crowding mechanism described in agents trading against agents, just distributed by a payment page instead of by research.

The parameters you were not shown

A backtest is a set of assumptions. The ones that most often differ from your reality:

When a bought bot is the right answer

It sometimes is, and pretending otherwise would be self-serving. A marketplace bot makes sense when the strategy is mechanical and well understood — a grid or DCA on a liquid pair — when the amount at risk is small enough that the cost of custom work cannot be justified, and when you are prepared to treat it as an experiment rather than an investment.

What you are buying in that case is convenience, not edge. Price it accordingly.

What to ask before buying

1. Show me performance for the parameter values next to the ones you recommend.
2. How many people run this configuration?
3. What fee tier, what slippage assumption, was funding included?
4. What is the longest losing streak in the test, and the deepest drawdown?
5. Do I get the source, or does it stop working when you stop supporting it?
6. Does it need withdrawal permission on my API key? (If yes: walk away.)

Honest sellers answer these. The answers also happen to be the same questions you should ask a developer building something custom — see how to choose a developer.