The ten mistakes that ruin a perfectly good trading system
Most automated systems do not fail on their own. Their owners break them. The ten mistakes that ruin a working trading system, and what each one really costs.
The system was identical. Everything that happened around it was not. Once the rules live inside a machine, the machine stops being the variable. You are.
An automated system fails in two ways: the edge stops working, or the owner intervenes. The second is far more common, far faster, and entirely under your control. Every mistake below was a decision somebody made on purpose, while feeling something.
| Mistake | What it feels like at the time | What it actually costs |
|---|---|---|
| Switching it off in the valley | Prudence | The recovery you paid for |
| Tweaking parameters | Improvement | Your entire track record |
| Too much size | Ambition | The account, on a normal drawdown |
| A second system too early | Diversification | Any ability to diagnose either one |
| Judging it by the week | Attentiveness | Every decision made out of noise |
| No logging | Simplicity | Any chance of knowing what went wrong |
| No restart plan | Optimism | The days the platform sat closed |
| Manual trades in the same account | Opportunism | Clean data, and often the drawdown limit |
1. Switching it off in the valley
The expensive one, and it always looks responsible. Four red days, the curve sags, and turning it off feels like risk management. It is the most reliable way to collect all of a system's losses and none of its wins.
2. Tweaking parameters mid-flight
Widening the target because the market feels lively. Tightening the stop because yesterday it got clipped by two ticks. Skipping Fridays because Fridays have been bad lately.
Each edit sounds sensible. Together they produce a system that has never been tested, has no backtest, and shares nothing with the one you bought except its file name. You replaced it with an untested version and kept the old confidence.
The moment you change a parameter, your track record resets to zero trades.
If you truly think a change is an improvement, prove it the way the original was proved: run it in the Strategy Analyzer across the full 88-month history on a separate profile and compare. That takes an afternoon. Changing it live and watching a week settles nothing.
3. Starting with too much size
Size does not change the quality of a system. It multiplies both ends of it, and your tolerance for pain does not scale with your account balance.
Standard sizing is 50k at one contract, 100k up to two, 250k up to five. Those are ceilings, not targets. Starting under the ceiling buys the only thing that matters early on: surviving long enough for the average to work. Test it in the calculator first.
5. Adding a second system too early
Two systems in one account is not diversification. It is a diagnostic problem you built yourself.
When the account goes red you cannot tell which one is responsible, whether both are losing at once, or whether one is quietly funding the other's mistakes. When it goes green, the same problem in reverse. You have doubled the exposure and halved the information.
Run one thing until you know what its normal looks like: the shape of its good weeks, the depth of its bad ones, how long a flat stretch runs. That takes months. If you still want a second system afterward, give it its own account.
6. Judging it by the week
Judge by the quarter, against the system's own history. Anything shorter and you are deciding out of noise. See how much a trading bot actually makes.
7. No logging
Something goes wrong six weeks in. The platform disconnected, a trade did not fire, or one filled at a price you did not expect. You open a ticket and the first question back is: what happened, exactly, and when?
If the answer is "I think it was Thursday, maybe," nobody can help, including you. Keep the platform logs, export the trade history, and write one line per trading day: machine on, system armed, trade taken, anything odd. Thirty seconds a day turns a vague grievance into a solvable problem.
8. No restart plan
Computers reboot. Windows updates at the worst moment. The internet drops. The platform locks up. All of that is normal, and none of it is a crisis if you decided in advance what to do about it.
Write the answers down before you need them: how you confirm the system is armed this morning, what you check if the machine restarted overnight, what you do if you return mid-session to an open position. A system that places real stop and target orders keeps its risk capped even when your machine is dead, because those orders live on the broker's servers. See how it works. Most people's restart plan lives in their head, which is exactly where it stays when the screen is blank at 8:29 AM ET.
9. Mixing manual trades into the same account
You spot something the system did not take. It looks obvious. You take it by hand, same account, because why not.
Three consequences, all bad. The account's numbers become a blend of two strategies, so the system's performance is no longer measurable. Your manual trade competes for the same risk budget, and a discretionary loss can push the account into a limit while the system is still managing a position. And it reopens the door automation was built to close. Trade by hand in a separate account, with separate money.
The pattern behind all ten
One thing connects every item: each mistake is an attempt to feel useful. Switch it off, adjust it, add to it, watch it harder, take the trade it missed. Doing something feels better than doing nothing, especially with money on the line.
But the reason to automate is that here, doing nothing is usually correct and almost nobody manages it unaided. Your job is the boring perimeter: platform current, machine on, events logged, rules known, size low enough that a normal bad stretch is survivable.
Frequently asked questions
When is it actually correct to turn a system off?
When its behavior stops matching its own documented history: the drawdown goes deeper than anything in the record, the average winner or loser changes shape, or trades fire in conditions the rules should have filtered. Those signals are structural and checkable against the backtest. A run of losses inside historical norms is not one of them.
How long should I run a system before deciding it does not work?
Longer than feels comfortable. An edge that breaks even at a 25% hit rate needs dozens of trades before the average carries authority, and a system can sit underwater across a whole quarter with nothing wrong. Compare your live results to the equivalent window in the backtest rather than to zero. If a quarter of uncertainty is unbearable, automation will not fix that.
What if I make one of these mistakes and the account survives?
That is the dangerous outcome, not the lucky one. Turning a system off in a valley and watching it go flat, or doubling size into a good week, teaches you the mistake was fine. It was not fine, it was unpunished. Judge a decision by the process behind it, not by how the next five trades landed.
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.
Twenty years, drawdowns included
Every engine publishes its full history repriced at today's values, with the losing years and the deepest drawdowns printed as plainly as the profits. Verify first, decide later.
More from Judging a system
Judging a systemDo Trading Bots Really Work? What the Evidence SaysDo trading bots work? Yes at executing an edge without flinching, no at inventing one. What the evidence supports, and what the marketing quietly leaves out.Judging a systemWhy cheap trading bots cost the most, in actual dollarsA cheap bot that burns months and a slice of your account costs far more than its price. What cheap trading bots really cost, in dollars.
Judging a systemWhat a Trading Bot Can and Cannot Legally Promise YouA US trading bot cannot legally promise guaranteed returns or say it is risk-free. Where the CFTC, NFA and FTC draw the line, and what honest claims look like.