Most automated systems that fail do not fail because the code broke. They fail because the person running the code made a decision the code was designed to prevent: doubled the size after a good week, switched the strategy off two days before it recovered, or ran it on a funded account whose rulebook forbade automation in the first place.
The rules below are the ones that decide whether a working system survives contact with its owner. None of them are clever. All of them are broken constantly.
Almost every rule here protects the system from you, not from the market. The software will run the same plan on the worst day of the year as on the best one. Whether it gets to is your decision, and that decision is where the money is actually won or lost.
Somebody else's report is a claim. Your own report is evidence. Before a single live order, load the strategy into NinjaTrader 8, run the period the vendor published, and confirm the numbers land where they should. Then run it forward in simulation and in market replay so you see how it behaves bar by bar. If a vendor cannot hand you something you can reproduce, you are not evaluating a system, you are evaluating a sales page. The method is in how to backtest in NinjaTrader 8.
The Algomyth engine figures most often quoted here describe a $50,000 account at the Medium sizing level. Every figure that comes out of a report, the profit, the average trade, the $5,273 worst drawdown, describes the size it was run at and no other. Double the contracts and you have doubled the drawdown as well as the profit, which means the numbers you read no longer apply to the account you are running. Start where the test starts. Scale later, deliberately, with the arithmetic in front of you.
This is the single most expensive instinct in trading and automation does not immunize you against it. Doubling after a loss feels like recovery and is mathematically a shorter path to zero, because the sequence of losses required to end you gets shorter every time you scale into weakness. The full math, and why every martingale variant eventually meets a streak it cannot survive, is in the math of ruin.
A system whose target sits at three times its risk does not need to be right most of the time: break-even arrives at a 25% win rate. Losing streaks are arithmetic, not malfunction. Turning the strategy off during one guarantees you take the losses and miss the recovery, which is the worst available outcome and the most common one.
The defense is to decide in advance, in writing, what would actually justify stopping: a drawdown deeper than the tested maximum, a run of trades outside anything the backtest produced, or a change in the system's own behavior such as entries firing at the wrong time. "It feels wrong this week" is not on that list. Learn to read the difference in what drawdown really is.
A bot does not turn a bad plan into a good one. It turns a good plan into one that actually gets followed, and only if you let it.
Ten trades tell you nothing. The published Algomyth engine backtest contains 1,164 of them for a reason: an edge worth roughly $75 a trade after costs needs a lot of repetitions before it separates from noise. In a short window a losing system can look brilliant and a winning one broken. Nor is a bad run the measure: at a 1.46 profit factor, with the target placed at three times the stop, this system pays for its losing trades through the size of its winners. Why that combination works is set out in risk, reward and win rate.
Hypothetical performance. The figures above come from a backtest of the Algomyth engine over historical data, not from a live account. Simulated results are prepared with hindsight, carry no financial risk, and cannot fully reflect real execution, slippage or liquidity. Past performance, real or simulated, does not guarantee future results.
If you trade funded capital, the prop firm's terms sit on top of your strategy and override it. Four clauses matter most, and every one of them can end an account that is making money.
| Clause | What it can do to a working system |
|---|---|
| Automation permission | Voids the account outright if strategies are not allowed, or not allowed in that account type |
| Trailing drawdown | Closes the account on a dip the strategy would have recovered from |
| Daily loss limit | Ends a session the system would have finished flat |
| Consistency rule | Blocks a payout because one good day was too large a share of the total |
Ask before you pay, get the answer in writing, and read funded account rules explained. A rule you did not know about costs exactly as much as one you ignored.
The full backtest is published with the account size, the contract count and the drawdown attached, along with the live signal log that archives every new signal with its outcome.
Duplicate strategies are the most common technical failure in retail automation. Two copies of the same strategy on two charts of the same instrument will both take the signal, and you are suddenly at double size without deciding to be. The same thing happens when a strategy is enabled in the Strategies tab and again on a chart, or when a disconnect leaves an orphaned instance running.
Check the Strategies tab before every session, confirm exactly one instance is enabled, and confirm the account it is pointed at. Ninety seconds of checking prevents the one error that is expensive every single time. More of these in the mistakes people make automating.
A system that misses its entry is not flat, it is a different system. If the trade fires once a day at 9:30 AM ET, the machine has to be awake, connected and logged in at that minute, every day. That means power settings that never sleep, automatic updates that never reboot during the session, a broker connection that reconnects, and a plan for what happens when your internet drops mid-position. Whether that argues for a VPS or a dedicated machine depends on your setup: VPS for automated trading.
When something looks wrong, you need to know whether it is wrong. That requires a record: the trades the strategy took, the fills it got against the fills the backtest assumed, the days it did not trade and why, and every change you made with the date you made it. Without it you are arguing from memory, and memory is generous about the weeks you did well. A working format is in the journal for an automated system.
Every rule on this list is a way of saying the same thing: the system's edge only exists across a long sequence of trades, and every intervention breaks the sequence. Size changes break it. Switching off breaks it. A missed session breaks it. A rule violation ends it entirely.
That is why the honest version of automated trading is boring, and why boring is the point. The software's advantage over you is not that it is smarter. It is that it does not get bored, scared or confident, three states that produce most of the decisions on this page.
Long enough to produce a sample that means something, which is measured in trades rather than weeks. A handful of trades tells you almost nothing about an edge worth a few tens of dollars per trade. Decide that threshold in advance, along with the drawdown that would genuinely justify stopping, and write both down before you start.
You can, but understand what you are doing: you are now running an untested system with the confidence of a tested one. Any parameter change invalidates the backtest that convinced you to run it, so the honest sequence is to test the change over the full history and out of sample first. Most changes made mid-drawdown are the drawdown talking, not the market.
Three things. A drawdown deeper than anything the backtest produced, behavior the strategy should not exhibit such as entries at the wrong time or orders without a protective stop, and a broken assumption such as the instrument being rolled or the platform losing its data feed. A losing week is none of those, and treating it as one is how people take the losses and miss the recovery.
Not necessarily. What you need is a machine that is reliably awake, connected and logged in when the strategy needs to act. A dedicated computer at home with sleep disabled and updates scheduled outside the session can do that, and a VPS simply makes it somebody else's problem for a low monthly fee. Choose based on how reliable your power and internet actually are, not on what sounds more professional.
Switching the system off inside a normal drawdown, closely followed by trading more contracts than the tested size. They arrive together: oversizing makes an ordinary drawdown feel unbearable, and that feeling produces the shutdown. Starting at the tested size makes the rest of the list possible to follow.
Everything we write comes back to one idea: verify, then decide. The live terminal exists for exactly that.
Unlock Algomyth Live