Is my trading bot actually working? Read the numbers

Your bot has been running three weeks. Account is up. Feels good. Then comes the question everybody asks themselves at that point: is this real, or did I just get lucky?

Honest answer: after three weeks you cannot know. Not because your bot is bad, but because you have too few trades to say anything. That is not hand-waving, that is how probability works. Below are the numbers you do judge a bot on, with the thresholds that matter in practice.

How do I know if my trading bot is actually working?

By reading three things together: trade count, profit factor and expectancy after costs. Below 30 trades any result is noise. From 60 trades you get a direction, and only around 100 to 200 trades do you know with reasonable confidence whether the edge is real or you are looking at luck.

Those three hang together. A pretty profit factor across 12 trades is worthless. So is a tidy expectancy you calculated before costs. And 300 trades at a profit factor of 1.05 mostly tells you that you are working hard for nothing.

Here is what you see constantly: somebody posts a screenshot of plus 14 percent in a month. Without the trade count attached, that is not evidence, it is a picture. Always ask for the number of trades and the deepest drawdown. Anyone unwilling to show those two usually does not have them.

How many trades do you need before a result means anything?

At least 30 before a number carries meaning, 60 or more for a first conclusion, and 100 to 200 for real confidence. Below that you are measuring chance.

Take a coin. Five heads in a row happens roughly once every 32 attempts. That is not rare, that just happens. So five winning trades in a row prove exactly as much as five heads: nothing. Yet that is precisely the moment people double their deposit.

It cuts the other way just as hard. Four losers back to back feels like a broken bot. At a 55 percent win rate you should expect at least one five-trade losing streak per year purely from the statistics. We worked through the arithmetic behind those dips separately in the maths of drawdown and recovery.

How fast you reach those counts depends on your style. A scalping bot taking ten trades a day hits 100 within two weeks. A swing bot that acts twice a week needs a year. That difference decides how patient you have to be before you are allowed to conclude anything.

What is profit factor and which number is good enough?

Profit factor is gross profit divided by gross loss. Above 1.0 you make money, 1.5 to 2.0 is decent, above 2.0 is strong. Below 1.2 the margin is thin enough to disappear as soon as spreads widen.

If you made 3,000 gross and lost 2,000, your profit factor is 1.5. For every unit you lost, you pulled in one and a half.

Now the part you rarely read: a high profit factor can be a warning. See 6.0 on a three-month test and you are almost always looking at a strategy optimised to death on exactly that period. Those numbers do not survive contact with a live account. A profit factor between 1.5 and 2.5 over a long window is worth far more than a 6 over one quarter. More on why test results flatter reality in backtesting and misleading results.

What does expectancy tell you that win rate does not?

Expectancy is what you make per trade on average: win rate times average win, minus loss rate times average loss. Win rate on its own says nothing about whether you make money.

Run the numbers. Bot A wins 40 percent of trades, average win 90, average loss 40. Expectancy is (0.40 × 90) minus (0.60 × 40), so 36 minus 24. Twelve per trade. Bot B wins 80 percent, average win 15, average loss 70. That is (0.80 × 15) minus (0.20 × 70), so 12 minus 14. A loss of two per trade.

Bot B looks better on every screenshot. Eighty percent winners, who would say no. And it bleeds out slowly anyway. This is exactly the pattern behind grid and martingale systems, where nearly every trade wins until the one that takes it all back. We wrote a separate piece on the danger of martingale and grid.

And always calculate after costs. An expectancy of 5 dollars per trade sounds fine, until you subtract 7 dollars of commission and spread per round turn. Then you sit at minus 2. On high-frequency bots this is the most common way to talk yourself rich.

In what order do you read an MT5 report?

Start with maximum drawdown, then the shape of the equity curve, then profit factor and recovery factor together, then the trade count, and read the Sharpe ratio last. The net profit at the top is the least informative number on the page.

Why drawdown first? Because that is the number that decides whether you last. A bot that makes 60 percent with a 45 percent drawdown on the way is one nobody sits through. You switch it off at the bottom, like everyone does.

The shape of the curve then tells you more than the endpoint. A line grinding steadily upward with small dips is a different animal from a flat line with one enormous jump in it. On that second one your whole result came from one or two trades, and those may never repeat.

Recovery factor is net profit divided by maximum drawdown. Above 2 is solid, above 3 is strong. It tells you how much profit you got for the pain you had to swallow along the way. Here are the thresholds side by side.

Why does a backtest look better than your live account?

Because the test leaves out three costs that exist live: delay between signal and execution, spread that widens at the worst moments, and commission that sits at zero in many demo templates. A strategy scoring 1.8 on paper often lands near 1.3 live.

What a test needs before you take it seriously: 200 trades or more, a period of two years or longer, and tick data at 99 percent modelling quality. Miss one of those three and you are looking at an indication, not proof.

That gap between paper and practice also shows up between demo and live, with the same bot on the same settings. We dug into where it comes from in profitable on demo, losing on live.

When is your bot genuinely broken instead of in a rough patch?

When drawdown goes deeper or lasts longer than the bot ever showed in testing, or when the ratio between average win and average loss shifts structurally while your win rate stays flat. A five-trade losing streak is just part of the deal.

The practical rule we use: write down beforehand what the deepest drawdown in your test was and how many trades it lasted. Go past that live and you have a reason to look. Stay under it and all you have is discomfort, and discomfort is not a signal.

Watch that second form especially, because it is subtle. Your bot still wins as often, but its winners shrink and its losers grow. That happens when the market changes character, for instance when daily range structurally collapses. Your win rate will not reveal it. Your expectancy will, and that is exactly why you track it monthly instead of once a year.

What we fix for our own bot before any money is on it: a maximum of 1 percent risk per trade, a hard stop at 3 percent loss on the day, no more than three positions at once, and a pause around high-impact news. Those rules are set in advance, so if results drift you know it is the market and not a tweak somebody made quietly. And because we take a 30 percent commission on your profit and nothing on losses, we have no reason to stretch those limits.

So do not start with the question of whether your bot makes money. Start with whether you have enough trades to be allowed to ask. Usually the answer is no, and then the best action is to leave it running and reopen the report in two months.