You bought a solid EA, ran it for four days, hit a losing streak, and turned it off. The EA was fine. Your first week was the problem.
This is the most common way good systems die — not from a flaw in the code, but from an onboarding that guaranteed a bad read. Wrong lot size, a laptop that went to sleep, a broker that slipped the fills, and a sample of six trades you treated as a verdict. By day five you had "proof" it did not work, and the proof was manufactured entirely by the setup.
You are not impatient for wanting to know if it works. You are reacting to a real fear of throwing money at another dud. But the first seven days are not for judging the EA — they are for building the conditions under which judging is even possible. Here is exactly what to do in week one, and the rule for when an EA has actually earned the off switch.
Day 0: The Two Things That Must Be Right Before It Trades
A VPS With Low Latency
An EA running on your home laptop is not really running. The machine sleeps, the Wi-Fi drops, Windows reboots for an update at 3 a.m. and misses the London open, and the execution latency to your broker is whatever your home connection happens to be that second. Every one of those is noise that will get blamed on the EA.
A proper VPS near your broker's server fixes all of it: the EA runs 24/5, and the round-trip to the broker drops to single-digit milliseconds. For latency-sensitive entries, aim for under ~5 ms ping to the broker's data center. This is not optional polish — it is the difference between testing the EA and testing your home internet.
A Broker That Does Not Add Its Own Noise
If the broker requotes on news, slips the open, or widens spreads under volatility, the EA's results get corrupted by execution the EA never chose. You will see losses and assume the logic failed, when the logic was fine and the fills were not. Clean, raw-spread execution is a prerequisite for a readable first week — you want the only variable to be the EA, not the venue.
Clean execution is the prerequisite for a fair test.
A new EA can only be judged on a broker whose fills do not add their own mess. IC Markets (raw spreads, institutional execution) and Axi are the two we run on for that reason. Affiliate links — if you sign up through them and hit any account issue, message me and I escalate it to my Axi manager directly. Pair either with a low-latency forex VPS and your week-one data is finally clean.
Days 1–2: Risk Percent, Not Fixed Lots
The single fastest way to kill a good EA in week one is to run it at the wrong size. Fixed lots on a small account is how a normal drawdown becomes a margin call — the EA does its job, the position size does not match the equity, and the account dies on a perfectly ordinary losing sequence.
Set risk as a percent of equity from the first trade: ~2% per trade as a standard starting point, and ~0.5% if you are running it toward a prop-firm challenge with tight drawdown rules. Risk-percent sizing means the EA's bad week is survivable by construction, instead of being the thing that ends the test. Fixed-lot sizing is the number-one first-year EA killer — fix it before the first trade, not after the first scare.
Days 3–5: Let the Sample Size Build
Here is the hard part for everyone: doing nothing while the number of trades grows. Six trades tell you nothing. A losing streak of three out of four is completely normal variance for a profitable system and statistically meaningless. If you judge on a handful of trades, you are reading noise and calling it a verdict.
The EA needs enough trades across enough different market conditions before its real behavior shows. In week one you are not evaluating performance — you are accumulating sample. The right action during a normal drawdown is usually to leave it alone and let the data build; doing nothing during a drawdown is frequently the smartest move available, because most "fixes" applied in week one are panic dressed as discipline.
Days 6–7: What "Working" Actually Looks Like
By the end of week one you are not looking for profit. You are looking for behavior matching the description. Did the EA take the kind of trades it said it would? Did it respect its stop logic? Did it sit out conditions it is supposed to avoid? Did position sizing behave as configured? A "working" first week can absolutely be slightly red — what matters is that the system did what it claimed, not that the curve is already green.
If the EA is part of an AI stack, "working" includes the reasoning being visible and sane, not just the P&L. The seven components a real AI trading system needs are what you are actually verifying in week one — that the machinery runs as described.
The Kill-Switch Rules (When It IS Broken)
Patience is not the same as blind faith. There are conditions that justify turning an EA off in week one — they are just specific, not emotional:
- It violates its own stated risk. If it is supposed to risk 2% and you see a single position risking 10%, that is a real defect, not variance. Off.
- It trades instruments or sessions it said it would not. Behavior outside its own spec is a red flag worth stopping for.
- The drawdown blows past the documented maximum. Exceeding the worst drawdown in the published track record this fast means something is wrong with the setup or the EA. Investigate before continuing.
- Execution is visibly broken — orders rejected, stops not honored by the broker. That is a venue problem, but it still means stop and fix before the data is worth anything.
Notice what is not on that list: "it had three losing trades," "it is down 1.5% on day four," "I have a bad feeling." Those are the reasons good EAs die, and none of them are the EA's fault.
The Honest Close
An EA is a multi-month instrument that most people judge on a long weekend. The first seven days are setup, not trial: the VPS, the broker, the risk percent, and the patience to let a sample exist before you read it. Get those right and you will finally be judging the EA instead of your own onboarding.
Give it the week it needs to show you what it actually is. Then judge it hard — on its drawdown, its behavior, and its data, not on day four's mood.
Want the setup checklists and the weekly forward-test notes — including the bad weeks? The DoItTrading newsletter sends them in plain language. And if you want an EA built to be watched in public from day one, Alpha Pulse AI is the one we forward-test live.
Frequently Asked Questions
Should I run a new EA on demo or live first?
A short demo run confirms the EA installs, attaches, and behaves as described without risking capital — useful for the first day or two. But demo fills do not fully reflect real spreads and slippage, so a small live or low-risk account gives a truer read once the basics check out. Either way, the first week is about clean setup and accumulating sample, not about the size of the result.
How many trades before I can judge a new EA?
Far more than one week usually produces. A handful of trades is statistical noise; you need enough trades across varied market conditions before performance is meaningful — typically dozens at minimum, and ideally a track that spans different regimes. In week one you are gathering sample, not reaching a verdict. Judging on six trades is the most common way good EAs get killed early.
What VPS specs does an EA need?
Low latency to your broker matters more than raw horsepower. Choose a VPS located near your broker's server and aim for under roughly 5 ms ping for latency-sensitive strategies. Enough RAM to run your terminal and charts smoothly, and 24/5 uptime so the EA never misses a session because a home machine slept, are the essentials. The goal is that execution conditions stop being a variable in your test.
What risk percent should I start a new EA at?
Use risk as a percent of equity rather than fixed lots from the first trade. Around 2% per trade is a standard starting point for a normal account, and around 0.5% if you are running toward a prop-firm challenge with strict drawdown limits. Risk-percent sizing keeps a normal losing streak survivable, which is the whole point of week one — staying alive long enough to actually evaluate the EA.
