From backtest to live: what to monitor every week once your EA is running

Live is a different job

Building a forex robot is a research task. Running one is an operations task, and the skills do not overlap much. The research question was "does this have an edge". The operations questions are "is it still running", "is it still getting the fills it needs", and "is it still the same strategy it was when I measured it". None of those are answered by looking at the balance, which is what most new developers look at.

What follows is a weekly routine that takes about twenty minutes on a live-synced ConfirmedTrades account page. It is deliberately boring. Surprises should come from the market, not from your monitoring.

1. Is it alive? The heartbeat

An EA fails silently. The VPS rebooted, Algo Trading got switched off, the chart was closed, the broker connection dropped and never came back. The balance looks fine because nothing happened. The Heartbeat page in the dashboard shows each connected account as Live-synced, Sync lost (no report for 30 minutes without a goodbye) or Connector unloaded (a deliberate stop), along with terminal health: ping to the broker, build, whether automated trading is allowed, and the open charts with the EA on each.

Turn on heartbeat alerts for the accounts that matter so you are told rather than having to look; EA heartbeat monitoring covers the setup. Weekly check: was there any gap last week, and did any trades fall in the gap? A strategy that missed Tuesday's session is not the strategy you backtested.

2. Is it getting the fills it needs? Execution in points

Open the Execution panel, pick the EA's main symbol, and look at two things. The spread histogram for the hours the EA trades, and the signed slippage on opens and closes (MT5 reports both from history; MT4 reports stop-loss and take-profit closes). Everything is in points with the digit count shown, so the numbers are comparable week to week and against other brokers.

Worked example: the EA's average winner is 60 points. Typical spread at its trading hour was 12 points last month and 19 points this week; average close slippage moved from −1 to −4 points. Round-trip cost went from roughly 14 to 27 points, from under a quarter of the average winner to almost half. Nothing in the strategy changed, and its expectancy just halved. This is the most common way a marginal EA dies and the balance chart shows it last. Broker cost audit has the full method.

3. Is it still the same strategy? Rolling statistics

Use Custom Analysis to filter the record to the last four weeks and compare the size-independent rows against the full record: expectancy in points, profit factor, payoff ratio, win rate, average holding time, trades per week. Then do the last twelve weeks. You are looking for drift, not a bad week; a bad week is expected.

Win rate is the noisiest of these and needs a sanity band. With a true 55% win rate, the standard error over 40 trades is about 7.9 percentage points, so anything between roughly 39% and 71% in a month is consistent with nothing having changed. Over 160 trades the band narrows to about 47% to 63%. If rolling win rate is outside the band for two consecutive windows, or holding time has shifted materially, look for a cause: broker change, symbol spec change, a parameter you "tweaked", or the market. Drift in holding time with no drift in win rate often means the stop or target is being filled differently, which points you back to item 2.

• Expectancy (points): last 4 weeks vs full record.
• Profit factor and payoff ratio: same comparison.
• Win rate: inside the sampling band for the window size?
• Average holding time and trades per week: unchanged?

4. Is this losing streak normal? Streaks against expectation

The account page reports the longest winning and losing streak, the current streak, and the Z-score of the win/loss sequence. Before you panic at a run of losses, compute what independence predicts. The expected longest losing streak over N trades is roughly the natural log of N divided by the natural log of 1 over the loss rate. Worked example: 45% loss rate over 200 trades gives ln(200) / ln(2.22), which is 5.3 / 0.8, about 6.6. A losing streak of 7 is ordinary. A streak of 12 in that record has roughly a one-in-200 chance and deserves a look; it is not proof the edge is gone, but it is a reason to check items 2 and 3 before the next trade rather than after.

If the Z-score is strongly negative with a high confidence, the streaks cluster more than chance and the independence estimate understates them. Add a couple of trades to the expected maximum before you set any circuit breaker.

5. How deep, and is it on record? Drawdown

Read Max DD (balance) and Max DD (equity) in Glance. These come from the permanent ledger and only ever get worse, so a week in which either moved is a week to write down. Compare the current equity drawdown with the Monte Carlo 95th-percentile figure you sized from in position sizing and risk of ruin. Inside it: keep running. Past it: you are now outside the range your own data predicted, and the honest move is to cut size, not to hope.

Also glance at the Ulcer index and the worst recovery in Performance depth. A drawdown that is shallow but has lasted longer than the historical worst recovery is the same kind of warning as a deep one.

6. Is it doing things you did not design? Flags and concentration

The AI review's deterministic checks run over the closed trades on every sync. Weekly, read the flags rather than the narrative. A new high-severity flag (stop-loss share under 15%, lots rising after losses, average loss more than 2.5 times the average win) on an EA you did not build that way means a bug, a settings change, or manual interference on the account. A concentration flag (over 85% of trades on one symbol) on a multi-symbol EA means the other symbols stopped trading. Check the by-symbol table with its long and short split: an EA that has quietly become long-only on one pair is a different strategy.

Check magic numbers in the trade table too. Trades without the EA's magic are manual trades, and a manual trade inside an EA's record contaminates every statistic above it.

The weekly checklist

Same order every week. Tick it or note why not.

• Heartbeat: any Sync lost gaps? Trades missed inside them?
• Execution: spread at trading hours and signed slippage, in points, vs last month. Cost as a share of average winner.
• Rolling stats: 4-week and 12-week expectancy, profit factor, payoff, win rate within its band, holding time, trades per week.
• Streaks: current and longest losing streak vs the expected maximum for the sample; Z-score confidence.
• Drawdown: ledger moved? Current equity DD vs the 95th-percentile Monte Carlo figure. Ulcer and recovery trades.
• Flags: new AI review flags, concentration, magic-number hygiene, long/short balance per symbol.
• Decision: keep size, cut size, or stop. Write the decision down with the numbers that drove it.

Why do this on a public record

You could keep all of this in a spreadsheet. A live-synced, broker-verified page is better for one reason: it is computed from the broker's data by someone who is not you, it cannot be edited, and the drawdown ledger remembers. Buyers get the same numbers you do, which is the whole point of selling a verified EA. Connecting and publishing is free; sign up and put the terminal on it this week.