EdgeWatchFXby ErgodicLabs

The RecordMethod

What changes when an EA goes live

A backtest answers whether a strategy could have worked. A live account answers a different question: is it still behaving like itself? Here is how to tell a normal bad run from an edge that has faded, and why it takes longer than people expect.

2 min read

Before an EA goes live, the question is whether the backtest is robust or fitted to its history. After it goes live, the question changes. The strategy is now running on data it has never seen, with real spreads and real fills, and what matters is whether it still behaves like the record it built.

Every EA against its own past

The fair comparison for an EA is not another EA, or a market average. It is its own earlier trades. A trend follower with a 35% win rate is fine at 35%; a scalper with an 80% win rate is not fine at 60%.

So the first part of a live record becomes a baseline: how much it usually made per trade, how often it lost, how deep its drawdowns went, how long it held positions and how large they were. Every later trade is checked against that.

Noticing a small change takes time

Suppose an EA made 0.15R per trade on average with results that scatter by about 1R, and its edge quietly drops to zero. How long before the record shows it?

Longer than most people expect. A running sum of results below the baseline, the standard tool for this, will notice on the order of a hundred trades later for a typical EA, and sometimes more. Before that, a faded edge and an ordinary bad run look the same.

A bad month is not evidence that an edge has gone. A bad month that goes on longer than the EA’s own history explains is.

Faster signals

Because results alone are slow, other signals help:

The losing streak

How likely was a losing run this long, at the EA’s usual loss rate? A run that its own history would almost never produce is worth attention sooner than the average.

The drawdown and the time under water

Resampling the baseline’s own days shows how deep, and how long, its drawdowns usually were. A current drawdown well outside that range stands out early.

Behaviour

Sizing, holding time, trades per week and legs per trade are visible from the first trade. An EA that suddenly holds positions three times longer, or opens twice as many, has changed, whatever its results say yet.

False alarms

Every monitor that reacts quickly also reacts to noise. The useful question is how often it would raise a false alarm on an EA that had not changed at all. That rate can be measured by running the monitor on simulated records with no change, and a monitor whose false-alarm rate is not known is hard to trust.

How EdgeWatchFX does it

Each EA is compared with its own record. The first half of its trades, up to 200, is the baseline; the baseline stops growing so that a faded edge does not slowly become its own yardstick. The monitors are a running sum of results below the baseline, the chance of the longest losing run, the drawdown and time under water against resampled copies of the baseline, and changes in sizing, holding time and trade frequency. An EA is Healthy, on Watch, Degraded or Broken; each monitor leaves a level only well below where it entered it, so states do not flicker, and the app shows roughly how often a false Watch would come up at that EA’s trading rate.

Figures in this note describe recorded results. None of it is a forecast or investment advice.

More from The Record