How the Machine Learning Filter in Our New EA Actually Works — Without the Buzzwords

Machine learning is one of the most overused phrases in EA marketing.

It appears in product titles, descriptions, and launch screenshots. Rarely with any explanation of what the system is actually doing.

We built an ML component into DVN Core Gold LION. This post explains exactly what it does — and what it does not do.

What It Is Not

It is not a neural network predicting price direction.

It is not a deep learning model trained on historical tick data to find patterns invisible to the human eye.

It is not a black box that makes decisions the developer cannot explain to a customer.

These approaches fail in production for a consistent reason: they overfit to past data, then fail when market behavior shifts. A model trained on 2022 conditions does not reliably trade 2025 conditions.

We did not build that.

What It Is

The ML filter in DVN Core Gold LION has one specific job.

XAUUSD trades differently across different sessions of the day. The London open behaves differently from the New York open. Midday liquidity behaves differently from the pre-close session. Overnight hours behave differently from both.

These differences are real. But they are not identical across every broker, account type, and spread environment.

A hardcoded session filter applies the same rules to every account. The ML filter learns from your specific account's trade history — and adjusts accordingly.

How It Learns

The filter activates after 50 closed trades.

At that point, the system reviews the outcome of every closed trade and identifies which 4-hour time window of the trading day it occurred in.

It assigns a performance score to each time window based on those results.

It ranks all time windows from strongest to weakest.

It permits trading only in the top three.

All other windows are suspended.

How It Stays Current

Every 10 new closed trades, the model retrains automatically.

It does not lock in an initial set of preferred hours and ignore what happens afterward. The filter is continuously updated based on the most recent trade data.

If your broker changes spread behavior. If a new news schedule affects liquidity at a particular hour. If the character of a session shifts over time. The filter adjusts — over time, incrementally.

What Changes in Practice

Without the filter active, all five strategies in DVN Core Gold LION trade across all available hours.

With the filter active, only the time windows with a demonstrated performance edge in your live account are permitted to trade.

The strategy logic does not change. Entries, stop losses, and exit rules remain exactly the same. The filter does not modify how trades are placed — it modifies when trading is permitted.

The result is fewer trades. Higher average quality per trade.

One Limitation Worth Stating Clearly

The filter requires 50 closed trades before it activates.

During this initial period, the system trades without time-based filtering — on all available hours.

If you are starting a new account, this warm-up period will produce results that do not reflect the fully filtered system. This is expected behavior and is documented in the setup guide.

We are stating it here because it belongs in the explanation — not in the fine print.

DVN Core Gold LION. Coming June 10.

Setup guide, documentation, and live signal: → https://www.mql5.com/en/channels/dvn_core