SuperTrend Martingale EA for XAUUSD: Phase A/B Self-Learning Engine Explained (MT5)

Most SuperTrend-based Expert Advisors on the market do one thing: they open a trade when the SuperTrend line flips, add martingale multipliers on losses, and hope the trend holds. They never actually learn anything from their own trading history. This article breaks down a different approach — a SuperTrend Martingale EA for MetaTrader 5 built specifically for XAUUSD (Gold) that includes a two-stage self-learning engine, referred to internally as Phase A and Phase B.

What Makes This EA Different From a Standard SuperTrend Martingale Bot

A typical SuperTrend martingale system uses a fixed multiplier and a fixed set of entry rules from day one. The problem is that market conditions on XAUUSD change constantly — spread, volatility, and trend strength in 2026 are not the same as they were even a few months earlier. A static rule set either overtrades in choppy conditions or under-trades in strong trends.

This EA instead runs in two learning phases:

• Phase A — Data Collection Mode: The EA trades live (or on a demo account) using its baseline SuperTrend + martingale logic while logging the outcome of every trade. This phase is not about performance yet — it is about building a reliable sample size. In testing, a minimum of 300 completed trades was used as the threshold before moving on, since fewer trades produce statistically unreliable adjustment signals.
• Phase B — Adaptive Mode: Once the 300-trade threshold is reached, the EA activates its learning engine. From this point forward, entry confidence, lot sizing, and martingale step behavior are adjusted based on the win/loss pattern collected during Phase A, rather than staying fixed forever.

Why XAUUSD (Gold) Specifically

Gold is one of the more volatile instruments available on most MT5 brokers, which makes it a demanding test case for any martingale-style system. A martingale approach amplifies both the strengths and the risks of the base strategy, so testing and refining the learning engine on XAUUSDm (the Exness-style Gold symbol) exposes weaknesses faster than testing on a calmer forex pair would.

Two Bugs That Break Phase A/B Logic (And How They Were Fixed)

Building a self-learning EA surfaces edge cases that a simple indicator-following EA never hits. Two real examples from development:

1. Daily Profit Target Lock Bug

The function responsible for checking whether the daily profit target had been hit, IsDailyProfitTargetHit(), had a silent bug: if the input DailyTargetPercent was set to 0.0 (meaning "no target"), the calculated dailyTargetAmount also came out to zero. Because zero technically satisfies "profit has reached the target," the EA's internal flag tradingStoppedForDay would get set to true — permanently, since nothing ever reset it. The fix was a simple guard clause: if dailyTargetAmount <= 0, the function returns false immediately, so a disabled target can never accidentally lock trading for the day.

2. Stale Global Variable Peak Equity Bug

The EA stores a rolling peak-equity value in an MT5 Global Variable (EA_PeakEquity_778001_XAUUSDm) to power its drawdown protection guard. During testing, this variable was found holding a stale historical peak from a previous test run, which meant the drawdown guard was comparing current equity against an outdated, unrealistically high number — triggering false drawdown stops. Clearing and correctly re-initializing this Global Variable at EA startup resolved the issue.

Both bugs share a common lesson for anyone building or debugging a similar EA: default/zero values and persisted Global Variables need to be explicitly handled, or they silently corrupt logic that only shows symptoms days later.

Key Takeaways for Traders Testing Similar EAs

• Don't judge a self-learning EA's performance during its data-collection phase — it is not optimized yet by design.
• 300 trades is a reasonable minimum sample size before trusting any adaptive adjustment layer on a volatile symbol like Gold.
• Always test daily-target and drawdown-guard logic with edge-case inputs (zero, negative, first-run) — these are the conditions that break martingale-based EAs in live trading, not the "normal" cases.
• MT5 Global Variables persist across restarts and even between different test runs on the same terminal — always verify they're correctly scoped and reset for a new instance.

This EA is under active development in parallel with an EMA-crossover-based system, so future updates will cover how Phase B's adaptive behavior performs once enough live trade data accumulates.