Gold Neuron EA Setup Guide — Broker Time Zones, Risk Presets, and Why the Defaults Are the Whole Setup

Everything a Gold Neuron buyer needs to configure correctly, in one page: the one broker setting that actually matters, what each risk preset measured on real ticks, and why you do not need a set file at all.

Gold Neuron EA (MT5): https://www.mql5.com/en/market/product/187329

Gold Neuron decides trade direction once per day with a gradient-boosting model embedded inside the EA as an ONNX resource (13 market features, no external files, no DLLs, works offline). Every position opens with a fixed take-profit and stop-loss. No martingale, no grid, no averaging. Because the whole configuration ships as the EA's default inputs, "setup" is really three checks — and two of them are usually already correct.

Step 1 — Attach to XAUUSD M15. The defaults are the published configuration.

Every published backtest figure for this EA was measured on the shipping binary with its default inputs. There is no secret set file, and you do not need to build one: risk selection is a single dropdown (RunMode), and everything else defaults to the tested values. Attach to a XAUUSD (gold) M15 chart, allow algo trading, and the EA is in the published configuration.

Step 2 — The one setting that actually matters: your broker's server time.

The AI features are computed in UTC. The EA converts your broker's server time to UTC automatically, and this conversion is the single most common thing to get wrong — a one-hour mismatch silently degrades the model's inputs.

- On EET-style servers (UTC+2 in winter, UTC+3 in summer — the majority of MT5 brokers): the defaults are already correct. Do nothing.

- On brokers whose server clock is fixed year-round at GMT+0 (some raw-spread brokers): set GMTWinterOffset=0 and AutoDST=false.

- Not sure which you have? Compare your terminal's Market Watch clock with UTC right now: +2 or +3 hours ahead means EET-style (leave defaults), equal to UTC means GMT+0 (change the two inputs).

Step 3 — Pick a risk preset with real numbers, not adjectives.

All four presets were measured on the shipping binary, 2025 full year, 99-100% modelling-quality real ticks (Dukascopy XAUUSD M15, $10,000 start, 1:500):

- Defensive (x0.5): +688%, max equity drawdown 14.6%

- Standard (x1): +5,170%, max equity drawdown 27.7%

- Aggressive (x2): +75,302%, max equity drawdown 22.9%

- Ultra (x2.5, the shipping default): +102,363%, max equity drawdown 52.4%

Two honest caveats. First, the model is trained on data through July 2026, so backtests inside the trained window measure fit to that history — they are not a forward guarantee. Second, those figures come from a tight-spread feed; on wide-spread standard-account feeds Ultra's drawdown gets much deeper, and we recommend Standard or Defensive there. If you run a standard account rather than a raw/zero account, start with Standard.

What v2.0 changes: the drawdown survival brake

Version 2.0 added a drawdown throttle (UseDDThrottle, on by default). When account equity falls 30% below its peak, lot sizing scales down progressively, reaching a 5% floor at 55% drawdown. In out-of-sample years before the training window (2017, 2019) the old build lost the whole account; v2.0 ends those same years around -44% to -54% — a serious loss, but a surviving account. The brake is why the Ultra default is usable at all; leave it on.

Why does it open three positions at once?

The three entries share one take-profit and one stop-loss — economically they are a single position, split for execution reasons: each order passes its own margin check (so a partially affordable signal is partially taken instead of rejected), and at large account sizes a single order would exceed the broker's per-order volume cap. We tested the alternatives — adding to the position only after price moves against it, and pyramiding only after it moves in favour — and both were dramatically worse on the same data (the win-rate structure means most winning days never pull back, so staged entries end up fully sized only on the bad days). Simultaneous entry is not an accident; it is the measured optimum.

What a normal week looks like

One decision per day at a fixed London-session time (04:15 UTC). On days blocked by the news filter, or when the model's probability does not clear the threshold, it simply does not trade — zero-trade days during big-event weeks (FOMC, CPI) are the system working, not failing. The dashboard on the chart shows the current state and the last signal probability.

Inputs you should not touch

EntryUtcHour, TP_USD, SL_USD and Splits define the strategy itself — they are part of what was tested, not tuning knobs. Changing them puts you outside every published number. The inputs meant for you are RunMode (risk), the two timezone inputs above if your broker needs them, and the notification switches.

Updates and support

The model is retrained periodically and shipped as a free update — the What's New tab shows the model date of each build; always run the latest. Questions and setup help: the product's Comments tab. We read it daily.