Aura Black Edition EA - neural networks in action

Aura Black Edition EA is designed to trade exclusively on Gold. According to the author, the robot uses neural networks in its trading algorithm. In particular, an artificial neural network with feed-forward (ANN), which is extremely difficult to verify - let's hope that the author is not disingenuous, although I very much doubt it. Aura Black Edition does NOT use dangerous trading methods such as martingale, grid, averaging, which means that it is relatively safe. At the same time, the adviser shows consistently profitable trading results from 2012 to this day, but for the period up to 2012, I have big questions. What prevented the Expert Advisor from trading profitably from 2003 to 2012? More on that below.

Сharacteristics of the Aura Black Edition MT4 EA
Platform: Metatrader4
EA version: 2.3
Currency pairs: Gold (XAUUSD)
Trading Time: Around the clock
Timeframe: D1
Minimum deposit: 500$ (per trading pair)

Trading algorithm
The author is trying to tell us about some kind of neural networks, but in my opinion everything is much simpler. The EA determines the direction of entering the market based on some indicators (possibly trend or channel ones) and opens a deal in their direction with the specified SL and TP, and then manages the open position using a trailing stop if the price goes in the right direction or closes the deal by Stop Loss if the price goes in the opposite direction.

And here is what the author says about his creation:

EA trained with a multilayer perceptron Neural Network (MLP) is a class of feedforward artificial neural network (ANN). The term MLP is used ambiguously, sometimes loosely to any feedforward ANN, sometimes strictly to refer to networks composed of multiple layers of perceptron's (with threshold activation). Multilayer perceptron's are sometimes colloquially referred to as "vanilla" neural networks, especially when they have a single hidden layer. An MLP consists of at least three layers of nodes: an input layer, a hidden layer and an output layer. Except for the input nodes, each node is a neuron that uses a nonlinear activation function. MLP utilizes a supervised learning technique called backpropagation for training. Its multiple layers and non-linear activation distinguish MLP from a linear perceptron. It can distinguish data that is not linearly separable.