Pipsgrowth EX15022 SMC-OrderBlock

Pipsgrowth EX15022 MarketPredictor is a research-flavored scalper whose trading signal is generated by a Monte Carlo price simulation rather than by the usual technical-indicator stack. The header comment advertises a 12-layer system (REGIME, SIGNAL, ENTRY, CONFIRM, NO-TRADE, CAPITAL CAP, RISK, SIZING, MANAGE, EXIT, SCALING, OnTester), but a careful read of the source shows that the live decision path reduces to two operations: every incoming tick, OptimizeParameters() rewrites the internal model coefficients, then PredictNextPrice() runs SimulatePrice() and forwards a single predicted price to ExecuteTrade(). The Sinusoidal, Fractal (FFT) and Sigmoid components that the header hints at are implemented as standalone functions — CalculateSinComponent, CalculateFractalComponentFFT, CalculateSigmoidComponent — but in PredictNextPrice their return values are computed and discarded (the source keeps the assignment lines commented out, leaving the Monte Carlo path as the only signal source wired into ExecuteTrade).

The Monte Carlo engine is the heart of the EA. SimulatePrice() seeds nothing itself, but OnInit() calls MathSrand((int)TimeCurrent()) so each EA session gets its own pseudo-random stream. Inside the loop, the function generates MonteCarloSimulations = 1000 draws. Each draw produces a random uniform factor between −0.5 and 0.5 by way of MathRand()/32767.0 − 0.5, multiplies that factor by the internal sigma variable (default 10.0), and adds the perturbation to P_t (the latest Bid). The mean of those 1000 perturbed prices is then returned as P_t1, the predicted next price. Because every draw is centred on P_t with a small symmetric perturbation, the average sits very close to the current Bid, so the only way the prediction can escape the ±sigma envelope is when MathRand bias happens to push the average over the threshold on a particular tick — which is exactly what ExecuteTrade() tests for.

The threshold logic in ExecuteTrade() is asymmetric. A buy is opened when P_t1 > currentPrice + sigma and no position is open on the symbol; a sell is opened when P_t1 < currentPrice − sigma and no position is open. The PositionSelect(_Symbol) guard means the EA never pyramids or hedges on the same chart: every fill is followed by a wait for the next entry signal. The orders themselves are raw OrderSend calls with TRADE_ACTION_DEAL, ORDER_TYPE_BUY or ORDER_TYPE_SELL, fixed volume 0.1, deviation 10 points, magic 22215022, and the comment 'Psgrowth.com Expert_15022'. The request does not populate request.sl or request.tp, so positions are placed with no stop-loss and no take-profit on the broker side. There is no CTrade wrapper used on the live entry path — the file declares CTrade g_trade_wrappers_EX15022 at the bottom, but ExecuteTrade bypasses it in favour of direct OrderSend.

OptimizeParameters() is invoked on every tick, not on a new-bar close, so the model coefficients are constantly re-anchored. The function calls CopyClose over the most recent 14 bars, computes a simple moving average, and assigns the result to the internal mu variable, overriding inputMu for the rest of that tick cycle. It also calls iATR(_Symbol, PERIOD_CURRENT, 14) and, if a positive ATR is available, sets alpha to 0.1 × ATR; otherwise alpha falls back to the input value inputAlpha. The other inputs — inputBeta (Fractal weight, default 0.1), inputGamma (Fractal damping, default 0.1), inputKappa (Sigmoid sensitivity, default 1.0) — are loaded once in OnInit into their corresponding internal variables and never touched again on the live path because the components that consume them are commented out. Adjusting those four parameters from the MT5 input dialog therefore only changes a value the live code never reads again after OnInit returns.

The CalculateFractalComponentFFT() function is the only place where the FFT machinery is wired in. It pulls 128 closing prices (N = 128, a power of two) with CopyClose, packages them into a Complex[] array (real part = close, imaginary part = 0), and calls FFT(data, false). FFT is a textbook recursive Cooley–Tukey implementation that splits the array into even and odd indexed sub-arrays, recurses into both, recombines them with complex twiddle factors MathCos/MathSin(2πk/n), and only divides by 2 in the inverse direction. After the transform, CalculateFractalComponentFFT computes the magnitude spectrum, scans indices 1..N/2 to find the dominant frequency (skipping the DC component at index 0), and returns beta × (maxIndex / N). The return value is the predicted-fractal price offset, but the live code does not use it.

CalculateSinComponent() and CalculateSigmoidComponent() follow the same pattern of being defined but inactive. Sin picks a random omega in [0, 2π] per call and returns alpha × sin(omega × t) where t is the bar count. Sigmoid fetches iATR(_Symbol, PERIOD_CURRENT, 14) and computes Delta_t / (1 / (1 + exp(−kappa × (P_t_local − mu)))). Both functions are reachable only from PredictNextPrice, and in PredictNextPrice the lines that would assign and combine them are commented out. The result is a file that reads like a research notebook: the math is all there, but the trading logic itself is the much simpler P_t + random-perturbation comparison against ±sigma.

The 3 retry helpers at the bottom of the file — TryClose_EX15022, TryClosePartial_EX15022, TryModify_EX15022 — are not invoked anywhere in the source. Each loops up to 3 times, calls g_trade_wrappers_EX15022.PositionClose / PositionClosePartial / PositionModify, retries on TRADE_RETCODE_REQUOTE, TRADE_RETCODE_TIMEOUT, TRADE_RETCODE_PRICE_OFF or TRADE_RETCODE_PRICE_CHANGED with 200 ms or 100 ms Sleep, and breaks on any other retcode. They exist as scaffolding for a future SL/TP management layer that the current build does not implement. There is no breakeven, no trailing stop, no time-based exit, no basket close, no equity stop, and no daily-loss cap. A position opened by ExecuteTrade stays open until the broker-side margin call or the trader intervenes manually.

The minDeposit recommendation in the database is $100 and the riskLevel is MEDIUM, which is a reasonable starting point given the 0.1-lot fixed size, the absence of protective orders, and the sigma threshold that gates entries. Users who want the EA to actually fire on a typical gold chart need to be aware that the default sigma = 10.0 is denominated in raw price units, not in pips, so on XAUUSD the ±10 envelope is roughly ±$1 of price movement and the Monte Carlo mean rarely escapes it. Tightening sigma to 0.5 or 1.0, or widening the prediction by adjusting MonteCarloSimulations and the random-factor range, is the practical lever — the input dialog is the only place to make those changes. A reader studying the source should treat the SMC-OrderBlock family label as historical and the 12-layer claim as design intent: what runs in the Strategy Tester is a Monte Carlo threshold-crossing system with no broker-side stops.