Free Optimization Guide

Range Brain AI Optimization Guide

CCI Momentum Classifier (Continuation + Failed-Breakout Fade)

This guide gives you EXACT Start / Step / End values to type into the MetaTrader 5 Strategy Tester optimizer for every tunable input, plus the order to optimize them in and how to tell a real edge from a curve-fit.

HOW THE SYSTEM WORKS (so the settings make sense)

Range Brain AI trades range breakouts BOTH ways, decided by a CCI momentum classifier at the moment each breakout happens:

Builds a price range during a session window you choose.
When price closes beyond the range, the EA checks CCI's direction right then: still moving further the same way as the break means momentum is genuinely building, so it trades the breakout as a CONTINUATION, immediately, at market.
CCI already turning back means the breakout looks like it's failing, so the EA arms a FADE in the opposite direction instead, and waits for CCI to reach an extreme and cross back, confirmed by a deterministic neural net (perceptron), before entering at market.
Each side re-arms independently. A fade stays armed and can fire again on the next fresh CCI extreme-and-cross-back, with no requirement that price return inside the range first. A continuation trade locks out further attempts until price closes back at/past the boundary it broke - only then does the next breakout get classified fresh.
UseMomentumClassifier = false runs the EA in pure fade-only mode (every breakout arms a fade, exactly like earlier versions), useful as a baseline to compare the classifier against.

Two things to remember:

This is a range-breakout tool that needs a REAL consolidation range to classify from. Pick instruments/sessions that actually range during your chosen window - a session window too short to form a meaningful range (e.g. one bar) starves the classifier of a real momentum read and skews results toward one side.
The neural net is DETERMINISTIC (no random numbers): same data + settings = same result every time, so the backtest reproduces live. Verify this once (see "Reproducibility check").

IMPORTANT: WHAT "POINTS" MEANS

All *Points inputs (StopLossPoints, TakeProfitPoints, etc.) are in POINTS, not pips.

5-digit FX (e.g. EURUSD 1.23456): 10 points = 1 pip. So StopLossPoints 150 = 15 pips.
Indices / metals / crypto: 1 point = the last price digit; typical moves are much larger, so scale the ranges below up accordingly.

Set the point ranges to your instrument's normal bar size, not blindly.

STRATEGY TESTER SETUP

Data: real tick data with real spreads (Dukascopy / Quant Data Manager). Garbage data = garbage results.
Modelling: "Every tick based on real ticks".
Period: 2+ years. Hold back the most recent ~25-30% for out-of-sample validation (or use MT5 Forward = 1/4).
Criterion: "Balance + max Profit Factor" or a Custom/Recovery criterion.
Optimization: "Fast (genetic)" once the grid gets large.

REPRODUCIBILITY CHECK (do once, first): Run the SAME single backtest twice. Results must be identical. If not, stop - something is wrong. This determinism is what makes the whole guide trustworthy.

NOTE: in the tables below, a BOOL is optimized as Start 0, Step 1, End 1 (tests both false and true). Tick its checkbox in the optimizer.

THE OPTIMIZATION TABLES (Start / Step / End)

Type these into the optimizer's Start/Step/Stop columns. Don't optimize every group at once - follow the PHASES section after the tables. Ranges assume 5-digit FX; scale *Points for other instruments.

GROUP A: CORE EDGE (optimize these first)

Input	Start	Step	End	Notes
RangeStartHour	6	1	16	broker server time
RangeStartMinute	0	30	30	on-the-hour vs half-hour
RangeEndHour	8	1	18	must be after start
RangeEndMinute	0	15	45	on-the-hour/quarter increments, incl. default 45
CCIPeriod	10	4	34	14/18/22/26/30/34...
CCIOverbought	100	25	200	fade trigger (up-breaks)
CCIOversold	-200	25	-100	keep symmetric with above
StopLossPoints	250	250	1500	the "break resumed" stop
TakeProfitPoints	250	250	3000	the reversion target
MaxTradesPerDay	1	1	4	0 input = unlimited (don't optimize 0 here)

(MaxPositions default is 1. Both re-arm paths respect it as a hard cap regardless of how many times they'd otherwise fire - raise it only if you deliberately want more than one open position at a time.)

GROUP B: MOMENTUM CLASSIFIER

Input	Start	Step	End	Notes
UseMomentumClassifier	0	1	1	bool: fade-only vs classifier on

(Compare UseMomentumClassifier=true against a locked Group A run with it forced false - that false run is your fade-only baseline from earlier versions. Keep the classifier only if it clearly improves the result over that baseline.)

GROUP C: PERCEPTRON (leave default unless clearly better)

Input	Start	Step	End	Notes
Coefficient	0.05	0.05	0.20	activation scaling
LearningRate	0.05	0.05	0.35	backprop step
BuyTargetOutput	0.30	0.30	1.50
SellTargetOutput	-0.40	0.10	-0.20

(This net is a confirmer for the fade side only - it does not affect continuation entries. Over-tuning it is a classic curve-fit. Optimize only with a tiny grid, and prefer the defaults 0.1 / 0.3 / 1.5 / -0.3 when results are close.)

GROUP D: TRADE MANAGEMENT (optional add-ons, off by default)

Input	Start	Step	End	Notes
StopMoveToBE	0	1	1	bool: break-even move
BreakEvenTriggerPoints	50	50	300	only if StopMoveToBE on
AddBreakEventPoints	0	5	20
UsePartialClose	0	1	1	bool: partial close
PartialTriggerPoints	100	50	400	only if UsePartialClose on
ClosedVolume	25	25	75	percent to close
UseTrailingStop	0	1	1	bool: trailing stop
TrailingStart	150	50	500	only if UseTrailingStop on
TrailingDistance	50	50	300
UseCloseTime	0	1	1	bool: time-based exit
CloseHour	16	1	23	only if UseCloseTime on

(CloseMinute: leave 0.)

SET, DON'T OPTIMIZE (choose to taste; they scale results, not edge)

MoneyManagement: pick ONE (MM_FIXED_RISK recommended). Not optimized.
RiskPercent: your risk per trade (e.g. 0.5 - 1.0). Scales P&L, not the edge - set it, don't optimize it.
FixedLots: only used if MoneyManagement = MM_FIXED_LOT.
DecreaseFactor: leave default.
MaxSpreadPoints: reject entry if spread exceeds this (points; 0 = disabled). Set per symbol/broker, don't sweep.

DO NOT TOUCH (visual / identity)

MagicNumber, DashboardX/Y, ShowDashboard, all color inputs, all Show*/Enable* visual toggles, PulseSpeed, ArrowSize, the Tokyo/London/NY text examples.

PHASES (the ORDER to optimize - don't do all groups at once)

Optimizing every group together explodes the search space and curve-fits. Lock each phase's winner before the next.

PHASE 1 - CORE EDGE (Group A only) Optimize the range window + CCI + SL/TP together with UseMomentumClassifier forced false (fade-only baseline). Keep Groups C/D at defaults (all bools off). This finds where the real edge lives. Lock the winning Group A values.

PHASE 2 - MOMENTUM CLASSIFIER (Group B) With Group A locked, flip UseMomentumClassifier to true and compare against your Phase 1 (fade-only) result. Keep it only if it genuinely improves the result - it's meant to add continuation trades on top of the fade edge, not replace it.

PHASE 3 - TRADE MANAGEMENT (Group D, optional) With A+B locked, test one management feature at a time (BE, then partial, then trailing). Add a feature only if it clearly helps. Most clean systems do fine with none of these - don't force it.

PHASE 4 - PERCEPTRON (Group C, optional, light) Only if you want the last few percent on the fade side. Tiny grid, prefer defaults.

After each phase: pick from a STABLE CLUSTER (see below), then re-run a single backtest with the locked values to confirm.

READING RESULTS - CHOOSING A WINNER

Don't sort by Net Profit alone. Look at:

Profit Factor > 1.2 (>1.5 strong)
Recovery Factor (net profit / max DD) - rewards smooth equity
Max Drawdown % - can you stomach it live?
Trade count - 100+ over the test is ideal; <30 is noise
Win rate WITH avg win/loss - 40% win + big winners can be great

THE CLUSTER TEST (most important): In the optimization results, the best settings should be SURROUNDED by other good settings. If your #1 row's neighbors are losers, it's a curve-fit - discard it and pick from a broad plateau of good results. A slightly-lower but robust setting beats a fragile peak.

OUT-OF-SAMPLE VALIDATION (the step that saves money)

Optimize on the older data (e.g. 2022-2023).
Take the chosen winner UNCHANGED and run it as a single backtest on the held-out recent data (e.g. 2024-now).
If it stays profitable with similar character, it's a real edge. If it collapses, it was curve-fit - go back and pick a broader, coarser setting. MT5 Forward mode (1/4) automates this.

PER-SYMBOL / PER-TIMEFRAME + GOING LIVE

Repeat Phases 1-2 separately for EACH symbol and timeframe. There is no universal setting - that's correct for a range-breakout tool.
Favor instruments/sessions that form a real consolidation range; lower TF = more trades + more noise, and a range that's too short (too few bars) starves the classifier of a meaningful CCI read.
Confirm your broker's server time matches your range window.
Because the EA is deterministic, the validated backtest is an honest preview of live. Demo-confirm, then scale up.
No martingale / grid / averaging. Risk is fixed by your SL and money-management settings. Keep it that way.