Pipsgrowth EX09006 MeanReversion

Pipsgrowth EX09006 is a candlestick-pattern mean-reversion EA. Rather than relying on moving-average crossovers or oscillator thresholds alone, the entire entry thesis is built around a single geometric bar shape — the Hammer for buy setups and its bearish twin the Hanging Man for sells — with a Commodity Channel Index confirmation standing in as the momentum filter. The pattern is detected on the most recent fully closed bar (shift = 1) using straightforward ohlc geometry, not pattern-recognition libraries. The bar's body has to sit inside the upper third of the range, and the lower wick has to be at least twice as long as the body and longer than the upper wick. The MAVal(1) gate then forces a context dependency: a Hammer is only valid when the closed bar's close is below the SMA(5) — i.e. the reversal is forming in a short-term dip, not after the trend has already turned. The Hanging Man mirror fires only when the close is above the SMA(5).

With the pattern shape verified, the EA looks at CCI(37) — a longer period than the typical 14, which means fewer signals and more weight on the most recent 37 closes. A buy requires CCI <= -100; a sell requires CCI >= 100. The further CCI sits past those thresholds, the higher the confidence score: 50 + |CCI - threshold| × 0.4, capped at 100. The score is logged but does not gate entries — the thresholds alone are the binary filter. The raw pattern direction (+1 buy, -1 sell, 0 none) is what propagates forward.

The regime classifier in ClassifyRegime() runs on every tick and bucketizes the market into one of seven states: STRONG_TREND, WEAK_TREND, RANGE, BREAKOUT, COMPRESS, EXPAND, or CHOPPY. The decision tree uses ADX(14), a Bollinger Bands width z-score (period 20, deviation 2.0, computed by hand over a 30-bar window), and an ATR(14) percentile rank (100-bar lookback). Mean reversion is only allowed in three of the seven states — RANGE, COMPRESS, and WEAK_TREND — which the source comment makes explicit: 'Block in StrongTrend/Breakout/Expand.' This is the single most consequential guardrail: it prevents the EA from fading trends it cannot win against.

Three confirmation checks must pass before a SendOrder call: HTF agreement, spread cap, and minimum R:R. The HTF filter pulls an EMA(50) on H1 and requires the signal bar's close to sit on the opposite side of it from the trade direction — a Hammer buy wants the close below the H1 EMA, meaning the higher timeframe is in a downtrend phase where a counter-trend bounce is more likely to be the start of a pullback, not the resumption of a move. The spread filter accepts the bar if current spread is below MAX_SPREAD_POINTS (35) or, if InpMaxSpreadOverAvg > 0, below three times a rolling average spread. R:R is enforced at MIN_RR_RATIO = 1.2, so even when a 1.67:1 nominal setup is in play (TP_ATR_MULT 3.0 / SL_ATR_MULT 1.8), the broker's stops-level can sometimes squeeze the actual R:R below the floor and block the trade.

SL and TP are ATR-driven: SL distance = ATR(14) of the closed bar × 1.8, TP distance = ATR(14) × 3.0, giving roughly 1.67:1. Lot sizing uses the standard risk formula in CalcLotByRisk(): riskMoney = effective_capital × InpRiskPercent / 100, divided by perLotRisk = slPriceDistance × tick_value / tick_size, then clamped to volume step / min / max and capped at InpMaxLots. With InpRiskPercent at 0.5 and a typical 200-point XAUUSD SL on M5, a $100 account is sized at the broker's minimum lot, which is why the recommended minimum deposit sits at $100 — anything below that and the position simply will not size up.

The management layer runs on every tick once a position is open. ManageOpenPosition() implements three independent exit mechanisms that can all fire. First, a one-shot break-even at 1R — when unrealized profit reaches 1× the original stop distance, the SL is moved to entry + 2 points and locked there, no further ratcheting. Second, a 50% partial close at 1R, which only fires if the remaining volume is at least twice the broker's minimum lot so the residual position stays tradeable. Third, an ATR(14) × 2.5 trailing stop with a 10-point step that only ever moves the stop in the favorable direction. The exit stack also includes an opposite-signal exit (Hanging Man while long, or Hammer while short), a 60-bar time stop on M5, and a hard regime-change close when the market flips into STRONG_TREND, BREAKOUT, or EXPAND — the mean-reversion thesis is considered invalidated in those states.

Capital and risk are gated by a stack of guards. The no-trade check evaluates: kill switch, broker trade mode, terminal connection, capital cap floor ($50 default), effective capital floor, server-time session (7:00–20:00), weekend block (Sun=0, Sat=6 in MQL day-of-week), spread cap, a ±30-minute news window around each hour boundary, regime permission, daily trade cap (5), 5-bar cooldown after the last loss, and the InpMaxConcurrent = 1 limit. Daily realized PnL is computed in RealizedPnLToday() from the deal history filtered by magic and symbol, and if it breaches -3% of effective capital, any open position is force-closed and new entries are blocked for the rest of the day. The weekly limit is 6% with the same mechanism.

The capital cap is the most under-appreciated input. InpCapitalCapAmount = 0 disables it (uses full equity); setting it to e.g. 5000 means the EA will never risk more than 0.5% of $5000 — useful for traders running multiple EAs on one account. The effective capital is recomputed on every tick in UpdateEffectiveCapital() as MathMin(cap, equity).

InpDryRun defaults to true. The EA will log '[DRYRUN] BUY lots=0.01 sl=... tp=... reason=SIGNAL' on every entry attempt and not place a real order until you flip this to false. This is a safe default for first-time deployment but a footgun for anyone who copies the EA onto a chart and walks away. InpKillSwitch is a one-click hard stop — when true, every entry is blocked and the reason printed is 'KILL_SWITCH.'

The OnTester() function returns a custom fitness of (net profit × profit factor) / (1 + drawdown%), with a minimum 30-trade floor. Zero trades, zero profit factor, or zero drawdown all return 0.0, so the optimizer is rewarded for net profitability, penalized for drawdown, and refuses to score under-sampled backtests. This is a quality-weighted criterion, not a net-profit-only one — a 2,000-trade sample with 1.3 PF and 12% DD will outrank a 50-trade sample with 3.0 PF and 4% DD in the optimizer's ranking.

TryClose_EX09006, TryClosePartial_EX09006, and TryModify_EX09006 each retry three times on REQUOTE / TIMEOUT / PRICE_OFF / PRICE_CHANGED with 200ms (or 100ms on modify) backoff, and the trade-send loop retries once on the same transient set with a 150ms sleep and a symbol refresh. The fills path is auto-detected via SetTypeFillingBySymbol in OnInit. The EA does not implement any MQL5 built-in news filter or any external feed — the ±30 minute news guard is a generic hour-boundary window that will fire even when no news is scheduled.

In a backtest, the EA's behavior on XAUUSD M5 should be characterized by low trade frequency — the pattern + CCI double filter is restrictive — and tight, fast winners. The partial-close at 1R typically resolves a third to half the position quickly when the bar bounces as expected; the trailing stop takes over from there. Losing trades usually exit on the opposite-signal rule or the 60-bar time stop rather than the original SL, which means the realized loss distribution is narrower than the SL would suggest, but the win rate is also lower than a typical trend-following EA. Run the Strategy Tester with 'Every tick based on real ticks' and tick-value confirmation for the lot sizing to match live behavior.