Pipsgrowth EX13002 RSI_MA

Pipsgrowth EX13002 RSI_MA is a contrarian pullback engine that buys the first oversold reaction and shorts the first overbought reaction, but only when the higher-timeframe trend agrees and the market is showing a tradable pulse. It runs on M5 or H1 bars of any XAUUSD or FX-major chart and is, in the source author's words, broker-portable. The compiled logic in Pipsgrowth_com_EX13002.mq5 is a strict 12-layer pipeline — regime detection, signal generation, entry, confirmation, no-trade gating, capital cap, risk, sizing, manage, exit, scaling, OnTester — wired into 30 functions and 19 user inputs plus 15 internal defines. The handle set is tight: iRSI(21) on PRICE_OPEN, iMA(21, MODE_SMA, PRICE_OPEN) on the chart timeframe, iATR(14), iADX(14), iBands(20, 2.0, PRICE_CLOSE) for width context, and an iMA(50, MODE_EMA, PRICE_CLOSE) on H1 for the higher-timeframe trend filter.

The signal itself is a textbook momentum-pullback cross. ComputeSignal reads the closed-bar RSI(21): a long fires when the previous bar printed at or below 100 minus InpRSIThreshold (30 by default) and the current closed bar climbed back above that oversold line with the previous bar's close sitting above the SMA(21) bias line. A short is the mirror — previous bar ≥ InpRSIThreshold (70) crossing back below, with the prior close below SMA(21). When a signal fires the function also computes a confidence score from the RSI excursion past the trigger, capped at 100, which is logged but not currently used to gate entries. The bias-MA guard is what stops the EA from buying every RSI pop in a strong downtrend — both the cross direction and the close-vs-SMA side have to agree.

That signal then has to survive a four-hurdle confirmation stack in ConfirmEntry before any order is built. First, an H1 EMA(50) agreement: the H1 closed bar must be on the same side of the 50-EMA as the proposed trade. Second, a spread filter that not only enforces InpMaxSpreadPoints=30 but also blocks trades when the current spread is more than 2× the rolling 50-tick average stored in g_spread_rolling_sum / g_spread_rolling_count — this is the EA's defence against latency spikes and post-news spread blowouts. Third, a server-time session gate from SESSION_START_HOUR=07:00 to SESSION_END_HOUR=21:00, with Friday from 20:00 onward blocked (close-proximity guard). Fourth, a reward-to-risk check: with the ATR-driven stop distance set to 1.5 × ATR and the take-profit at FULL_TP_R=2.0 × the stop distance, the realised R/R must clear InpMinRR=1.2 — note that the final TP/RR pair is locked in even when the test is satisfied, so this acts as a sanity filter rather than a dynamic target.

Regime is computed once per closed bar in ComputeRegime, which sorts the last 100 ATR(14) values and last 100 Bollinger-Band-width samples to derive the 20th, 80th and 25th percentiles. The classifier reads the current ATR against those percentiles to flag COMPRESS (ATR ≤ 20th percentile AND BB-width ≤ 25th percentile) and EXPAND (ATR ≥ 80th percentile) states, then folds those with ADX thresholds to assign one of seven regimes: STRONG_TREND (ADX ≥ 25), WEAK_TREND (ADX ≥ 18), RANGE, BREAKOUT, COMPRESS, EXPAND, or CHOPPY (ADX < 12). The only regime the entry logic explicitly rejects is CHOPPY; STRONG_TREND, WEAK_TREND, RANGE and EXPAND are all considered tradable. On top of regime, OnTick also enforces InpMinADX=18 as a hard pre-signal gate so the EA does not waste a signal in a directionless market.

Risk and sizing are computed in ComputeLots from effective capital, not raw equity. ComputeEffectiveCapital takes the current equity, subtracts nothing, then optionally caps it to InpCapitalCapAmount if that input is set above zero — at default InpCapitalCapAmount=0 the cap is disabled and effective capital equals equity. The result is cached for 5 seconds (g_eff_capital_last_calc) to keep sizing consistent across a bar but responsive enough to track drawdown. Lots are then solved as (effective capital × InpRiskPercent=0.5%) divided by the loss-per-lot at the current ATR stop distance, finally clamped to the broker's lot step and bounds. ComputeSLTP applies the same 1.5 × ATR stop distance, with a safety pad of stopsLevel + 5 points so the broker never rejects the order as too close, and rounds the result to the symbol tick size via NormalizePrice.

Layer 7 of the no-trade gate is the deal-event tracker TrackRecentDealEvents, which scans history from midnight to now for the EA's own InpMagic-tagged DEAL_ENTRY_OUT / DEAL_ENTRY_INOUT records, sums profit+swap+commission, and feeds two distinct state machines. The first is the daily-loss state machine: when g_today_realized_pnl (computed by ComputeTodayRealizedPnL on the same history scan) drops the account by ≥ InpDailyLossLimitPct=3% of effective capital, all new entries are blocked for the rest of the server day. The second is the consecutive-loss cooldown: every negative deal increments g_consecutive_losses, sets g_cooldown_until = now + 3 × PeriodSeconds (COOLDOWN_BARS_AFTER_LOSS=3), and only a positive deal resets the counter to zero. This is a real three-bar pause on the chart timeframe, not a fixed-seconds cooldown.

Position management runs on every tick in ManageOpenPositions and follows a fixed five-step order. Step 1 is break-even-plus-2 lock: once price has moved in favour by InpBE_RMultiple=1.0R, the stop is moved to entry plus BE_OFFSET_POINTS=2 ticks. Step 2 is an ATR trailing stop at InpTrailATRMult=2.5 × ATR that only ratchets in the favourable direction, only fires when the new stop is at least one tick beyond the current stop, and only modifies when the remaining distance to price exceeds the broker's stops level. Step 3 is a partial take-profit at PARTIAL_TP_R=1.0R closing PARTIAL_TP_RATIO=50% of the position — PartialAlreadyTaken is a no-op stub so this branch effectively re-fires every time ManageOpenPositions runs after TP1, which can over-close; in practice the partial-close path is only safe when volume still allows a non-zero remainder. Step 4 is a time exit at MAX_BARS_IN_TRADE=200 bars. Step 5 is the opposite-signal exit: a long is closed when RSI re-crosses up through OB, a short when RSI re-crosses down through OS.

Two more behaviours are worth highlighting. First, InpDryRun defaults to true — the EA is shipped in a logged-only mode where ComputeLots, ComputeSLTP and SendOrder all run normally but SendOrder short-circuits to PrintFormat('DRYRUN signal=...') and returns success. The only way to go live is to flip InpDryRun to false in the inputs dialog. Second, InpKillSwitch=true is a soft stop: when on, NoTradeGate returns true on the very first check, which blocks new entries but leaves the manage-open loop running so existing positions still get their break-even, trail, partial and time-exit treatment. SendOrder itself retries once on REQUOTE, PRICE_OFF or TIMEOUT before declaring failure. Close / close-partial / modify paths use the per-EA retry helpers TryClose_EX13002, TryClosePartial_EX13002 (both: 3 attempts, 200ms Sleep, retry on REQUOTE / TIMEOUT / PRICE_OFF / PRICE_CHANGED, break on any other retcode) and TryModify_EX13002 (3 attempts, 100ms Sleep, retry on REQUOTE / TIMEOUT only — PRICE_CHANGED and PRICE_OFF break the loop, a deliberate asymmetry with the close helpers).

OnTester returns 0 when the strategy traded fewer than 30 times — useful for filtering out under-sampled runs in the optimiser — and otherwise returns (net × profitFactor) / (1 + equity-drawdown-percent). When the drawdown comes back at zero or negative the formula falls back to the simpler net × pf. The OnTester formula is the only way the strategy self-evaluates during MT5's genetic pass; it explicitly rewards trades that produce profit per unit of drawdown, not raw net.

The backtest profile of this EA on XAUUSD M5 is straightforward: expect a low-to-moderate trade frequency dominated by the post-pullback continuation trades, occasional re-entries on the same bar if the signal reverses quickly, and visible drawdown clusters during the COMPRESS / CHOPPY regimes that pass the spread and ADX gates but ultimately don't trend. The PYRAMID_ENABLED=false define is a hard wall — there is no martingale, no grid, no averaging. The MAX_SYMBOL_EXPOSURE=1 cap means you cannot run a hedge or grid of this EA on the same symbol from the same terminal; either run it on a separate magic on a different symbol, or accept the one-position cap. The EA is suitable for accounts from $100 upward, but a $500+ balance gives the daily-loss state machine enough headroom to actually engage on a losing streak. Run on an ECN/RAW-spread broker with sub-30-point typical gold spreads, server time aligned so that the 07:00–21:00 server window overlaps the London and New York sessions, and verify the broker's GMT offset matches your chart's session labels before going live.