Pipsgrowth EX13001 RSI_MA

EX13001 RSI_MA is a mean-reversion pullback EA that hunts for the moment RSI(21) snaps back through its extreme line while the slow moving average still confirms the prevailing trend direction. The setup is small in count — one oscillator, one bias filter, one higher-timeframe agreement, one regime classifier — but the gate that decides whether a signal becomes a trade is wider than the signal itself, stacking a 7-state market classifier, an ATR-anchored R/R check, a session filter, a daily loss circuit breaker, and a three-bar post-loss cooldown before the order is ever sent. The order it does send is then managed in three layers: a 50% partial close at +1R, a break-even-plus lock at +1R, an ATR trailing stop that ratchets behind price, a 200-bar time-based exit, and a hard opposite-signal exit when RSI walks back to the other extreme.

The signal core sits in ComputeSignal(). It pulls two RSI bars from the chart timeframe and one MA bar. The long trigger requires that the previous bar's RSI was at or below the oversold line (100 − InpRSIThreshold, i.e. 30 by default), the current bar's RSI is above that line, AND the previous bar's close finished above the slow MA(21) — proving the pullback is happening into an uptrend, not into a falling market. The short trigger is the mirror. When the cross fires, the function assigns a confidence value based on how far RSI has already overshot the threshold (50 + 2 × excursion, capped at 100) and tags the reason for the audit log.

Confirmation runs in ConfirmEntry() and adds four extra hurdles. First, the H1 EMA(50) must agree — for a long signal the H1 close must be above the H1 EMA, for a short signal it must be below. The H1 timeframe is hard-coded via the HTF_TREND_TF define and the EMA(50) handle is created on OnInit. Second, the current spread must be below InpMaxSpreadPoints (30 by default) AND not more than 2× the rolling 50-tick spread average that UpdateSpreadRolling() keeps in a running sum. Third, the broker-server hour must be inside the 07:00–21:00 window and the regime must not be CHOPPY. Fourth, the projected R/R must be at least InpMinRR (1.2) — using SL distance = 1.5 × ATR(14) and TP distance = 2 × SL.

The regime classifier in ComputeRegime() is the most distinctive piece of plumbing in the EA. It does not lean on a single trendiness metric. It pulls 102 bars of ATR, 102 bars of Bollinger Band width, and 3 bars of ADX, then computes two ATR percentiles (20th and 80th) and one BB-width percentile (25th) over the last 100 bars. From those it derives boolean flags: COMPRESS (ATR ≤ 20th-pctile AND BB-width ≤ 25th-pctile), EXPAND (ATR ≥ 80th-pctile), then folds them with ADX bands of 25 (strong) and 18 (weak). The output enum has seven states — STRONG_TREND, WEAK_TREND, RANGE, BREAKOUT, EXPAND, COMPRESS, CHOPPY — and the entry path explicitly rejects CHOPPY at two places (OnTick pre-signal and ConfirmEntry). The classifier runs every closed bar regardless of whether a signal exists, so the gate has fresh context by the time ComputeSignal() returns.

Risk and capital control run in two separate functions. ComputeEffectiveCapital() returns the account equity, capped at InpCapitalCapAmount when set; the cap can be set to zero to disable. If the cap is enabled and effective capital drops below InpCapitalCapFloor (50 by default), NoTradeGate() refuses to open new positions. ComputeLots() then sizes each trade off the projected SL distance: risk_money = effective_capital × 0.5%, loss_per_lot = (sl_distance / tickSize) × tickValue, lots = risk_money / loss_per_lot, then clamped to the broker's volume step/min/max. The same effective-capital number is reused to enforce a 3% daily loss circuit breaker (TodayLossRatio() walks deal history from midnight, sums profit + swap + commission for the day's exits tagged with InpMagic, and returns the loss as a percentage).

Position management lives in ManageOpenPositions(), which runs on every tick as long as the ATR handle returns a value. It walks PositionsTotal in reverse, filters by magic and symbol, and for each surviving position applies five steps in order. Step one is break-even: once price has moved InpBE_RMultiple × SL-distance (= 1.0R by default) in favor, the SL is set to open_price + 2 points (BE_OFFSET_POINTS = 2) — a true breakeven-plus lock, not a flat breakeven. Step two is ATR trailing: SL becomes bid − 2.5 × ATR for buys, ask + 2.5 × ATR for sells, but only if the new SL tightens (greater than current SL for buys, less for sells) and stays farther than the broker's stops-level. Step three is partial TP at PARTIAL_TP_R = 1R: 50% of the position is closed and the remainder is left to run to FULL_TP_R = 2R. Step four is a time exit at MAX_BARS_IN_TRADE = 200 bars on the current timeframe. Step five is the opposite-signal exit: if the position is a long and current RSI is now ≥ InpRSIThreshold, the position is closed; the mirror closes shorts. After every step the function calls ModifyPositionSL(), CloseThisPosition(), or ClosePartial(), each of which routes through TryClose/TryClosePartial/TryModify_EX13001 — three-attempt retry loops with 200ms back-off on REQUOTE/TIMEOUT/PRICE_OFF/PRICE_CHANGED for closes and only REQUOTE/TIMEOUT for modifies.

The 19 input parameters are organized in six groups: identity (magic 22213001, comment, dry-run flag defaulting to TRUE, kill switch), capital cap (cap amount, floor), risk & sizing (risk percent 0.5%, daily loss 3%, max concurrent 1, min R/R 1.2), signal (RSI period 21, RSI threshold 70, MA period 21), regime/confirm/exec (min ADX 18, max spread 30, slippage 10), and exit/manage (ATR-SL mult 1.5, BE R-multiple 1.0, trail ATR mult 2.5). The dry-run flag defaulting to TRUE is deliberate — it forces the EA to print every intended order to the log without sending it, so a user can observe entries, exits, and modifications on a live chart before flipping the switch. The kill switch forces a close-only mode without needing to remove the EA from the chart. The session filter is server-time 07:00 to 21:00 with a 1-hour Friday wind-down (Friday from 20:00 onward is blocked), not a London/NY overlap, so the EA's behavior depends on the broker's server clock. The OnTester custom criterion returns (net_profit × profit_factor) / (1 + equity_dd_percent) when total trades ≥ 30, rewarding strategies that pair positive expectancy with low drawdown. There is no martingale, no grid, no basket, and no news blackout module — the NEWS_BLACKOUT_MINUTES define is wired but the CalendarValueHistory call is left as a placeholder.