Rủi ro & lệnh kẹtRisk & stuck orders

1. Part 1 — Fixing the losses: Risk Management
2. Part 2 — Option 1 vs Option 2: handling stuck orders
3. Part 3 — v1.41.1 fork notes

From the ReportTester-11280950 backtest analysis:

• A 46-trade losing streak: -120,789 USD (almost the whole account)
• Poor R/R: average loss (-581 USD) ≈ 4.75× average profit (122 USD)
• Drawdown > 100%: Balance and Equity drawdown exceeded 100% → the account nearly reached zero
• Gross Loss > Gross Profit: 410,918 loss vs 309,895 profit

Logic:

• Track g_equityPeak (highest equity since the session started).
• Every tick: drawdown = (equityPeak - equity) / equityPeak * 100.
• When drawdown ≥ InpMaxDrawdownPercent or ≥ InpMaxDrawdownUSD:
  • Set g_maxDrawdownTriggered = true
  • If InpMaxDrawdownCloseAll = true: close all positions
  • If false: only block new orders (Tia/Clear Stuck may still close positions)
• Reset once equity recovers above the peak.

Purpose: avoid excessively deep drawdowns (>100%) like the one in the backtest.

Logic:

• Track g_balanceAtDayStart (balance at 00:00 server time).
• Every tick: profitInDay = Balance - balanceAtDayStart.
• When profitInDay <= -InpMaxLossPerDayUSD: set g_maxLossPerDayTriggered = true → block new orders.
• Reset at each new day (00:00 server time).
• Reset the flag when the day turns profitable (profitInDay > 0).

Purpose: cap risk per day, so one very bad day does not snowball into a losing streak.

Behavior:

• When a threshold is breached: block new orders (Init/Add/Lot Balance).
• Still allowed: Tia cluster close, Clear Stuck, Global TP/SL, Hedge (to reduce exposure and recover).

• Max Drawdown 80%: stop opening new orders when equity drawdown ≥ 80% from peak (avoiding >100% as in the backtest).
• Max Loss Per Day 5000 USD: stop opening new orders when the day's loss ≥ 5000 USD.
• CloseAll = false: only block new orders, do not close everything (so Tia/Clear Stuck get a chance to recover).

1. Lower maximum drawdown: from >100% to ≤80% thanks to the hard stop.
2. Fewer large losing streaks: Max Loss Per Day caps the daily loss, avoiding a 46-trade -120k streak.
3. Better R/R: Tia closes earlier (3 USD) and Clear Stuck acts earlier (60 pips) → smaller average loss, larger average profit.
4. Less time holding stuck orders: Clear Stuck with MaxLossPct 65% → more stuck orders cleared on profitable days.

1. Re-run the backtest on the same period (2026.01.19 - 2026.02.16) with the GoldenRobber_hybrid.set file.
2. Compare: did max drawdown fall below 80%? Is the large losing streak (-120k) still there? Did average loss shrink and average profit grow? Did Profit Factor improve from 0.75?
3. Adjust if needed: drawdown still high → lower InpMaxDrawdownPercent to 70% or 60%; still losing streaks → lower InpMaxLossPerDayUSD to 3000-4000; Tia closing too early → raise InpTiaProfitUSD to 3.5-4.0.

1. MQL5/Experts/GoldenRobber/GoldenRobber.mq5
   • Added the Risk Management inputs (lines ~127-133)
   • Added globals g_equityPeak, g_maxDrawdownTriggered, g_maxLossPerDayTriggered (lines ~177-179)
   • Initialized in OnInit() (lines ~778-780)
   • Added CheckMaxDrawdown() (after CheckGlobalSLClearStuck())
   • Added CheckMaxLossPerDay() (after CheckMaxDrawdown())
   • Updated OnTick() to call them and block new orders when a threshold is breached (lines ~2380-2390)
2. MQL5/Experts/GoldenRobber/GoldenRobber_hybrid.set
   • Added a Risk Management section
   • Lowered InpTiaProfitUSD: 5.0 → 3.0
   • Lowered InpGlobalSLClearStuckDistancePips: 80 → 60
   • Raised InpGlobalSLClearStuckMaxLossPctOfDayProfit: 50.0 → 65.0

Conclusion: risk-protection mechanisms were added to address the main causes of loss (large losing streaks, drawdown >100%, poor R/R). A new backtest is needed to confirm they work.

• Option 1: change the code so GoldenRobber groups stuck orders into a cluster and trims them together once there is profit.
• Option 2: tune the set file as proposed (enable Global SL Clear Stuck + suitable parameters).

• Tia (cluster): closes only when the cluster's total P/L ≥ InpTiaProfitUSD (e.g. $5). Any combination that includes a "far" order (large loss) sums negative → never reaches the target → cannot close.
• Flexible partial close: runs when there are both winning and losing positions. Profit creates a "budget", which is then used to trim losers (far orders first if InpTiaPrioritizeByDistance = true). When only one side is open and all of it is losing (e.g. 7 Sells, 0 Buys) there is no profit → no budget → nothing can be trimmed.

Idea: when several combinations satisfy sum ≥ minTarget, instead of closing the first one found, pick the one containing the most "far" orders (by price distance), so that when there is profit the far stuck orders get trimmed at the same time.

How:

1. Scan masks as today.
2. For each qualifying mask (BotDCA style, sum ≥ minTarget), compute a "priority score" = the number of losing orders in the cluster whose distance to the current price ≥ a threshold (e.g. InpGlobalSLClearStuckDistancePips or a dedicated input).
3. Choose the mask with the highest priority score (prefer clusters with many far orders).
4. Close that cluster (winners first, then trim losers within budget as today).

Pros: the "close only with profit" rule is unchanged; only which cluster gets chosen changes when several reach the target → stuck orders are grouped into the cluster that gets trimmed with profit.

Cons: needs an extra "best mask" variable and a distance loop; with more than 14 orders it is still bounded by TIA_MAX_POSITIONS.

Idea: today budget = P_sum - minTarget - 0.50. An option could be added (e.g. only when distance priority is on) to allow budget = P_sum - X with a small X (e.g. 0 or 1), spending almost all the profit on trimming losers → more far orders trimmed in one pass.

Pros: once there is both profit and loss, trimming together is more aggressive, especially for far orders (already sorted by distance).

Cons: the net after closing is tiny (only X USD left), easily skewed by slippage/commission; needs careful testing.

• Achievable: prefer clusters with many far orders (1A) + optionally widen the trimming budget (1B) → when there is both profit and loss, stuck orders are grouped into a cluster and trimmed together with profit.
• Not a replacement: when one side is all losing you still need Global SL Clear Stuck (or accept holding until the other side is profitable).

• DistancePips: 80 suits XAUUSD; raise it (e.g. 100) if orders should be "farther" before clearing.
• Mode: keep 2 (FarOnly) to close only truly far orders; switch to 4 (FarPartial) to partially close far orders within the % of day profit limit (less of a "shock" than closing all far orders).
• MaxLossPctOfDayProfit: 50 is balanced; lower it (e.g. 30) to clear only on days with more profit.

• No code change needed; only the set file (and the set at the time was already close).
• Clear Stuck also handles the all-losing-one-side case (7 Sells, 0 Buys): once price is far from the worst loser and the % of day profit condition holds, it closes (per the chosen mode).

• Prefer Option 2: keep/refine the set file with Clear Stuck as above (already present in the test set). It is enough to handle far/stuck orders, including the all-losing-one-side case.
• If you want to "group stuck orders into a cluster and trim them together with profit": also build Option 1A (prefer masks with many far orders); consider Option 1B only after thorough backtesting. Option 1 does not replace Option 2 for the all-losing-one-side case.

• Base: forked from v1.41 (Multi-TF Trend Confirm), WITHOUT stair-step (1.42–1.47).
• File: GoldenRobber-1.41.1.mq5 / .ex5 · Compile: 0 errors.
• Date: 2026-07-14.
• Backward compatibility: defaults MinPipsMode=Fixed, DCAPosEnable=DCANegEnable=true → behaves exactly like 1.41. Behavior changes only when you opt in.

• Disabling a branch → the EA skips all add processing for that branch every tick (SR check, IsNearSupportResistance…) → saves CPU.
• Open positions of that branch are NOT touched (TP/close still managed normally) — only NEW adds are blocked.
• Init (opening a new round) is NOT affected — only ADDs (averaging within a round).
• Use case: a config like pass10-hybrid where DCANeg rarely fires → InpDCANegEnable=false for a leaner, lighter EA.

• Uses the Classic pivot: R1 − S1 = previous period's range → base = half of that range, converted to pips, cached once per bar.
• Applies only to ADD spacing (both DCAPos/DCANeg branches). Init-round spacing stays fixed (round-opening behavior unchanged).
• Self-adapts to volatility: gold moving hard → large range → wider spacing → fewer counter-trend orders get stuck; calm market → tighter spacing → faster clean-up.

1. Baseline: original pass10 on EA 1.41.1 (all defaults) → must equal the old pass10.
2. DCANeg off: the config above (DCANegEnable=false) → compare profit/DD with the baseline. If roughly equal → DCANeg really is "useless" here and can be switched off for a lighter EA.
3. Pivot MinPips: set InpDCAMinPipsMode=1, keep TF=H4, K=1.0 → backtest 2023-2026, reconstruct equity DD, compare with Fixed. Check whether the 05/02/2025 dip gets shallower and whether whipsaw gets worse.
4. Tune PivotK (0.7–1.5) and PivotTF (H1/H4/D1) if step 3 looks promising.