mql-trading-bots/notes/Kimi_Code_CLI-Analysis_1-Recent_Changes_and_Vault_Synthesis.md

Kimi Code CLI — Analysis 1: Recent Changes & Vault Synthesis

Date: 2026-05-12 Scope: Uncommitted changes since last git commit (2026-03-30) + conversation history + audit files Confidence: 85% (will be refined in Analysis 2)

---

1. Executive Summary

The repository contains massive uncommitted evolution across both primary EAs and their preset files. The last git commit was 2026-03-30 (n8n/SSH workflow infrastructure). Since then, ~6+ weeks of trading system development exists only in the working tree. This represents significant risk: no version control, no rollback capability, and mixed fix states.

Key Evolution:

• MultiSignal_Confluence_EA.mq5: v1.16 → v1.20 (per-EA realized P&L tracking, weekly drawdown, warmup, pivot cache, state persistence)
• OrdersEA_Smart_Grid.mq5: v3.1 → v4.1 (massive architectural rewrite: adaptive filters, session filter, breakeven, correlation caps, range drift, state persistence, edge cleanup)
• Presets: GetOut=Y emergency shutdown bug FIXED in working tree (was critical per Apr 23 audit). ADXMax relaxed from 18-25 → 40 on most grid presets. Lots standardized to 0.03.

---

2. MultiSignal_Confluence_EA.mq5 (v1.20) — Detailed Changes

2.1 New Inputs Added

Input	Default	Purpose
InpWarmupBars	20	Bars required before filters activate (prevents early-bar noise)
InpPivotThresholdPoints	100	Proximity threshold in POINTS for "near pivot" detection
InpRangingATRFactor	0.3	Ranging "near pivot" = ATR × factor
InpMaxWeeklyDrawdown	10.0	Weekly drawdown cap per-EA

2.2 Architecture Improvements

• Per-EA realized P&L tracking: SumRealizedPnLSince() scans HistorySelect() filtered by DEAL_MAGIC + DEAL_SYMBOL. This is a genuine improvement over whole-account equity snapshots.
• State persistence: GlobalVariableSet/Get for dailyStartEquity, lastEquityReset, weeklyStartEquity, lastWeeklyReset. Survives EA restart.
• Warm-up guard: IsWarmedUp() requires InpWarmupBars bars + one full bar period since init. Prevents indicator-initialization gaps.
• Pivot cache: Daily pivot values cached to avoid recalc every tick.
• Market filling helper: GetMarketFilling() probes SYMBOL_FILLING_MODE per symbol, using IOC→FOK→RETURN fallback.
• Structured logging: LogS() prefixes every message with [_Symbol:MagicNumber].

2.3 Drawdown Logic — PARTIALLY FIXED (Critical Finding)

// DAILY RESET — still uses WHOLE-ACCOUNT EQUITY
dailyStartEquity = AccountInfoDouble(ACCOUNT_EQUITY);

// CHECK — uses PER-EA P&L
double dailyPnL = GetRealizedPnLToday() + floating;

Problem: The baseline is whole-account, but the measured P&L is per-EA. This is the "shared drawdown budget" contamination identified in 2026-04-03 conversation history. If Grid EA drops account equity by $4,000 overnight, Confluence EA’s 5% limit shrinks from $5,000 → $4,750. If both EAs run together, Confluence gets penalized for Grid’s losses.

Severity: HIGH. Not fully fixed despite explicit intent.

2.4 Preset Changes (confluence-*.set)

• InpRiskPercent: 1.0% → 1.5% (aggression increased)
• InpMaxDailyDrawdown: 3.0% → 5.0% (relaxed)
• InpMaxWeeklyDrawdown: 10.0% added
• InpMinATRPercent: normalized per pair (EURUSD 0.3%, GBPJPY 0.5%, XAUUSD 0.7%, etc.)
• InpMinADX: 18.0–22.0 range (was 20–28)
• InpWeekendCloseHour: varies by pair (EURUSD 14:00, others mostly 17:00)
• InpDebugMode: true across all presets

Context: The Apr 7 diagnosis found ATR thresholds at 0.8% blocked ALL trades silently. The presets were corrected in this working tree.

---

3. OrdersEA_Smart_Grid.mq5 (v4.1) — Detailed Changes

3.1 Major New Features

Feature	Description	Risk Level
Adaptive Filters	Relax ADX/lot size after InpRelaxFilterAfterDays of no trade	Medium
Adaptive Entry	Auto-scale grid spacing from ATR (InpEntryATRFactor)	Medium
Spread Filter	InpMaxSpreadPoints — blocks grid placement if spread too wide	Low
Stop Loss Toggle	InpUseStopLoss — opt-in SL (default false = original behavior)	High if enabled
Session Filter	InpUseSessionFilter — restrict trading to Asia hours (default false)	Low
Breakeven	Move remaining SLs to entry + buffer after first TP hits	Medium
Correlation Cap	InpMaxLongSymbols / InpMaxShortSymbols — limit cross-symbol exposure	Low
Range Drift	InpRangeDriftEnable — re-center grid when mid drifts N points	Medium
Profit Protection	InpStopAfterProfit — stop new grids after profitable cycle	Low
Cycle Report	Notification on cycle end	Low
Master One-Shot	masterShutdownDone prevents repeated master shutdown	Low
Edge Cleanup (v5.0)	Close wrong-side positions when price exits grid bounds	High if wrong
State Persistence	GlobalVariables for grid state, drawdown baselines, cycle data	Low

3.2 Drawdown Logic — SAME CONTAMINATION BUG AS CONFLUENCE

// Grid EA also resets from whole-account equity
dailyStartEquity = AccountInfoDouble(ACCOUNT_EQUITY);
// But measures P&L per-EA via realized + floating

Same shared-drawdown-budget problem. The April 23 audit explicitly called this out. It is NOT fixed in the working tree.

3.3 Preset Evolution

• GetOut=Y → GetOut=N on: GBPJPY, GBPUSD, USDJPY, XAUUSD. This was a critical fix.
• Lots: 0.01 → 0.03 on most presets (3× position size increase)
• ADXMax: relaxed to 40 on most (was very strict 18–24; this addresses Apr 13 "filters too strict" complaint)
• RSILower/Upper: widened to 20/80 on most (was 25–35 / 65–75)
• InpMaxDailyDrawdown: standardized to 5.0%
• InpMaxWeeklyDrawdown: 10.0% added

3.4 Unresolved Issues from April 23 Audit

1. Older presets still missing v4.1 safety controls: grid-eurusd, grid-audusd, grid-usdcad, grid-usdchf, grid-nzdusd, grid-eurjpy do not define InpUseSessionFilter, InpMaxSpreadPoints, InpUseBreakeven, correlation caps, adaptive settings, or range drift. They fall back to source defaults.
2. Same-side pending cancellation: When a BUY fills, remaining BUY_LIMITs are deleted. This may be intentional "one-fill-per-side" risk reduction, but contradicts the "multi-level averaging grid" expectation. Comments and behavior remain misaligned.
3. Breakeven implementation: Sets SL to open - buffer (BUY) or open + buffer (SELL). This parks the stop on the loss side of entry, NOT at true breakeven or profit-lock. The Apr 23 review noted this.
4. Cycle P&L contamination: cycleStartEquity captured from whole-account equity. If Confluence EA has floating P&L, Grid EA’s cycle profit/stop logic is contaminated.

---

4. Vault Notes Integration

Conversation History Timeline

Date	Key Finding	Current Status
Mar 21	Stop-loss cross-symbol bug fixed (v1.11)	✅ Still in place
Mar 30	Weekend gap + short signal fixes (v1.14/v3.1)	✅ Evolved further
Apr 1	Grid clustering bug fixed, .set param names fixed	✅ Clustering fix still present
Apr 3	Grid EA caused ~$60K loss; equity at 47%	⚠️ Risk controls added but contamination persists
Apr 5	Grid EA shut down to free shared drawdown budget	✅ Grid later re-enabled with fixes
Apr 7	ATR thresholds 0.8% blocked ALL trades silently	✅ Fixed in presets (now 0.3–0.7%)
Apr 13	Grid filters too strict; user frustration	✅ ADXMax relaxed to 40, RSI widened
Apr 23	GetOut=Y preset bug, equity contamination, breakeven flaws	⚠️ GetOut fixed; contamination & breakeven NOT fixed

Audit Files Status

• 2026-04-23 OrdersEA Smart Grid - Fix Checklist.md: 15 items. Most code-fix items remain unchecked in working tree.
• 2026-04-23 OrdersEA Smart Grid - Preset Mismatch Audit.md: Preset values updated but structural mismatch (missing v4.1 controls) remains.
• 2026-04-23 OrdersEA Smart Grid - Read-Only Code Review.md: 6 findings. Only 1 (GetOut=Y) clearly fixed.

---

5. Cross-Cutting Risks

5.1 Version Control Risk (CRITICAL)

~1,900 lines of diff across 57 files exist ONLY in the working tree. If the server crashes or files are overwritten, weeks of development vanish. This includes:

• The GetOut=Y fix
• The ATR threshold fix
• The ADX relaxation
• The per-EA realized P&L tracking
• The v4.1 architecture rewrite

Recommendation: git add + git commit immediately after review.

5.2 Compounding Risk from 3× Lot Increase

Grid presets moved from 0.01 → 0.03 lots. With MaxLevels=5 and potential multi-chart deployment, total grid exposure could be significant. The daily drawdown was also relaxed (3% → 5%). This is a more aggressive risk posture.

5.3 Silent Failure Path Still Exists

Both EAs have InpDebugMode=true in presets currently. If a future session turns this off, filter blocks become silent again (the Apr 7 root cause). The code structure still gates filter-block prints behind if(InpDebugMode).

5.4 State Persistence Conflicts

Both EAs use GlobalVariableSet with keys like "Confluence_" + MagicNumber + "_" + Symbol + "_dailyStartEquity". If two EAs accidentally share a MagicNumber, state collides. The keys are sufficiently namespaced but rely on MagicNumber uniqueness per symbol.

---

6. Immediate Action Items (Ranked)

Priority	Action	Rationale
🔴 P0	git commit all working tree changes	Prevent data loss
🔴 P0	Fix shared drawdown budget contamination in BOTH EAs	Root cause of Apr 3 crisis
🟡 P1	Normalize all older grid presets with explicit v4.1 controls	Prevent silent fallback to risky defaults
🟡 P1	Clarify grid strategy intent: true multi-level vs one-fill-per-side	Align code, comments, and presets
🟡 P1	Fix breakeven to actually lock at breakeven or profit, not loss-side	Current implementation misleading
🟢 P2	Add per-ticket retry backoff to PositionModify()	Prevent log floods
🟢 P2	Fix adaptive relaxation for never-traded symbols	lastTradePlacedTime == 0 blocks relaxation
🟢 P2	Add InpDebugMode safety check — ensure critical warnings bypass debug flag	Prevent future silent failures

---

7. Confidence Assessment

Area	Confidence	Gap
Code change inventory	95%	Large diff fully reviewed
Vault note synthesis	90%	All conversation history + audit files read
Root cause identification	85%	Shared drawdown confirmed; breakeven logic understood
Fix completeness assessment	80%	Cannot guarantee all edge cases without compilation + backtest
Live MT5 state	60%	No live log access in this session; assuming presets match repo

Overall Confidence: 85% — Analysis 2 will apply first-principles reasoning to push above 90%.