An ONNX reader can produce accurate node and tensor reports, but a log list hides topology. Parallel branches and joins read like a linear chain because connections are only implicit via value names.

A visualization pipeline in MQL5 addresses this by extending the parser to keep tensor shapes, node attributes, and limited weight samples with statistics. Internal value shapes are recovered by parsing the graph’s value-info entries, not just exposed inputs/outputs.

Rendering uses a chart bitmap with anti-aliased primitives, clipping, and cached text measurement. Layout assigns node depth as one plus the maximum depth of its producers, found by matching outputs to inputs, excluding weights. Boxes are placed by rank, edges are labeled, and UI supports pan, zoom, and node inspection.

👉 Read | AppStore | @mql5dev

Competitive Swarm Optimizer (CSO) is explored as a fix for classic PSO stagnation in parameter tuning: instead of pulling every particle toward a global best, agents compete in random pairs so only the loser updates, preserving diversity and delaying premature convergence.

The loser’s velocity blends inertia, attraction toward the paired winner, and a controlled pull toward the swarm centroid. The centroid is the only global reference; strong solutions propagate indirectly through repeated encounters rather than instant broadcast.

An MQL5 implementation is provided as a plug-in class for a unified test bench (Init/Moving/Revision), with practical details like zero-initialized velocities, per-axis velocity caps, boundary clamps, and Fisher–Yates pairing. Experiments on Hilly/Forest/Megacity across 5/25/500 dimensions evaluate solution quality under a fix...

👉 Read | NeuroBook | @mql5dev

Extralonger treats market time series and cross-asset structure as a single representation, avoiding the usual split between “temporal” and “spatial” modeling. This drops complexity from exponential growth to quadratic, enabling faster training, lower memory use, and forecasts that extend from hours to multi-day or weekly horizons.

The architecture is a three-branch Transformer: a temporal route for long-range patterns via self-attention, a spatial route using global-plus-local attention to model cross-market dependencies, and a mixed route that fuses both into a unified signal.

On the MT5 side, the work ports these ideas to MQL5+OpenCL: a configurable temporal embedding layer (CNeuronTempEmbedding) that builds multi-period embeddings and concatenates them with inputs, plus a Global-Local Spatial Attention kernel that runs dense global heads and s...

👉 Read | NeuroBook | @mql5dev

Candlestick encoding updates replaced the catch-all “_” with directional fallbacks: N for bullish unclassified and n for bearish unclassified. The original A/a, G/g, H/h, E/e, and D mappings remain unchanged, keeping the text-based pipeline stable.

The frequency pipeline was re-run on 1,500-candle samples for GBPUSD and XAUUSD on M15 and H1. The prior dominance of “_” is now treated as coverage data, while N/n preserve bias for candles that miss body-to-wick thresholds.

GBPUSD M15 singles: A 22.20%, a 21.07%, n 18.93%, N 16.80%. GBPUSD H1 singles are similar: a 21.80%, A 20.67%, n 17.80%, N 17.60%.

GBPUSD doubles: 103–104 unique pairs. Top pairs are Marubozu-heavy, but N/n appear in 12 of the top 16 patterns on both M15 and H1, indicating frequent transitions previously hidden by “_”.

👉 Read | AppStore | @mql5dev

Win rate can’t answer “should this still-open trade be held?” because the relevant probability is conditional on having already survived N bars. Survivors are a different population than new entries, so the chance of a profitable exit over the next horizon can rise materially with age.

The article ports survival analysis to MT5 trade duration: Kaplan–Meier for overall survival, hazard for per-bar resolution rate, and Aalen–Johansen cumulative incidence to handle competing terminal outcomes (profit vs loss). Treating losses as “censored” is the common mistake; it biases profit probabilities upward because stopped trades cannot later win.

Implementation details matter: MT5 history is deals, so positions must be reconstructed by DEAL_POSITION_ID, partial closes merged, and durations measured in bars (not wall time) to avoid weekend distortion. The indic...

👉 Read | Freelance | @mql5dev

Strategy Tester has no economic calendar history, so many MQL5 news filters built on CalendarValueHistory become inactive during backtests. This can push EAs into NFP, CPI, and rate decisions and cause large risk drift from live results. Another issue is timestamp conversion: historical calendar times are shifted using the current server offset, producing a recurring one-hour misalignment across winter/summer on DST brokers.

A three-file workflow addresses both. NewsCalendarExport.mq5 exports events to Terminal\Common\Files\NewsCalendar.csv and recalculates each event using the server offset valid on that date, then runs a winter/summer consistency check. NewsFilter.mqh provides one CNewsFilter for live and tester: live reads the terminal calendar with a 1-minute refresh cap, tester reads the CSV, and only the symbol’s currencies are applied. NewsFilter_Exam...

👉 Read | Quotes | @mql5dev

Daily drawdown breaches remain a primary failure point in prop firm challenges, often outweighing strategy quality. A dedicated monitoring layer can reduce operational risk during active trading.

AlphaQuant Prop Firm Daily Drawdown Dashboard is a free Expert Advisor focused on funded-account constraints. It overlays a clean on-screen risk panel rather than adding chart objects, and tracks account equity in real time to quantify remaining loss capacity versus the configured daily limit.

Core functions include equity-based monitoring (balance and equity), large color-coded readouts suited to multi-screen workstations, and status changes from SAFE to WARNING to CRITICAL as the threshold approaches. The baseline can auto-reset at server midnight to match common prop firm rule timing.

Setup is minimal: attach to any chart, set Max Daily DD % (for example 5.0), ...

👉 Read | Quotes | @mql5dev

Hull Moving Average indicator renders up to four configurable periods on a single chart, with direction-based coloring: green on rising values, red on falling values, and grey on flat sections. Line count can be set to 2, 3, or 4 outputs to match the required signal density.

The implementation uses a ring buffer plus incremental running-sum calculations, keeping CPU and memory usage constant regardless of available history. This avoids per-bar recomputation and reduces sensitivity to long backtest ranges.

Buffer layout is documented and stable, supporting reliable access via iCustom() from Expert Advisors and other tooling. Output buffers are aligned for predictable indexing across configurations.

👉 Read | Freelance | @mql5dev

SMC Gold is an intraday Smart Money Concepts EA designed specifically for XAUUSD on M15. It detects a sweep of a confirmed swing high/low, requires a reversal close through the sweep candle body, and only trades in the direction of the H1 trend using a 50 EMA filter.

A key implementation issue surfaced during testing: tick value reporting for XAUUSD can be wrong on some demo feeds. Using SYMBOL_TRADE_TICK_VALUE produced lot sizes around 10x larger than intended, pushing drawdown to 78% on 107 trades. Switching sizing to OrderCalcProfit() aligned actual risk with the configured 1% per trade. Any EA that sizes from tick value should validate XAUUSD contract properties.

Post-fix testing favored M15. M1/M5 underperformed due to spread impact, and H1 produced too few trades. The execution model is single-position, no grid or martingale, with SL/TP attached immedia...

👉 Read | AlgoBook | @mql5dev

Asian Sweep Scorecard adds a self-auditing layer to the common Asian range sweep setup. It marks the Asian session high/low, flags London/NY sweeps that reverse back into the range, then grades each historical signal in R with spread included. Signals are non-repainting and exposed via buffers for EA use.

Range is defined between configurable server-time hours, with days filtered out if the range is too narrow or too wide in ATR terms. A sweep requires a closed bar to wick beyond one side by a minimum ATR amount and close back inside. Confirmation must arrive within a fixed bar count by closing beyond the opposite end of the sweep bar; entry is assumed next bar open, SL is ATR-buffered beyond the wick, TP is set by an R multiple, and trades are force-closed at a daily cutoff.

The panel reports signal counts, win rate, profit factor, average R, and to...

👉 Read | VPS | @mql5dev

A single LLM prompt wired into MT5 tends to overcommit: when asked for buy/sell, it optimizes for producing a signal, not for admitting uncertainty. This mirrors confirmation bias in trading and becomes an architectural flaw, not a data issue.

The article proposes a debate pipeline: three parallel model roles (bull, bear, risk manager) analyze the same market brief, then a low-temperature judge applies hard rules. If risk is HIGH, the result is always hold; trades require aligned direction from at least two voices.

Indicators are computed locally in pure NumPy (MAs/EMAs, multi-period RSI, stochastic, ATR, StdDev, Bollinger position, momentum, candle anatomy) and fed as a structured briefing. A new DEBATE command keeps V17 compatibility, returns JSON plus full rationale for logging, enabling later prompt tuning from stored outcomes.

👉 Read | Quotes | @mql5dev

This article builds a full Hawkes-process toolkit in MQL5 to measure volatility clustering as event timing, not return size. Large moves are converted into an event train, fitted by maximum likelihood, and summarized by the branching ratio n: the expected number of follow-on events per event, interpreting reflexivity on a 0–1 scale.

Key implementation choices make it practical in MetaTrader: event times are measured in bars for numerical stability, and a trailing volatility window of ~100 bars avoids “normalizing away” clusters during event detection.

An exponential kernel enables O(N) intensity updates via recursion and a closed-form compensator, keeping likelihood evaluations fast. The fitter optimizes in (mu, n, beta) to enforce stationarity (n<1) as a simple bound, and verification relies on brute-force checks, numerical integration, Python cross...

👉 Read | VPS | @mql5dev

Most platforms report a single lifetime Profit Factor. That hides time structure and can make different risk profiles look identical.

Rolling Profit Factor restores chronology by recomputing PF over a fixed window of N consecutive closed trades, sliding forward one trade at a time. Trade-count windows keep sample size constant, unlike calendar windows.

Implementation notes for MQL5: compute Gross Profit and absolute Gross Loss per window. Handle GrossLoss≈0 with a high finite sentinel, and mark all-zero windows as undefined. If history < window size, return failure.

An incremental update subtracts the outgoing trade and adds the incoming trade, reducing complexity from O(N×W) to O(N). Architecture splits trade sourcing, history reading, rolling computation, scaling, reference lines, canvas rendering, and test sources via an ITradeSource interface.

👉 Read | Calendar | @mql5dev

Extralonger targets a common pain point in market forecasting: most models handle minutes or hours, but fail on multi-day horizons. It treats market data as spatiotemporal, aiming to stay stable over long histories while keeping compute and memory costs low.

The core idea is a unified space-time representation that merges price “space” and time into one encoding, avoiding duplicated passes across both dimensions. Architecture-wise, it runs three parallel routes (temporal, spatial, mixed) and uses a Global-Local Spatial Transformer to capture both distant cross-asset links and short-range session-level interactions, with a full receptive field over history.

The MQL5 implementation centers on CNeuronExtralonger, adding adaptive attention pooling so route weights are computed per input, not fixed. Modules include shared temporal projection, separate ...

👉 Read | Docs | @mql5dev

A consensus of LLM agents tends to average away rare but valuable signals and stays static even when specific agents underperform. This article replaces voting with a market-style selection mechanism: each agent trades independently and is judged by PnL, not agreement.

The framework runs 10 distinct trading “philosophies” in parallel, tracks per-agent capital and streaks, and rebuilds each system prompt every cycle via build_prompt() to reflect risk posture. Results persist across sessions, and closed trades trigger a REWARD protocol that updates agent state without extra model calls.

On MT5, true separation is achieved with per-agent magic numbers, allowing simultaneous long/short positions on the same symbol. Three modes cover research and production: observer, full parallel trading, and selection that disables bankrupt agents.

Backtest (EURUSD M15) sho...

👉 Read | AlgoBook | @mql5dev

Most spread indicators average over ticks, which underweights quiet minutes and smooths out the spreads paid during rollover, pre-open, and low-liquidity sessions. A timer-based sampler avoids that bias by taking one spread snapshot per second so each minute has equal weight.

PropSpread is built as a monitoring utility for prop firm and funded accounts. It shows live spread in broker points, as raw price difference, and in the symbol’s natural unit derived from digits and calculation mode (Forex pips, index points for CFD indices, otherwise price units). A legend is always printed to verify conversions.

The panel reports rolling stats over a configurable window (min/avg/median/max and time above threshold), plus session stats since server midnight. It adds spread as a percent of ATR from the last closed bar, and includes threshold alerts with hold time...

👉 Read | Freelance | @mql5dev

EMA Cross Demo EA for MetaTrader 5 was released as a compact reference Expert Advisor focused on development method rather than trading advantage. Signals are evaluated once per new bar using closed candles only, preventing repainting. A fast/slow EMA cross triggers entry; an opposite signal closes the current position and opens the new one on the next tick of the new bar.

Risk is based on ATR: stop-loss uses ATR multiplied by a configurable factor, take-profit is derived from the stop distance via a reward:risk input. The EA enforces the broker’s minimum stop and freeze distances, normalizes volume to symbol limits and step size, and skips entries when spread exceeds a configured maximum.

Position handling is safe for both hedging and netting accounts via magic number and symbol lookup. Recent fixes include correct stop-distance flooring against Bid/Ask on...

👉 Read | VPS | @mql5dev

ATR Trailing Stop ported from TradingView Pine Script to MQL5, kept consistent with a Python reference implementation and validated bar-by-bar.

Rules across all versions:
nLoss = ATR multiplier x SMA(true range, InpATRLength). The stop ratchets under price while closes remain above it, and above price while closes remain below it. A close through the stop flips it to the other side and draws an arrow.

Translation details:
True range on the first bar uses high-low due to missing previous close. SMA remains undefined until enough bars exist (first value at bar n-1). Previous stop uses “previous stop else current price” while undefined. Persistent state maps to an indicator buffer in MQL5 and an array in Python. Crossovers are strictly false when either side is undefined, affecting initial bars.

Repository: github.com/RAo797bit/pine-python-mql5-conversion

👉 Read | Forum | @mql5dev

Trade Reporter Lite sends MetaTrader 5 trade events to a Telegram chat in real time. It only reports activity and does not open, close, or modify orders. Monitoring covers all positions on the account, including manual trades and any EA. Restarts do not re-announce positions already open.

Open messages include side, symbol, entry, stop, target, pip distances, and lot size. Close messages include pips, R-multiple based on the first stop, P/L in account currency, entry-to-exit, and time held. Failed deliveries retry with back-off at 5 s, 15 s, 45 s, and 135 s. An on-chart status box shows sent, queued, and failed counts plus a readable error.

Setup requires a Telegram bot token from @BotFather and a chat ID (channel @name or numeric ID). WebRequest must allow https://api.telegram.org. In Strategy Tester, messages are written to the journal as TRL[tester].

👉 Read | Quotes | @mql5dev

CME Gap Tracker for gold and Bitcoin measures weekend gaps using CME Friday close and Sunday open, mapped into broker server time with Chicago DST handling. It plots each gap as a box, keeps unfilled gaps extended until price trades back to the Friday close, then marks the fill and records hours-to-fill. Gaps below a configurable minimum percent are ignored.

A panel summarizes recent history: counts by direction, fill rates within 24 hours and 7 days, median and slow-tail fill times, and “heat before fill” in gap-multiples. A built-in fade test enters at the CME open toward the fill, uses a stop sized by gap multiple, times out after N days, charges entry-bar spread, and treats stop/target conflicts conservatively.

Gold backtests on broker spot data showed high fill rates (roughly 83%–97%), but negative results for a fixed fade strategy due to freque...

👉 Read | AppStore | @mql5dev

Gold breakout EA for XAUUSD on H4 uses a Turtle-style entry: buy when the last closed H4 bar finishes above the prior 20-bar high, only if price is above the 200 EMA. Default is long-only, one position at a time, with a 2x ATR(20) stop and a fixed 2R take-profit. Orders are opened at market, then SL/TP are attached to the fill; failure to attach triggers an immediate close. Risk is position-sized from stop distance, spread is filtered (skip when spread >10% of stop), and entries are blocked near session close.

Backtest (Jan 2020–Sep 2026, 10k deposit, 1% risk, spread+swap, random delay) reported +9,029.72 net, PF 1.77, max DD 11.38%, 198 trades, 46.97% win rate. Returns were concentrated in 2024–2025; 2020–2023 ranged from -2.9% to +5.1% per year. Enabling shorts reduced performance in this sample, with sells negative while buys remained positive.

👉 Read | NeuroBook | @mql5dev