Short rolling Sharpe readings in MT5 often look like “alpha” but are mostly estimation noise: the standard error shrinks only with 1/√n, so 40–60 bars can’t reliably separate skill from randomness. The article formalizes this using Lo’s Sharpe standard error, then annualizes both Sharpe and its uncertainty to build a practical confidence band.

RollingSharpe.mq5 plots annualized Sharpe plus upper/lower bounds at ±1.96·SE. The core rule is simple: if the band crosses zero, the observed Sharpe is not statistically significant.

On the engineering side, it addresses MT5’s recalculation pitfalls (prev_calculated=0, partial history) by using stateless per-bar, two-pass variance computation for the indicator, while providing reusable O(1) circular-buffer classes for EAs where sequential updates are safe.

👉 Read | Calendar | @mql5dev

Weekend gap trading can be automated by pairing a custom indicator with an MQL5 Expert Advisor that executes orders from indicator buffers.

The indicator must expose buffers read via CopyBuffer(). Six buffers are used: buy/sell arrows plus TP/SL for each direction. Buffers are registered with SetIndexBuffer(), and the EA connects through iCustom() to avoid re-implementing gap detection.

The EA structure centers on inputs (lot size, slippage, magic, closed-bar confirmation, duplicate handling, opposite-position logic, midpoint SL rules), a CTrade instance, an indicator handle, buffer arrays, and state variables. Utility functions centralize buffer reads, empty checks, duplicate-bar tracking, position lookup, stop validation, and optional reversal.

OnTick() runs once per new bar, copies buffers, validates setups against broker stop distances, places Buy/Sel...

👉 Read | CodeBase | @mql5dev

This part turns MMAR parameter estimates into a full synthetic price path via a single, stateless CSimulationEngine class designed for repeatable Monte Carlo runs. It relies only on MT5 Standard Library components: ALGLIB FFT and linear algebra plus Normal/Gamma/Poisson RNG, while keeping each intermediate array for inspection and validation.

Stage 1 builds multifractal trading time with a binary multiplicative cascade. Random multipliers come from the fitted distribution using M = b^(-V), are clamped for numerical safety, pair-normalized to conserve mass, then integrated into a CDF theta(t) that creates fast/slow market time.

Stage 2 generates fractional Brownian motion with H using either exact Cholesky (small n) or FFT-based Davies–Harte (large n), with fallback on failure, then rescales to match observed volatility.

Stage 3 composes X(t)=B_H[theta...

👉 Read | Freelance | @mql5dev

Manual chart-object monitoring fails for three reasons: reaction latency, poor scalability across symbols/timeframes, and subjective “did it touch?” decisions. The article extends earlier MT5 object enumeration/normalization work into an automated pipeline that turns drawn tools into consistent, testable triggers.

The solution reuses normalized geometry in SComplexObjectInfo, then adds interaction logic that evaluates sloped objects at the current time (trendlines, channels, pitchfork median/levels) instead of comparing against static anchors. It also handles mixed object models: HLINE/VLINE single-axis coordinates, rectangles as zones, and Fibonacci levels read from the object’s level arrays.

Architecture is split into modules: an updated collector (now includes HLINE/VLINE), an InteractionDetector that outputs touch/cross/breakout events with dir...

👉 Read | Signals | @mql5dev

This CoSO implementation treats optimization like managing a research lab: “funds” act as discrete compute/iteration budget, allocated between proven agents and new entrants. AssignFunds splits budget by omegaCurrent, then distributes the main share via rank-weighted lottery so higher-ranked researchers get more opportunities without fully eliminating randomness.

Diversity is injected through CreateOutsiders, which spawns a limited number of new agents, initializes coordinates within bounds (with step snapping), normalizes per-agent probability vectors, and enforces population caps with controlled growth.

HireResearchers adds exploitation: funded supervisors spawn nearby variants that inherit best-known positions and bias parameters, with Gaussian perturbations to keep local search active.

ComputeStdDev measures population dispersion, and UpdateOm...

👉 Read | VPS | @mql5dev

This MT5/MQL5 deep dive focuses on the last three chart object events that matter when objects interact with users: delete, change, and end-edit. One key detail: some of these events are not generated unless explicitly enabled, so handlers can be correct yet never fire.

For CHARTEVENT_OBJECT_DELETE, the article shows how to catch deletions and immediately recreate “protected” objects, including edge cases around when notifications are disabled and how UI lists lag behind recreated objects. A practical pattern is keeping an internal snapshot of object properties for reliable restoration.

For CHARTEVENT_OBJECT_CHANGE, it explains why user edits don’t persist after recreation unless the EA captures updated properties itself. Since MT5 reports only the object name, developers must selectively read and store relevant properties, with strict filtering to avoid...

👉 Read | NeuroBook | @mql5dev

SQLite access in MetaTrader 5 still needs defensive testing, especially around result handling. Query execution and result reading are coupled: ExecRequestOfData must run before any read, because DatabaseReadBind returns rows from the last executed request.

A generic wrapper around DatabaseReadBind is used to mitigate MQL5 type constraints. The binding API effectively behaves like a void reference in C/C++, so templates are used to accept varying row structures without rewriting code when result shapes change.

Testing shows strict ordering rules. Field order in SELECT must match the structure used for reading, including joins. Extra columns can be returned and selectively ignored by switching structures, but mismatches between expected fields and returned columns produce unexpected output and require careful validation.

👉 Read | NeuroBook | @mql5dev

The logic targets sessions with high-impact (red) economic events. It counts qualifying news items for the current date, then evaluates the next scheduled release time.

Before the release window, the EA places a pending order based on the configured direction, either Buy Stop/Buy Limit or Sell Stop/Sell Limit. Order placement is gated by the news count and the remaining time to the event, to avoid triggering outside the defined pre-news period.

Typical safeguards include preventing duplicate pending orders for the same event, applying spread and slippage limits, and removing or disabling orders after the release if the setup is no longer valid.

👉 Read | Quotes | @mql5dev

This article moves MT5 metric export from backtest-only CSV dumps to a live streaming pipeline that stays useful during active market hours. The goal is continuous observability: indicator values, session counters, and equity snapshots become an auditable stream instead of an end-of-session artifact.

The design is a decoupled three-layer system: an MQL5 include that buffers rows and flushes in batches, a daily rotating CSV in the common files folder, and a Python tail daemon that reads appended rows, maintains rolling windows, and logs anomalies.

Key engineering details: open-write-close I/O to avoid long-held handles, configurable buffering to control latency vs. data loss risk, midnight-UTC rotation to cap file size, and per-symbol/timeframe filenames to prevent multi-chart conflicts. A demo indicator shows gating to avoid exporting historical back...

👉 Read | Signals | @mql5dev

This article turns candlestick charts into analyzable data by encoding each bar as a single symbol (A/G/H/E for bullish types, a/g/h/e for bearish, D for doji, and “_” for unclassified). Using 1,500-bar samples of GBPUSD and XAUUSD on H1/M15/M5, an MQL5 script generates an encoded series plus per-symbol counts and percentages in a TXT report.

On GBPUSD, Marubozu-like candles (A/a) dominate at ~20–22% each, while 32–36% of bars fall into the unclassified bucket, signaling where the taxonomy may need refinement. Bullish and bearish totals stay nearly symmetric across timeframes, and M5 shows a noticeable rise in doji frequency.

The practical output is a reproducible market “profile” that can feed next-step modeling: two-symbol pattern frequencies and transition probabilities for systematic strategy research.

👉 Read | AppStore | @mql5dev

A new MQL5 Wizard custom signal class, CSignalUKFCapsNet, targets the “noisy middle” where classic indicators and regime models struggle on fast, erratic markets.

Engine 1 uses an Unscented Kalman Filter-style hidden-state estimator to treat price as a noisy measurement and produce a low-lag baseline trend without moving-average whipsaw.

Engine 2 adds a Capsule Network as a structural validator, checking whether the proposed direction matches current volatility boundaries (ATR) and momentum pace (RSI). Low-confidence setups are suppressed via vector squashing.

The class supports UKF-only or UKF+CapsNet testing, and offers four modes: volatility breakout, mean reversion, trend following, and consolidation squeeze—giving developers switchable logic for different intraday conditions.

👉 Read | NeuroBook | @mql5dev

Part 2 replaced math analogues (cosine/sine/exponential) with a virtual 3‑qubit processor: 8‑state Hilbert space, unitary gates, normalization, and measurement behavior. This shift targets superposition, entanglement, and non‑commuting operations that classical pipelines miss. Reported limits: LSTM ~58% accuracy, transformer noise sensitivity, ARIMA degradation, and overfitting instability.

Circuit construction is data-driven: rotations parameterized by price moves, controlled-phase gates by feature correlations, Hadamard by uncertainty. Evolution order matters; decoherence strength tracks regime changes. Measurements extract Z/X statistics and two‑qubit correlations, using non-destructive reads in simulation.

Hybrid output feeds classical components and transformer attention (Q/K). Implemented as an MT5 EA in MQL5. 2024–2025: +13.3% annual, 2.39%...

👉 Read | CodeBase | @mql5dev