ATR-smoothed Heiken Ashi candles are drawn directly on the chart: blue for bullish and…

This update finalizes the MT5 replay/simulation system for training by aligning simulated pricing with live-server behavior. A small fix removes hardcoded SYMBOL_DIGITS side effects, and a new config option lets users set per-symbol decimal precision so the position indicator formats prices correctly (e.g., instruments with 0.5 ticks). Database stability is improved by moving table-creation SQL into an external script embedded as a resource and adding constraints to prevent invalid or duplicate records, reducing corruption risks without extra application logic. Take Profit and Stop Loss become functional by adding a close-price check inside the position indicator and firing a custom event to the EA to close positions. The same event-driven pattern can be adapted to simulate pending orders by changing the trigger conditions and emitting the appropri... 👉 Read | Freelance | @mql5dev

PDF generation in MQL5 can work without DLLs when the format is treated as plain text plus a bottom index. A minimal PDF has five regions: header, numbered objects, xref table, trailer, and %%EOF. Only the object list grows; most of the file is boilerplate. The body is a flat set of indirect objects referenced by “N 0 R”. Pages do not embed content or fonts directly; they reference a content stream and resource objects. PDF values are limited to eight types, with names (/Helvetica) distinct from strings ((Hello)). Streams require an exact /Length byte count. A single page typically needs Catalog, Page Tree, Page, Contents stream, and Font. Correct xref byte offsets and startxref are critical; a one-byte shift breaks the file. Content streams use postfix operators (BT, Tf, Td, Tj, ET). Multi-line layouts rely on relative Td moves and can switch fonts mid-s... 👉 Read | Docs | @mql5dev

The article shows how MQL5 operator overloading can be used to express data-structure operations with stream-style syntax similar to C++ input/output, improving readability when done with clear intent. A pointer-based queue is rebuilt using overloaded operators, where a small header-only change switches behavior between LIFO (stack) and FIFO without rewriting the calling code. This demonstrates separating policy (order) from usage. The same idea is extended to a linked list: adding a subscript operator enables array-like access, then the implementation is revised so assignments append new nodes instead of overwriting existing values. The final design uses a doubly linked list and avoids traversal by maintaining links during insertion, making updates predictable for trading utilities like event queues and order pipelines. 👉 Read | Quotes | @mql5dev

Average True Range (ATR) is derived from True Range (TR), defined as max(High-Low, abs(High-Close[1]), abs(Low-Close[1])). Initialization typically computes TR over Length bars, then uses the SMA of those values as the first ATR. RMA uses alpha=1/Length and updates as: rma = alpha*TR + (1-alpha)*prev_rma. SMA mode recalculates the simple average of TR over the last Length bars on each candle. EMA uses alpha=2/(1+Length) with: ema = alpha*TR + (1-alpha)*prev_ema. WMA applies linear weights: sum = N*TR[0] + (N-1)*TR[1] + … + 1*TR[N-1], then wma = sum / (N*(N+1)/2). Common validation setup: XAUUSD on H1, comparing RMA, EMA, SMA, and WMA outputs side by side. 👉 Read | Docs | @mql5dev

SAGDFN targets noisy, redundant market data by keeping only neighbors that measurably influence the system. After implementing Significant Neighbors Sampling in OpenCL, the focus shifts to Sparse Spatial Multi-Head Attention to extract structure from the selected links while staying compute-efficient. A key optimization avoids per-pair embedding concatenation. By splitting the first linear layer into separate query/key projections computed once per node, pair logits become simple vector additions, cutting complexity from O(NMhd) to O(Nhd)+O(NMh) and reducing memory pressure—well-suited to GPU execution in MQL5+OpenCL. For attention normalization, iterative α-Entmax is replaced with Sparse-SoftMax to keep sparsity without expensive τ searches. The OpenCL forward kernel computes head-wise sparse weights with local reductions, NaN/Inf guards, bounds-che... 👉 Read | Calendar | @mql5dev