Week ending 5 November 2026

TL;DR

5 November 2026: No new MetaTrader 4 or MetaTrader 5 desktop release is listed for 30 October–5 November. A new MQL5 example uses confirmed swings to distinguish RSI continuation from reversal, without placing trades. Before sizing an EA, test alternative trade sequences: this week’s sequence-risk study shows why one backtest’s drawdown is an incomplete risk estimate.

At a glance

This week’s most useful MetaTrader developments are in the way trading systems are researched, tested and connected. New community articles examine RSI signals in context, statistical evidence for mean reversion, the order of winning and losing trades, reliable Python integration and a physics-inspired optimizer.

There is no new desktop build announcement dated 30 October–5 November in the official release histories checked for this edition. MT5 Build 6230 from 24 September and MT4 Build 1490 from 2 October remain the latest listed desktop releases. Their implications are covered below as earlier updates, followed by this week’s newly published technical reading. MT5 release history · MT4 release history

For developers, the practical theme is to make every decision inspectable: know which data produced a signal, when that information became available, what the execution system did with it and how the result changes under less favorable assumptions.

MT5 platform and MQL5 updates

Build 6230 from 24 September 2026

Build 6230 expanded the AI Assistant’s ability to retrieve Expert Advisor input parameters and prepare Strategy Tester runs. It can also work with EAs, scripts and indicators on charts and retrieve built-in Economic Calendar information.

The same release added sorting for vectors and matrices, .NET CoreCLR 8+ integration, matrix printing and more operations on complex matrices and vectors. These include products and solutions such as covariance, convolution, linear solves and pseudoinverses. Web-terminal fixes addressed unexpected chart scrolling when a new bar appeared and an account-connection error. Build 6230 release notes

What to do with this: treat improved AI access as a reason to formalize the research brief. Before asking an assistant to optimize a robot, write down the symbol, timeframe, date range, execution assumptions, permitted parameter ranges and evaluation metric. Preserve that specification alongside the result.

For numerical work, add small correctness tests around any newly adopted matrix operation. A test with known inputs and outputs is easier to diagnose than an unexpected change in a complete strategy’s equity curve.

The earlier Build 6180 context

Build 6180, released on 3 September, had already added tools for obtaining tester reports, checking tester settings, launching optimization and reading tester, terminal and EA logs. It also extended several operations for complex vectors and matrices, including handling missing numeric values and manipulating matrix shape. Build 6180 release notes

The distinction matters: report retrieval and optimization tools were not introduced this week, nor were they all introduced by Build 6230. Together, the September releases support a more connected workflow, but a generated explanation still needs to match the actual tester configuration and logs.

A useful review habit is to ask three separate questions: Did the requested test run? Did it run with the intended inputs? Does its result answer the research question? A convincing narrative cannot substitute for any of those checks.

MT4 platform and MQL4 updates

Build 1490 from 2 October 2026

The latest listed MT4 desktop release combines security, bug and stability work with changes to Stop Loss and Take Profit handling in FIFO mode.

When an existing position on an instrument has SL/TP levels, opening the New Order window for that instrument populates those levels. Changes in that window can apply to both the existing and new position. Setting levels where multiple positions previously had none applies them across the instrument’s positions, and One Click Trading inherits existing levels. These are FIFO-mode behaviors, not a statement that all MT4 accounts behave identically. Build 1490 release notes

Developer review: on an appropriate demo account, check a small matrix of situations: one position versus several, existing stops versus no stops, and order-dialog entry versus One Click Trading. Record what the terminal and broker actually accept. If an EA assumes each position’s protection can be managed independently, compare that assumption with the account’s rules before relying on it.

The release notes do not announce new MQL4 language features. This week’s technical selections are MQL5 examples; porting an underlying trading idea to MT4 remains a separate implementation task.

This week in algorithmic development

RSI signals with confirmed market structure

Published 5 November 2026 · Clemence Benjamin

The article builds an analytical EA that combines closed-bar RSI threshold crossings with confirmed swing highs and lows. It classifies continuation, reversal, pending and expired situations rather than treating “overbought” as an automatic sell.

Implementation is separated into RSI reading, swing detection, structure classification, candidate management and chart or Journal output. The program places no orders and includes no money-management layer. Its confirmed pivots require subsequent bars, so recognition is delayed by design. The reported test uses generated ticks, and its limited Journal excerpt does not establish full-year signal accuracy or profitability. RSI and market-structure article

Suggested experiment: keep the original analytical version separate from any trading version. Build a review file with the event time, RSI values, swing levels, confirmation time and eventual classification. Manually examine a sample of continuation, reversal and expired cases before adding orders.

This offers a manageable first project because the initial acceptance criterion is observable behavior: does each state transition follow the declared rules? Profitability is a later question, requiring explicit entry, exit, sizing and cost assumptions.

A variance ratio test for regime assumptions

Published 4 November 2026 · Kayode Michael Oyetunde

This MQL5 implementation examines whether the variance of multi-period returns scales as expected under a specified random-walk or martingale-difference null. It separates a reusable CVarianceRatio engine from a script that retrieves data and reports results across horizons.

A ratio below one suggests a balance of negative serial dependence; above one suggests the opposite. The magnitude alone is insufficient: statistical uncertainty matters. The article includes classical and heteroskedasticity-robust statistics, warns about scanning multiple horizons and notes that short-interval microstructure effects can resemble useful mean reversion. Rejection of the null is not proof of a tradable edge. Variance-ratio implementation

Suggested experiment: declare a lookback and a small set of horizons before looking at results. Save all outcomes, including inconclusive ones. If the test becomes a strategy filter, compare the filtered and unfiltered strategies on the same untouched period with identical costs and execution rules.

Require the filter to justify its complexity. A reduced number of trades can make a result look cleaner while leaving too little evidence to evaluate. Report coverage alongside returns: how often did the filter permit trading, and which conditions did it exclude?

Sequence risk beyond one equity curve

Published 3 November 2026 · Patrick Murimi Njoroge

This installment uses block bootstrap resampling and block permutation to distinguish variation in sampled trades from variation in their ordering. It examines drawdown distributions, loss-limit breaches and walk-forward degradation.

The worked example uses an RSI rule on EURUSD H1 data from 2018–2024, with bid-ask spread as its only transaction cost. It also considers server-day equity checks, which are different from analyzing closed trades alone. The example’s findings belong to that rule and dataset; they are not universal drawdown estimates. Sequence-risk article

Suggested experiment: alongside the original backtest, keep a distribution of possible drawdowns rather than one favorable number. State the simulated horizon and exactly how each loss limit is measured. A fixed percentage of starting balance, a trailing high-water mark and a daily equity rule are different constraints.

For your own review, add explicit stress assumptions for costs omitted from the baseline. The useful decision is whether the strategy remains acceptable under a plausible range of paths and frictions, not whether one resampled path looks attractive.

Reliable Python and MQL5 communication

Published 4 November 2026 · Stanley Kimathi Kibaara

The Redis Streams prototype makes message identity, acknowledgments and recovery explicit. MQL5 supplies events, Python proposes trades and the EA retains execution authority. Despite the title’s emphasis on events, the EA still polls on a timer; the improvement is in the delivery protocol.

Stable IDs and pending-message handling help address duplicates and restarts, while unusable messages need a rejection path. A delivery acknowledgment does not by itself establish a broker fill. The article treats crashes, ambiguous broker outcomes and retries as distinct problems. Redis Streams integration article

Suggested experiment: test recovery before adding a complex model. Deliberately interrupt the connection before sending an order, after submission and before recording the outcome. In every case, establish whether the system can reconcile its records without submitting a duplicate trade.

Define expiry rules for signals, too. A message that was valid when generated may be unusable after a restart. Put a timestamp, strategy version and identity on each proposal, and specify how an operator can inspect unresolved work. More infrastructure is worthwhile only when it makes failure behavior clearer.

Archimedes Optimization Algorithm in MQL5

Published 4 November 2026 · Andrey Dik

This article implements the Archimedes Optimization Algorithm in the C_AO framework. AOA represents candidates using properties inspired by density, volume and acceleration, with exploration and refinement modeled through their interactions.

The underlying algorithm dates to work published in 2021; the new item here is this MQL5 article and implementation, not the invention of AOA. The article evaluates it on benchmark functions with a fixed budget of 10,000 fitness evaluations. Such optimization benchmarks are not evidence that a trading strategy will make money. AOA implementation and benchmark discussion

Suggested experiment: compare any new optimizer with a simple baseline using the same objective, bounds and number of evaluations. Run multiple seeds and report variability, not just the best outcome.

Keep the final evaluation period outside the search process. Improving the search algorithm cannot repair a badly chosen objective, contaminated validation data or unrealistic transaction costs. If a more elaborate optimizer only finds a better in-sample peak, the extra machinery has not yet demonstrated practical value.

MQL4 and MQL5 implementation corner

The same indicator name has different interfaces

RSI is a useful example of why a strategy cannot be ported by changing the filename extension. In MQL4, iRSI returns the indicator value for the requested shift. In MQL5, iRSI returns an indicator handle, and values are retrieved from its buffer. A failed MQL5 handle must be detected, and the handle can be released when no longer needed. MQL4 iRSI reference · MQL5 iRSI reference

For a closed-bar rule, explicitly distinguish the latest completed bar from the forming bar. In MQL5, CopyBuffer position zero refers to the current bar. With a fixed two-element receiving array and a request starting at position one for two values, the older requested value occupies the beginning of physical memory. Check the returned count against the number requested; missing or incomplete history is not a usable signal. CopyBuffer reference

A practical porting review:

1. Write the signal rule in plain language, including the exact time at which each input becomes available.

2. Match those inputs across the two platforms using the same symbol, timeframe and comparable history.

3. Log the calculated values before connecting the signal to order code.

4. Compare event timestamps, not merely the appearance of two charts.

5. Review execution and position management separately from indicator calculations.

This sequence makes discrepancies easier to locate. If the signal inputs already differ, changing the order code will not resolve the underlying mismatch.

A practical development plan for the week

For an existing EA, prioritize one controlled improvement rather than combining all five ideas. Choose according to the failure you are trying to understand.

If entries are hard to explain: start with the analytical RSI example and add structured event logs. Your first deliverable should be a traceable decision, not an optimized return figure.

If a mean-reversion assumption is untested: run the variance-ratio diagnostic as a research tool. Decide in advance what result would change your view, then retain the negative and ambiguous findings.

If a backtest looks too comfortable: examine drawdown and cost sensitivity before increasing complexity. Make the loss-rule definition explicit and keep floating-equity questions separate from closed-trade statistics.

If Python integration is unreliable: concentrate on duplicate detection, stale signals and uncertain execution outcomes. A simpler model with auditable execution is a better debugging target than several interacting models.

If optimization is the bottleneck: benchmark the optimizer under a fixed computational budget. Separate faster search from better out-of-sample behavior.

For each experiment, retain a compact record: platform build, code version, data source, account mode, symbol, timeframe, date range, settings, costs, expected behavior and observed result. Finish with a decision: keep, revise or reject the change, with a reason that another developer could reproduce.

Closing perspective

The most valuable contribution of this week’s reading is a set of ways to ask sharper questions. Does an RSI crossing have structural support? Does the statistical assumption hold at the chosen horizon? Could another ordering of trades produce unacceptable losses? Can the execution system survive a restart? Does a different optimizer improve anything beyond the training result?

Those questions connect research to implementation. Start with a narrow hypothesis, preserve the evidence and expand only after the behavior is understood.

The linked community articles are educational implementations, not independently validated trading systems. None of the techniques reviewed here establishes profitability or removes trading risk.