Inside the PMTS Signal Engine: From Raw MT5 Tick Data to Trading Decision — May 29, 2026

One of the most frequent questions institutional allocators ask before subscribing capital to PMTS is conceptually simple but technically dense: what happens, mechanically, between a tick arriving from the broker and an order being sent back to the market? The answer is a multi-stage signal engine. This article documents the architecture of that engine as it operates on MetaTrader 5 for the XAUUSD market on May 29, 2026, and shows the live performance footprint it produces.

Why the signal engine matters

A trading system is only as good as the discipline embedded in the pipeline that converts noisy market microstructure into a binary decision: trade, or stand aside. Discretionary traders compress this pipeline into intuition. Systematic platforms, by contrast, must express it as code — reproducible, auditable, and resilient under regime change. PMTS has been engineered around that constraint from day one.

The result is visible in the platform-wide metrics published in our backend. In the last 30 days, PMTS executed 4,964 trades across the full account population with a 59.11% win rate and aggregated net profit of USD 3,107,536.84. In the rolling 7-day window between May 22 and May 29, 2026, the engine generated 213 trades, 74.18% of which closed in profit, contributing USD 244,302.59 to aggregate equity. Those numbers are not the goal of this article — they are the consequence of the architecture described below.

Stage 1: tick ingestion and microstructure normalization

The pipeline begins at the broker bridge. Each PMTS Expert Advisor is attached to an MT5 terminal connected to one of the seven brokers we currently support (MetaQuotes, DarwinexZero, FTMO, MultiBank Group, MEX Atlantic, among others, distributed across 19 production accounts). The EA subscribes to the OnTick event for XAUUSD and writes every quote into a circular in-memory buffer of fixed depth.

At this stage, the engine is not yet trying to predict anything. It is performing three deterministic operations:

• Outlier filtering. Quote feeds occasionally contain stale prints, broker pre-open snapshots, or wide-spread spikes during low-liquidity windows. A median-deviation filter rejects any tick whose bid-ask midpoint diverges from the rolling median by more than a configurable threshold.
• Spread-aware mid-price reconstruction. Because XAUUSD spreads can widen materially during the Asia-London handoff and around FOMC headlines, the engine computes a synthetic mid-price weighted by recent depth rather than the naive (bid+ask)/2. This synthetic price is what every downstream module consumes.
• Timestamp alignment. Broker server times are normalized to UTC and re-bucketed into one-second, one-minute and five-minute bars. The engine maintains these bars in parallel, because different sub-strategies operate on different horizons.

Stage 2: feature engineering — turning prices into questions

Stage two is where raw quotes become decisions in waiting. Each new bar triggers the computation of a feature vector covering five families: trend, volatility, momentum, liquidity, and macro context.

Trend features include adaptive moving averages, slope-of-regression on multiple lookbacks, and a Heikin-Ashi-style smoothing for visual confirmation in the back-office dashboard. Volatility features include ATR on 14, 50, and 200 periods, realized variance estimators, and Yang-Zhang range estimators that are robust to overnight gaps. Momentum features cover RSI, Stochastic, and a proprietary divergence detector that scores when price highs are not confirmed by momentum highs.

Liquidity features quantify how much the market is willing to absorb at the current price level — derived from the spread, the rate of quote updates, and the inferred depth via order book proxies. Finally, macro context features inject the current proximity to scheduled events: minutes-to-FOMC, time-since-NFP, and whether the Fed blackout window is active. The engine intentionally treats macro proximity as a feature, not as a hard switch — we want the system to down-weight signal confidence near major prints, not blindly stop trading.

Stage 3: regime classification

The single most important architectural choice in PMTS is the explicit separation between regime detection and signal generation. Most retail Expert Advisors collapse the two: they fire entries the moment a pattern is detected, regardless of whether the market is trending, mean-reverting, or compressing into a range. PMTS does the opposite. Every feature vector is first routed through a regime classifier that assigns the current market state to one of four buckets: trending bullish, trending bearish, mean-reverting, and high-volatility transition.

Each bucket activates a different sub-strategy ensemble. A trending bullish regime, for instance, will activate breakout-continuation modules and disable mean-reversion shorts entirely. A high-volatility transition regime, common in the hours preceding a FOMC statement, biases the engine toward smaller position sizes and tighter risk caps. This is why PMTS rarely suffers the classic algo failure mode of fighting a strong trend with a counter-trend module — the classifier has already silenced the wrong tools.

Stage 4: signal scoring and confluence

Within an active regime, multiple sub-strategies independently produce scores in the range of −1 to +1. The signal engine then aggregates these scores using a confluence rule rather than a simple average. A trade is only considered if (a) at least two independent sub-strategies agree on direction, (b) their combined score crosses a threshold calibrated per regime, and (c) the macro-proximity feature does not veto execution.

This confluence requirement is the most important reason behind the quality of recent execution. On account 25 over the May data window, PMTS produced a profit factor of 2.5793 with a 64.63% win rate across 82 trades. On the XAUUSD symbol specifically, the historical profit factor stands at 2.0942 with an average win of USD 158.75 versus an average loss of USD 81.03 — a roughly 2:1 win/loss ratio that is structurally enforced by the engine, not the consequence of cherry-picking trades.

Stage 5: risk-adjusted execution

A signal is not yet an order. The final stage applies a portfolio-level risk filter that considers: open exposure across all 19 production accounts, current correlation between active positions, distance to the daily and weekly loss circuit breakers, and available free margin per account currency (EUR, USD, GBP). Only after the risk module signs off does the order leave for the broker via MT5.

Execution itself is split into market and pending orders depending on the regime. In trending regimes, the engine prefers market orders to capture continuation; in mean-reverting regimes, it favors pending orders at calculated reversion levels. Slippage is logged on every fill and fed back into the next iteration of the spread-aware mid-price model — a closed loop that gradually tightens execution quality without manual recalibration.

Auditability: every decision leaves a trace

Every stage of the pipeline writes structured log entries to a server-side database. For each trade, you can reconstruct the exact feature vector at the moment of entry, the regime classification, the sub-strategy scores, the confluence outcome, and the final risk decision. This is what makes PMTS auditable in a way that black-box AI systems usually are not. Investors can verify, on the platform dashboard, the exact metrics behind every account: Sharpe-style ratios, Sortino deviation, Calmar ratio when sufficient history is available, win rate, profit factor, and drawdown geometry.

Across the institutional account snapshot, PMTS maintains a maximum drawdown of just 0.7277%, a metric that is more representative of system integrity than the headline profit numbers. A signal engine that produces strong returns at the cost of large drawdowns is, structurally, a less valuable system than one that produces moderate returns at low drawdowns — institutional allocators understand this; the architecture above is built around that understanding.

What this means for the investor

The institutional client does not need to know the implementation detail of every module. What matters is the operating contract: PMTS will (1) only trade when an active regime is identified, (2) only execute when independent sub-strategies confirm each other, (3) automatically scale down around macro events, and (4) reject any trade that would breach portfolio-level risk caps. That contract is what produces the 30-day footprint of nearly 5,000 trades at a 59.11% win rate without runaway drawdowns.

If you want to evaluate PMTS against your own risk parameters, you can open an evaluation account and monitor the same metrics shown above on your personal dashboard in real time. The engine described in this article is the same engine producing those numbers — there is no separate production version.

Past performance does not guarantee future results. Trading involves substantial risk of loss and is not suitable for every investor. The KPIs cited in this article reflect aggregated platform performance as of May 29, 2026 and may not be representative of future periods. Always consult a qualified financial advisor before allocating capital.