Shapley attribution of strategy filters in native MQL5: one unfiltered backtest that records every filter's verdict per trade contains all 2^n filter combinations exactly, and the report separates each filter's credit, its removal cost, its overlap with others and the share any random veto would have earned.
Economic calendar releases turned from categories into numbers in native MQL5: actual minus forecast standardised by each indicator's own dispersion, and anchored to the minute bar that actually moved by a server clock recovered from price. Includes the per-event response regression that says which releases move a symbol and which, payrolls among them, do not.
Path-dependent volatility in native MQL5: the Guyon-Lekeufack model as four exponential accumulators updated in O(1) per bar, splitting volatility into a trend component and an activity component. Includes the out-of-sample evidence against GARCH(1,1) and the ablation that prices what the sign of the path is worth.
A per-quote market activity gauge in native MQL5: autoregressive conditional durations on the waits between quote changes, with the time-of-day rhythm removed, plus the evidence on whether activity improves short-horizon volatility forecasts beyond live price movement.
Toeplitz Inverse Covariance Clustering in native MQL5: sparse per-regime precision matrices, non-repainting regime ribbon and dependency graph panel.
Symbolic Fourier Approximation in MQL5: a Fourier front end with MCB binning for historical analog search, measured head to head against SAX on identical windows.
Intrinsic time and the directional-change operator in MQL5: scaling laws verified on live ticks, and the Alpha Engine coastline trader built on top of them.
Reconstructing the FX volatility smile in MQL5 from broker quotes, using Garman-Kohlhagen pricing and the delta-space conventions the FX market actually quotes in.
Survival analysis applied to MetaTrader 5 deal history: Kaplan-Meier and hazard curves that answer how long a trade lasts and when it is most likely to die, not just how often it wins.
Entropy pooling in native MQL5: history reweighted by relative entropy instead of cut by a lookback window, with volatility conditioning under a scenario budget, views on volatility, tails and correlation, and an expected-shortfall sizing EA with walk-forward evidence against rolling windows and kernels.
MiniRocket in native MQL5: 84 fixed convolution kernels and a ridge classifier in the dual, with a leak-free audit harness whose controls prove the measurement works before it is trusted
Kronos is a pretrained transformer that forecasts candlesticks the way a language model predicts words.
We implement a research paper's network-momentum trend-following strategy entirely in MQL5. The EA detects lead-lag relationships between markets with Derivative Dynamic Time Warping, learns a weighted network by convex optimization, and trades the momentum that spills across it.