LiquidBrain Liquidity-Draw ML Portfolio
A machine-learning direction model per instrument — predicting which liquidity pool price sweeps first — each with per-pair walk-forward-optimized entries and fixed R:R exits. 9 instruments, all net-positive.
Honest methodology. every graph is a real MetaTrader 5 every-tick (1-minute OHLC) backtest at 0.5% risk per trade, with positions held through the weekend. The model: LiquidBrain here is trained on data before 2021 only — and this is the exact model delivered to clients, not a different one used for the brochure. Every prediction across 2021 – July 2026 is therefore on data the model has never been shown. The settings: stop, target and entry gating were chosen by MT5's genetic walk-forward optimizer across 2021–2026, so unlike the model they were fitted on the period these graphs display. Read the full-window Profit Factor as the primary number. The "PF 2025–26" row is that pair's profit factor over the most recent stretch, which the model had never seen; it is shown because it is a harder test than the full window, though the settings were still fitted with sight of it. Each run starts two years before the window shown and those warm-up trades are discarded, so no figure here benefits from the strategy trading on incomplete data at start-up. On conv-sized instruments, risk per trade scales with model conviction.
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOS
MT5 1-min OHLC · model-OOSPortfolio avg PF ≈ 1.40 · avg Sharpe ≈ 1.27 · all 9 instruments net-positive on both the full window and the 2025–26 model-unseen stretch. Sharpe is annualised from monthly returns. Correlation note: Gold&Silver form one metals leg; the USD pairs share a dollar factor — size at portfolio level accordingly. Instruments that failed walk-forward validation are not sold, so this is the surviving set, not every instrument tested. Delivered over the AgroAlgo signed ONNX-OTA channel; models refresh automatically, no reinstall. These results are measured with positions held through the weekend. Enabling the FTMO weekend/holiday flatten (required on some funded-account types) closes positions before the weekend and will change the results — typically fewer, shorter trades and lower returns. Past performance does not guarantee future results. © Synapse Dynamics · synapsedynamix.com · generated 2026-07-26 (9.2 release).