LEARN · EN · easyquanttrading.com
System Forge: design, then prove
The Forge is where trading rules are born: pick a market sample, run multi-algorithm discovery, and get candidate strategies ranked by evidence — not by curve beauty. In-sample profit is only a filter; the validation pipeline decides what is deployable.
- In-sample profit hard filter on deliverables
- Export capability matrix: pure MQL vs bridge
- Optional auto-queue into robustness pipeline
- Multi-algorithm search: GA, Tree-GP, RL and more
What a strategy forge actually does
A forge automates the search part of strategy design. Instead of manually testing indicators and parameters, you set the market sample and the algorithm family, and the system evolves thousands of rule candidates — recombining, mutating and evaluating them against history.
The word that matters is 'evaluate'. If evaluation is honest (cost-aware, look-ahead-free, out-of-sample), the survivors deserve attention. If evaluation is sloppy, the forge is an overfitting machine with a nice interface.
The forging workflow, step by step
- Pick a market sample — XAU/USD, BTC/USDT, or your own dataset — and choose the timeframe.
- Choose the algorithm family: genetic algorithms, tree-based GP, reinforcement learning, or others; each explores rule space differently.
- Run the search. The forge filters out candidates that fail the in-sample profit screen, so the result list only shows research-worthy candidates.
- Review the candidates by their evidence: rules are human-readable, metrics come from a look-ahead-free engine with costs included.
- Queue the survivors into the robustness pipeline — walk-forward, Monte Carlo, DSR/PBO — and only pipeline-passed strategies become deployable.
Why 'in-sample profit' is a filter, not a verdict
The in-sample profit screen exists to keep the result list useful — losing candidates are noise. But in-sample profit is exactly what an overfit strategy shows best, so it can never be the verdict. The verdict comes from the fixed pipeline: out-of-sample walk-forward segments, Monte Carlo shuffles, and the DSR/PBO statistics that price in how many trials the forge ran.
This is why EasyQuant separates 'research candidate' from 'deployable'. A strategy can pass the forge filter and still be correctly flagged by the pipeline — that is the system working, not failing.
Choosing between algorithms
Different search families explore different parts of rule space. Genetic algorithms are fast and good at tuning parameterized templates; tree-based genetic programming can evolve more expressive rule structures; reinforcement learning searches sequential decision policies. None is intrinsically 'better' — the honest way to compare them is to run the same validation pipeline on each family's output and compare survivors.
On EasyQuant, switching algorithms never switches the validation standards: the same fixed pipeline applies to every family, so comparing results across engines is meaningful.
FAQ
- In-sample profit means live profit?
- No. It is only a research filter. OOS/WF/MC gates still apply before deployable.
- Do I need to write rules myself?
- No. The forge searches rule space for you. You choose the market, the algorithm family, and the constraints; the system returns candidates with readable rules.
- How many strategies does a forge run generate?
- Typically thousands of candidates per run. That scale is exactly why multi-trial statistics (PBO/DSR) are mandatory before anything is marked deployable.
- Can forged strategies be exported to MT5?
- Yes — strategies that pass the pipeline can be exported as pure MQL5 packages or an optional signal bridge for your own account.
- What is the difference between Forge and Pipeline?
- Forge discovers candidates (research). Pipeline validates them (evidence). A strategy is not deployable until it passes the pipeline in fixed order.
More guides
Not investment advice. Historical results do not guarantee future performance. EasyQuant is a research factory — you execute on accounts you control.