Methodology
How the engine decides
We do not re-rank your MT5 table. We look for stable zones — and tell you how strong the evidence is.
- 01 Gates Your minima in both periods
- 02 Quality Rebuilt without Result
- 03 Neighborhood Do neighbors also hold?
- 04 Plateau Connected good zones
- 05 Pick Center of the plateau, not the peak
- 06 Unseen Optional clean check
1. Minima before rankings
A setup only enters if it clears your floors (PF, drawdown, trades, profit) in both periods. Rankings always invent a “top 5%” — even when everything loses.
2. Quality without Result
MT5’s Result column is whatever you optimized for. We rebuild quality from objective exports (PF, recovery, Sharpe, drawdown, trades). Score = worse of in-sample and forward — not the average.
3. Plateaus, not peaks
Isolated spikes are rejected. A plateau is a zone where neighbors also clear your minima. We pick the center (maximin), not the profit tip.
4. Evidence strength — not a trade signal
The verdict is strong / moderate / weak / insufficient. It never means “go trade”. Too little data is not the same as “no plateau”.
5. Stress-tested on your actual trades
The optimization grid only has aggregated per-pass metrics — never enough for a real Monte Carlo. But the backtest report of your chosen configuration on the unseen period does carry the trade-by-trade list, and that unlocks three checks the grid alone cannot give you: a stationary-bootstrap Monte Carlo over your real trades, a sample-size check (can this many days actually tell your edge from noise?), and a cost-stress table for what happens if your real live spread, slippage or commission are worse than what you optimized with. Drop the report — a real MT5 backtest, not an optimization export — anywhere on the "Unseen period" step.
6. What we do not claim
- A plateau is stability in your sample — not proof of future edge.
- We do not invent classic PBO / Reality Check / SPA: MT5 optimization files lack the equity curves those tests need.
- Forward already helps select, so it is slightly “contaminated”; the cleanest check is a true unseen period.
- We do not tell you to buy, sell, or fund an EA. Evidence ≠ permission to trade.
- With many parameters or a genetic (non-grid) search, neighborhood and plateau size are capped conservatively — it may understate support, never invent it.