The order
- Set the symbol and range, and reserve out-of-sample bars under Data range parts in Edge Finder. A Build task uses the same control, so what is reserved in one is reserved in the other.
- Measure and Deploy. The reserved bars are excluded from every table and used only for a direction check, shown as holds, reverses or too few.
- Run the Builder on the configured task, with its out-of-sample range inside the window you reserved.
- Apply StrategyQuant’s retest, walk-forward matrix and robustness checks to the survivors.
- Name and export the ones you keep.
The move that spoils it
Letting a measurement see bars that validation later treats as unseen. If Edge Finder, NN Lab or the Builder has looked at a stretch, a walk-forward result on that stretch says less than it appears to. The reserved range in Edge Finder is there to prevent this, and the Held out column is the only thing it is used for.
Choices between runs are not corrected
Edge Finder’s correction covers the cells in a table. It does not cover choices made between runs: changing the holding period after seeing the hold curve, trying five timeframes and keeping the best one, or deploying a window found on the full range. The manual says so, and says a fresh date range is the confirmation. A walk-forward matrix is one way to get fresh data, so it belongs after the measurement and on the Builder’s survivors, not as a way to tune the measurement.
What the matrix is not asked to do
Edge Finder does not run walk-forward and the matrix does not decide which side or hours to measure. Each stays in its own place. How StrategyQuant builds and reads the matrix is described in StrategyQuant’s own documentation.