The measurement
Edge Finder takes the symbol on the Data tab and computes one thing over and over: what happened from the open of the bar after a condition was true to the close of the bar a set number of bars later. The default is five bars. It groups that move by side, regime, prior bar, calendar or block signal, and compares each group with the bars outside it. There is no backtest and no genetic search, so the same data and settings give the same table every time.
Four questions
Which side. Long and short are measured as separate populations, so a strong side cannot hide a weak one. The result can be long, short or both, and it says whether the two sides look symmetric.
Which hours. Time Windows looks for runs of hours on twelve weekday sets and charges for the whole search. See time windows.
Which style. Does the market continue after an up bar or revert? Seven block sets, about 160 conditions in all, are each tested once and ranked. When it cannot separate two styles it says they are tied.
Which blocks. Every condition block and numeric indicator in SQX is scored as an entry filter. Indicators are split at five fixed quantiles of their own distribution, so the rule is the same on every instrument and is set before any return is looked at.
Reading the tables
Look at the q column and the Held out column before the means. The q is a p-value adjusted for the number of cells in the table, so a row at q below 0.10 is one where roughly one in ten accepted rows is expected to be noise. Held out says whether the cell points the same way on the bars you reserved. A cell that reverses there is the one to distrust. Grades run strong, clear, weak, not detected and too few bars, and only strong and clear count as survivors.
What goes into the Build task
Deploy writes the measured side, the symmetry, the chosen style’s block set, any Block Edge survivors you add and, if you choose, a time range. The style’s own entries and exits come with it, as when you load its block file by hand, so read the line that says what the build will enter and exit with.
What it cannot tell you
The move is entered at the next open and held for a fixed number of bars. There is no stop, no target and no sizing, so the figure is a bias detector and not a forecast of what a strategy will earn. Correction covers a table, and does not cover the choices you make between runs, such as changing the holding period after reading the hold curve. A fresh date range is the check on those. More on that in false discovery rate in a StrategyQuant X workflow.