Three settings decide numbers that look like results. A conversion window decides how many of the sales that happened get counted, not how many happened. A second, shorter window applied to one class of touch removes everything that lands after it. And a membership duration multiplies against the daily inflow, so halving it does not trim a list, it halves it. All three are arithmetic, and all three can be worked out before the setting is changed.
With the lag field empty the calculation uses an exponential model in which exactly half the conversions have arrived by the median day. Paste measured data and the model is not used at all - day 1 means within the first day after the touch, and a window of n days captures every line up to and including day n.