How the reach column is computed
First in, a seat either opens or folds. So the chance the action arrives at a seat with nobody having entered is the product of one minus the opening width of every seat before it. At 100bb the button opens 48.6% of hands when it gets the chance, but it only gets the chance 44.5% of the time, so it opens 21.6% of all hands dealt. Averaging the widths would have told you 30.5%, which is not a frequency of anything.
What this is not
This counts open-raises. It is a floor on preflop raise frequency, not the whole of it - 3-bets, squeezes and raises over limpers all count as raises in a tracker and none of them are here. It also says nothing about VPIP, which needs the calling and defending trees. We have not aggregated those, and rather than estimate them we have left them out.
If you want to compare your own numbers against the solve rather than against a population average, the Hand Tracker grades your actual opens against these same solves and only issues a verdict once your sample is large enough to prove a gap. The full ranges behind every width here are free in the solver, and the method is on methodology.