The readings the calculated tolerance does not admit, the furthest out first.
The sample, as the grouping named it.
The sample as counts per bin, with the curve to draw over it.
How many times the scatter a reading must lie out before it is taken for a departure rather than for chance. It is fixed by the size of the sample and not at a round three, so that fewer than one of the sample's own readings is expected to pass it.
How far the sample departs from the normal shape, at either end.
Where the readings sit and how far they scatter, at either end.
The window that follows: k times the scatter, centred on the
median. Its offsets run in the direction of the sample, the
copies named less the rest, so a window measured over the
parent's copies carries them negated. toleranceAcross is what
turns these into the window a derivation stores, and it asks
which side was named; storing one of these directly does not.
How far one sample of readings scatters, and the tolerance that follows from it.