A split is a simple idea: how long did each kilometre take. It stops being simple when your data is a list of timestamped positions and the kilometre ends between two of them. The splits on Cadence's run summary come from a small function that is worth getting right.
The approach: walk the accepted fixes, accumulate distance, and when the total crosses a kilometre, interpolate the time at the exact crossing between the two surrounding samples. That time minus the previous crossing is the split. Pauses are subtracted, so a stop at a crossing does not turn into a slow kilometre.
The last kilometre is almost never whole. It is shown with its real distance and pace, in a lighter tone, and kept out of the fastest-split calculation, because a 300 m sprint to the door should not win. The winner gets the volt-lime highlight and the rest of the list stays quiet. A table that shouts everywhere says nothing.