> The changing scales and meaning while zooming in and out makes it hard to derive conclusion from these measurements.
It is way simpler than that: at best it is misleading, at worst it shouldn't implement such algo nor display this thing.
The "macro zoom out" until "max zoom in - 1 step" visuals suggest a continuous, factual measurement when it is actually just a speculative mathematical model.
Only the maximum zoom in (displaying...raw data points) is correct. Until you stumble upon tracks sharing the same spots with one full red and the other full green. Miscalibration, shiny green people from earth: whatever it may be.