I should have realized an exponential function (orange) would fit better than a cubic (blue). Thus:
y = 0.0284598 × 3.62961^x + 1.54071
Unsurprisingly, this has a better R² value of 0.9998.
It makes sense, because every zone should weight higher *per percentage difference* than the previous zone.
This is not true with the cubic between training zones 1 to 2 and 2 to 3. It is true with every zone using the exponential fit.
Anyway, I still haven't heard anything from the researchers.
</nerd>