The datum radius in a direction, before terrain.
For a spheroid this is radius, exactly as it has always been. For a body
with a figure it is that body's measured ellipsoid, and the difference is
not cosmetic: radius is a, the largest half-extent, so a sphere of it
sits above the real surface everywhere except at the two tips of the long
axis. Haumea is 1050 × 840 × 537 km, which puts its poles 513 km inside a
sphere of its own equatorial radius — a ship would latch landed half a
Haumea-radius above anything, with the altitude readout at zero. Phobos was
the smaller version of the same bug and a genuine regression: its datum was
the 11.27 km mean radius until it gained a figure, and became 13.3.
The lumps below the ellipsoid are not accounted for here and cannot be:
they live in data/shapes/, and packages/universe may not read a file.
That leaves an error bounded by the body's own roughness — a median of 9% of
the mean radius across the measured set — against the 50% a sphere of a
leaves on a body like Haumea. It is the difference between a datum that is
approximately the surface and one that is definitely not.
A spheroid's own flattening is also not accounted for, and that is deliberate rather than an oversight: it has always been that way, Earth's 21 km of polar flattening is the largest case that can be landed on, and changing it moves the ground under every existing save.