The closest the camera may sit to a body, as a multiple of its radius.
Half a radius of clearance — 3,186 km over Earth, 869 km over Luna — and it is a clearance from three different things at once, which is why it is not the 1.02 it used to be.
The air. The atmosphere shell is drawn at (radius + hazeHeight) over
the sunk sphere, and the tallest one in the Solar System model is Titan's, at
1.155 radii — 400 km of tholin smog on a 2,575 km moon, which is what
makes it a featureless orange ball in every photograph before Cassini's
radar. Triton is next at 1.030, then Mars at 1.018 and Earth at 1.016 — so at
1.02 radii the camera sat well inside the first and level with the rest,
which is the single worst place to put it. The shell covers the whole
viewport, the ray's near end clamps to zero so every pixel marches the full
chord, and none of it is a picture of anything: it is the inside of a hazy
ball. A phone runs out of fragment budget there long before a laptop does,
which is how the bug was reported. observer.test.ts states the clearance
over every haze the model produces, generated as well as vendored, so the
number that has to change if a thicker one is authored is this one.
The ground. The terrain streamer draws whenever a body's relief covers more than a few pixels, which on a planet is a long way out, and refines all the way to its own detail floor. Standing on one is what the surface arm and flight mode are for; the orbit arm wants the body as a body.
The frame. At 1.5 radii a sphere subtends 84° of diameter, so at the default 65° lens the limb is just off the edges and the disk is still legible as a curved world. Nearer than that the picture is ground with no horizon in it — the planetarium is showing you less the closer it gets, which is the opposite of what zooming in is for.
The presets survive it: RANGES's closest framing is fill: 0.95, which at
the default lens is 1.95 radii and never reaches the floor.