SPATIAL INFERENCE

Reconstruct 3D space from 2D visual cues.

Depth is not given in the retinal image. It is computed. The posterior parietal cortex — specifically the intraparietal sulcus — reconstructs three-dimensional space from a fundamentally two-dimensional input, using occlusion, relative size, texture gradient, linear perspective, and atmospheric haze as evidence.

This computation is embodied — built on a lifetime of moving through physical space, of reaching and misjudging and correcting. Human depth inference generalizes to any scene because it is grounded in physics, not statistics. AI depth estimation models are grounded in training distributions. They fail on novel scenes in ways humans do not.

Depth Field measures monocular depth inference specifically. No stereoscopic tricks. Only the cues available to one eye in a flat image — the same cues a painter uses, a surgeon uses, a pilot uses. This is a learned perceptual skill. The scores reflect that.

Your SPATIAL INFERENCE is unmeasured.