A term from drawing, applied to submissions

Foreshortening is an old term from drawing and photography: an object angled toward or away from the viewer appears shorter along that axis than it would viewed side-on, because the camera is recording a projection, not the object itself.

The definition, precisely

A length pointed straight at the lens contributes almost none of its true extent to the frame - it is compressed nearly to a point. The same length held perpendicular to the lens, level, shows its full extent. Every angle between those two extremes compresses the length by an amount that depends on the tilt, not on anything about the subject.

The diagram in words

Picture the lens as a single point and the subject as a line. Rotate that line from perpendicular (full length visible) toward pointing straight at the lens (length collapses toward zero). The apparent length at any angle in between is the true length multiplied by the cosine of the tilt from perpendicular - a small tilt costs little, a large one costs a lot, and the cost accelerates as the tilt increases. Distance adds its own projection effect on top: Ward and colleagues (2018) modelled face photos and found a 12-inch shot enlarges the nose by 30% in men and 29% in women against a distortion-free projection, with "essentially no difference" at 5 feet.

What it means for a rating

A rating tool scores what is in the frame, and what is in the frame at a downward or upward tilt is a foreshortened version of the subject, not the subject as it would appear level. This is the mechanism behind why camera angle moves a score so directly - a small, hard-to-notice-by-feel change in tilt produces a real change in what the tool sees, which is why level is the one angle that can be reproduced reliably rather than approximated.

It also means a multi-angle set will show genuine variation for reasons that have nothing to do with anything except geometry - averaging that variation away hides the fact that it was foreshortening, not the subject, that moved between angles.

Rate Cock's per-axis entries make this kind of geometric artefact easier to spot than a single total would, because a proportion axis that swings between two otherwise-similar submissions is a reasonable prompt to check the angle before concluding anything else changed. The vision models behind these tools read the frame as given; nothing in how they process an image corrects for the tilt it was taken at, so the correction has to happen in the photo, not after it.

A tape measurement is immune to this entirely, since it reads length directly rather than through a lens at an angle - foreshortening is a photography problem, not a measurement one. A human reviewer can often compensate for it by eye in a way a model does not reliably do, which judges account for informally when they ask for a level shot rather than whatever angle was easiest to take.