The rule that makes a comparison readable

Change one variable per comparison, because a score change between two submissions can only be pinned on a cause when exactly one thing differed. If more than one thing changed, there is no way to say which change did it, and dividing credit between two causes from a single number is not something the number supports.

Why this happens so easily

It rarely happens on purpose. A new phone tends to arrive with a new distance you stand at, a new light in the room you happened to be in, and a slightly different crop habit, all inside the same submission. Any one of those alone would be a clean, readable test. Together, a score change is uninterpretable - it could be the lens, the light, the framing, or some combination the number cannot separate out.

This is the same problem underneath mixing a mirror shot with a front-camera shot, where switching "how you take the photo" silently swaps two or three variables at once inside what feels like one decision. It is also why subject state has to be held constant on its own rather than allowed to drift alongside a camera change - two moving parts, one result, no way to split it.

The discipline

Before changing anything for a comparison you actually care about, ask what single thing is different this time. If the honest answer is "a few things," the comparison is not ready yet - fix everything except the one variable you want to test, and only then submit. A geometric change like angle or distance is worth isolating this way specifically because its effect is large enough that it would swamp a smaller variable riding along with it, and you would never notice the smaller one moved at all. How large one geometric change can be is documented outside rating tools: Ward and colleagues (2018, JAMA Facial Plastic Surgery) calculated that a photo taken 12 inches from a face makes the nose appear 30% larger in men and 29% larger in women than an undistorted projection, while at 5 feet the difference essentially disappears.

What this buys you

A single-variable comparison is attributable. A multi-variable one is not, no matter how carefully you log the conditions around it - the log tells you what happened, but only a clean comparison tells you what caused it. The same logic holds outside photography entirely: a measurement taken with a different tape and a different method on the same day is two changes, not one, and the same discipline that keeps a photo comparison honest keeps a measurement one honest too. A tool that shows a breakdown by axis, the way Rate Cock does on public entries, makes a clean single-variable test more informative still - you can see which specific axis moved rather than only that the total did, which is most of the reward for doing this properly in the first place. A human judge forms an impression from everything at once and rarely isolates a single variable at all, which is a different kind of feedback than this discipline is built to produce - useful in its own way, but not a substitute for it.