The subject is not constant either

Yes, the same person photographed at a different hour can score differently, because the subject is not constant between shots. Skin and surface appearance respond to temperature, time of day, activity and general physiological state, and a model has no channel through which to know any of that changed.

Why the model cannot tell

Nothing in a scoring pipeline carries information about when a photo was taken relative to your body's state, and even if a timestamp were available, the model was never trained to use one. It receives a grid of pixels and treats it as a complete, self-sufficient description of whatever it shows, because that is the only kind of input it has ever been given. Two photographs of the same person, taken an hour apart under different conditions, arrive at the model as two unrelated images with no shared identity attached. The model is not failing to notice that they are the same subject in different states; it has no representation of "subject" and "state" as separate things to begin with, only the image in front of it.

What actually varies, in general terms

Skin and surface appearance respond to temperature, and rooms, seasons and time of day all shift ambient temperature in ways a person barely registers but a camera and a downstream model record faithfully. Circulation and skin tone shift with temperature and activity level, changing colour, sheen and surface texture, all of which are properties a vision model reads directly, because texture and tone are exactly the kind of low-level signal a scoring head has learned to weigh. The circulatory swing is large: in a review of human thermoregulation, Charkoudian (2003, Mayo Clinic Proceedings) notes that skin blood flow can reach 6 to 8 litres per minute during hyperthermia, under the control of sympathetic nerves that both dilate and constrict skin vessels. General physiological state - rested versus tired, relaxed versus not, recently active versus not - shows up as small differences in posture, muscle tone and skin appearance that a person would attribute to "just how I looked right then" and a model has no way to discount, because discounting requires knowing a baseline exists, and the model was never given one.

None of this is a claim about size, and none of it should be read as one. This is about surface appearance and shape-adjacent cues shifting between shots of the same person under different conditions, which is enough on its own to move a score without anything about the subject's underlying dimensions having changed at all.

Why "the same photo" is doing more work than it sounds like

When someone says they retook a photo of "the same subject" and got a different score, the implicit claim is that only the camera and the framing changed. Often that is not true. The room was warmer or cooler, the time since waking or since activity was different, and the state of the subject at the moment of the shutter was not identical to the first attempt, even though a person looking at both photos might describe them as showing the same thing. The broader catalogue of what makes the same subject score differently covers camera and framing variables in detail; this is the one variable on that list that lives in the subject rather than the setup, and it is easy to miss precisely because it feels like nothing changed.

A useful way to think about it: a model given two photos is not asking "is this the same person as before." It is asking "what does this image, on its own, look most like," independently each time, and a subject in a different physiological state produces an image that is, in the narrow sense the model works in, genuinely a different input. This is also why a lower retake after an unusually high first score is not necessarily a sign of anything - regression toward the mean explains part of that pattern on its own, before physiological variation is even considered.

Where this stops

This post stops at the observation that state varies and the model cannot see it happen - it does not extend into anything about physical measurement, which is a different question with a different answer entirely. Measure My Cock's method pages are where any question about controlling conditions for a physical figure belongs, because a tape and a fixed protocol are the tool built for holding a baseline steady, not a scoring model. For the score side specifically, Rate Cock returns a number for the photo in front of it and nothing more, which is worth remembering the next time two attempts, taken hours apart, come back different. A human reviewer can sometimes ask about timing directly rather than guessing at it from the image alone - Rate Penis's judge process allows for that kind of context a still photo cannot carry - and reading a single result against that background is exactly the caution Penis Rater writes about from the user's side.