The first thing to do with any tool

The first thing to run on any rating tool is a retake test: take a photo under conditions you can hold, take a second under the same conditions, and submit both. The gap between the two numbers measures everything except the subject, and every comparison you run afterward inherits it.

A different test from resubmitting one file

This is not the identical-file test, which resubmits the exact same image to see how much the tool alone varies. The retake test asks a slightly bigger question: given the same phone, same distance, same angle, same light, held as tightly as you can manage, how much does a fresh photo differ from the last one? That gap includes the tool's own noise, but it also includes your ability to actually hold "the same conditions" in practice - your hand is never in exactly the same place twice, the light source flickers slightly, the crop is close but not pixel-identical. Metrology has a name for the setup you are aiming at: the international vocabulary of metrology (JCGM VIM, 2.20) defines a repeatability condition as the same procedure, operators, measuring system, operating conditions and location, with replicate measurements "over a short period of time". The full protocol this rides on sets out what "same conditions" means in the first place; the retake test is what tells you how well you actually held it.

How to run it

Shoot the second photo in the same session as the first, minutes apart, changing nothing you have control over. Submit both, and record both numbers with the date and the fact that they were a retake pair, not two separate baseline entries. Do this once before you start trusting the tool for anything comparative, and do it again periodically, because your own consistency at holding a protocol tends to improve or slip over time in ways worth knowing about.

Reading the gap

A small gap is the expected outcome and tells you the setup is tight enough to trust for moderate comparisons. A large gap is information, not failure - it says something in the conditions moved more than you assumed it did, and it is worth checking the log against the room before blaming the tool. The opposite surprise is worth naming too: a retake pair that lands almost identical, closer than the expected noise would predict, deserves a second look rather than being read as an unusually clean result. How much two retakes should actually differ, and what a larger-than-usual gap points at goes through this in more detail once you have a pair of numbers in hand to interpret. On a tool that publishes a per-axis breakdown, like Rate Cock, the retake test is more informative still, because a retake pair can show which specific axis moved rather than only a single total shifting by an unexplained amount.

What a retake pair cannot tell you

A retake pair does not separate your own consistency from the tool's own noise - it reports the combined size of both. That is a limitation worth naming rather than working around here, because splitting the two apart needs a second, different test rather than a cleverer reading of the same pair. Treat the retake gap as a ceiling: the true noise contributed by the tool alone is somewhere at or below that number, never above it, which is still useful even without the split.

Why this comes before anything else

Every later question - is this tool more generous than that one, did this change actually move the score, does a half-point shift mean anything - is a question about a gap between two numbers. None of those questions are answerable until you know the size of gap that shows up from pure retake noise, because that gap is the floor beneath which nothing else is distinguishable from chance. Reliability research treats it the same way: Koo and Li (2016, Journal of Chiropractic Medicine) describe test-retest reliability as the variation in measurements "taken by an instrument on the same subject under the same conditions". Five retakes rather than two give you a centre and a spread instead of a single pair, which is the natural next step once the basic retake test has told you the noise is worth taking seriously.

Where the same logic applies outside a rating tool

A documented measuring method reports its own repeat-measurement variation for the same reason - a single measurement is a draw, and knowing how much a repeat typically differs is what makes a later comparison meaningful rather than superstitious. A model's output has an internal noise floor of its own, separate from anything you contribute through the camera, and the retake test does not separate the two - it measures both together, which is exactly what you need if the two numbers you are about to compare will also carry both. The retake test is cheap: two photos, one sitting, one number to write down. It is the first thing worth running with any tool, including a human reviewer if you intend to send them a series rather than a single image, and skipping it means every later comparison is resting on an assumption you never actually checked.