The term this site keeps using

A standardised submission is a photo taken under conditions you wrote down and can reproduce: the same distance, angle, light and crop as last time, recorded well enough that "last time" is not a guess. The point of writing them down is not the photo itself. It is the second photo, the one that gets compared against it, which only means anything if the two shared the same conditions.

What it is not

It is not the most flattering photo, and it is not the sharpest or best-lit photo you are capable of taking. Optimising for the best single result and standardising for a repeatable one are different goals, and only one of them survives a second attempt - the flattering photo gives you a number you cannot get again on purpose, the standardised one gives you a number you can. It is also not a claim about accuracy. A standardised submission removes variables that are not the subject; it says nothing about whether the tool's judgement of the subject itself is right.

What it excludes

Anything that changes the geometry or the light without your noticing does not belong: moving closer between shots, switching which lens the phone selects, shooting one session in window light and the next under a ceiling bulb. None of those are dramatic on their own, but each is a hidden second variable riding along with whatever you meant to test, and a comparison with two things changed at once cannot tell you which one moved the number. The full list of what has to stay fixed, and what to do about the conditions that genuinely cannot be pinned down, is the reference conditions protocol - this post is the definition, that one is the checklist. Even a careful protocol leaves a residue behind: the handful of variables no amount of standardising actually removes is worth reading alongside this definition so the term is not mistaken for a promise of full control.

Why the definition matters

The word gets used loosely, including on this site, so it is worth being precise about what it buys: a result you can put next to another result and trust the gap between them means something, on a tool that reports enough of a breakdown to see what moved rather than just a total that swallows the difference. How a model actually reads a submitted file is unaffected by whether the submission was standardised, but how much you can trust a change in its output is entirely dependent on it. The photo side of this discipline exists because the tape side already has one - method-first measurement is the older habit, and a standardised submission is the same idea applied to a camera instead of a tape. It also carries over if the photo is ever going to a person doing the judging instead of a model: consistent conditions read as more comparable to a human reader too, just for slightly different reasons.