A complete protocol for comparable submissions

A protocol is not a way to score higher. It is a way to make a result mean the same thing twice, and that is a different goal with a different set of rules. Ratecock.com's photo guide will tell you how to take a photo a tool reads well. This is the other document: how to take a photo you can take again, on purpose, and trust the comparison that follows.

Eight variables move a result without touching the subject. Fix all eight, write down what you fixed them to, and a second submission becomes evidence instead of a fresh guess. None of the eight is exotic or expensive; the whole cost of the protocol is discipline, applied consistently, over the entire series rather than once.

Device

Use the same phone every time, not the newest one. A device change swaps the lens, the processing pipeline and the colour response all at once, and none of those changes are about the subject. Write the model down in the log the first time, not just "my phone" - a note that reads "my phone" is useless the day the phone changes and you need to know exactly what changed. Why the same old device beats a better new one is worth reading before you upgrade mid-series, because the temptation to switch the moment a better camera arrives is exactly the moment the protocol is most at risk.

Distance

Pick a distance and note it in a form you can reproduce: "arm's length," "one metre," a mark on the floor or a length of string held taut. Phone lenses are wide-angle, which means anything close to the lens is rendered larger relative to anything further back, within the same frame - move six inches closer and the proportions the tool reads have already changed before anything about the subject has. Distance is the single variable that moves a score the most, and it is also the easiest one to fix, because it costs nothing to stand in the same spot twice. A physical mark - tape on the floor, a fixed piece of furniture used as a reference - removes the guesswork better than trying to remember a distance by feel.

Angle

Shoot level, not down. A downward tilt foreshortens the subject, and how much depends on precisely how many degrees you tilted, which is not something you can eyeball consistently between sessions - a tilt that feels the same to you can differ by several degrees from the last one, and the tool has no way to tell you that happened. The geometry of a downward tilt is the same geometry that makes distance matter: both compress or stretch what the lens is reading before the tool ever sees it. What that geometry means for the tool's read is a subject that site covers on its own terms; here it only matters as a reason to keep the angle fixed. A simple check: prop the phone rather than hand-hold it, using a level line in the room - a doorframe, a shelf edge - as a reference for level, the same way each time.

Light

Diffuse, side-on, and from a source you can return to. Window light on an overcast day repeats better than anything involving direct sun or a flash, because it does not depend on the hour or on distance to the source the way a lamp or a flash does. Flash flattens texture and varies with how far you are standing from whatever you are pointing it at, which makes it two variables pretending to be one: intensity changes with distance, and direction changes with wherever the flash unit happens to be aimed that day. Even with window light chosen, the hour still matters - the same window delivers different colour and direction at different points in the day, so the light source and the time of day have to be fixed together, not separately.

Crop and framing

Decide how much of the frame the subject fills, and repeat it. Cropping changes what fraction of the image is subject and what is context, and both of those are inputs the tool reads, not neutral framing choices - a tighter or looser crop is not a stylistic decision here, it is a variable with the same status as distance or angle. A background that changes between shots adds noise to exactly the axes that are already the least stable. Background clutter is not ignored by the tool; it is scored along with everything else in the frame, which is why a plain, consistent backdrop is worth setting up once and reusing rather than improvising each session.

State

Note the state the subject is in. A submission series only holds together if the subject is in a comparable state each time; state is a variable like any other, and no protocol before or after the shutter corrects for it. This is the one line in the log that cannot be verified from the photo alone the way distance or crop can be checked against a mark on the floor - it depends on honest self-report at the time, which is exactly why it belongs in the log rather than in memory.

Which variables matter most

Not all eight carry equal weight, and knowing the order helps when time or patience for a full setup is short. The submission variables ranked by their effect on a score put distance and angle near the top and framing choices like background further down - useful to know if a session has to be abbreviated, because it tells you which corners are more expensive to cut than others.

Recording, not remembering

Eight facts, written down every time: device, distance, angle, light source, time of day, crop, background, state, and the date. A log is what turns a pile of results into a series; without it, a March result and a June result are two unknowns compared to each other, not one measurement repeated under known conditions. This is also where a baseline earns its name - the first entry in the log, the one every later entry gets compared against, not because it is a better photo but because it is the one whose conditions are on record and can be checked against.

Taking more than one

A protocol that is followed once produces a single data point, and a single data point cannot tell you whether a later change is real or noise. Five retakes under the same fixed conditions give you a centre and a spread rather than one number standing in for a whole distribution, and the spread is the part that keeps later comparisons honest. This costs one extra photo session, not a change to any of the eight variables above - the protocol stays identical, it is simply run more than once before you draw a conclusion from it.

A protocol like this pays off whether the destination is an automated tool or a person. A human judge still responds to whatever material you send, and a consistent set beats an inconsistent one there for the same reason it does here: variance that has nothing to do with the subject is still variance, whoever or whatever is doing the looking.

Common ways the protocol fails in practice

The most frequent failure is not a dramatic one - it is quietly loosening one line of the eight over successive sessions because it feels close enough. A crop judged by eye instead of against a fixed reference, a distance estimated rather than marked, a light source that has drifted from "diffuse, side-on" toward whatever was convenient that day: none of these look like a broken protocol from inside a single session, and all of them are. The second most frequent failure is skipping the log itself, on the assumption that the setup is memorable enough not to need writing down - it rarely is, months later, when the comparison actually needs the record. Both failures share the same fix: treat the log as the protocol, not as paperwork attached to it, and check the current session against the log rather than against memory.

What the protocol is not

It is not a measurement method. A length in centimetres comes from a tape and a documented method, never from a photograph, and nothing in this protocol substitutes for that - if the property you care about is a physical length, this document is the wrong one to be reading. It is not a route to a better number either. A standardised photo and a flattering one are frequently not the same file, and optimising for the higher score is a different project with a different, worse outcome for anyone trying to track a real trend rather than chase a single good result.

Running it

Once the conditions are fixed and logged, the protocol is only useful if you actually use it more than once. A single standardised photo is better than an unstandardised one, but it still only tells you where one number landed on one occasion. The way to find out whether a gap between two results is real or noise is to hold the protocol still and take a second shot under it - a separate procedure from the standardisation itself, worth reading on its own terms once this one is in place, since it asks a different question of the same fixed setup. None of this requires expensive equipment or an unusual amount of discipline. It requires writing eight facts down and being willing to stand in the same spot, at the same time of day, holding the phone the same way, more than once. That is the entire cost, and it is the only thing that turns a photo into data rather than an anecdote with a number attached to it.