Every variable that sits between the subject and the number

A rating is the end of a chain: subject, device, optics, light, framing, file, tool. Every link can move the number, and they do not move it equally. Distance, angle and light direction are strong effects; resolution above a threshold is weak, and much else is folklore. This guide walks the chain and routes to each deeper post.

Why a chain, and why order matters

A score describes the submission, not the person, as argued in the score describes the submission. A submission is whatever arrives at the tool, and between the subject and the tool there are several stages, each of which can change what arrives. Thinking of it as a chain helps in two ways. First, it tells you where to look when a result surprises you: work back from the number, link by link. Second, it tells you where effort pays, because the links are not equal, and spending care on the weak ones while the strong ones drift is the most common waste in this whole subject.

The ranking of effect sizes in this guide is qualitative. The ranked version gives an order, and no source publishes a universal number, because the effect of a variable depends on the tool. Where a magnitude is not known, this guide says direction only.

Link one: the subject's state

Before any equipment, the subject itself varies. State, as in what condition the subject is in at the time of the shot, is a variable like any other, and no camera protocol can correct for it. Same subject, different state is the post on this, and the one-line version is that a series only holds together if state is held, logged or both. It belongs first in the chain because it is easy to forget that the chain starts before the camera.

Link two: the device

The phone matters through its lens, its sensor and its processing. A new phone changes all three at once, so for comparison the same device every time beats a better one. Phones with several lenses switch between them silently as you change zoom, and each renders proportion differently, as covered in focal length as a hidden variable. The camera app's processing, including scene detection and automatic enhancement, is part of the device too; turning it down where possible reduces variation.

Is a phone good enough? Yes, with the condition that it be the same phone; that question is answered directly. This link is strong for switching devices and weak for choosing a particular good one.

Link three: the optics, including the dirt on them

Optics here means the lens as it is on the day. A clean lens is a variable you can fix in seconds, and a dirty one is a quiet source of haze and lost contrast; Gu, Ramamoorthi, Belhumeur and Nayar (SIGGRAPH Asia, 2009) describe lens dirt as a layer that attenuates light and scatters stray light toward the camera. A smudged lens covers the practical side. Blur from focus or motion belongs here too, since it removes the detail some axes depend on, and focus and motion blur covers it.

Strength: modest per instance, but variable between sessions, which makes it a source of unexplained noise out of proportion to its size.

Link four: distance, angle and perspective

This is where the largest effects live. Moving the camera changes the proportions the tool reads. Distance is the single largest source of score movement unrelated to the subject, and tilt foreshortens in ways you cannot see from where you stand. The geometry of why is in foreshortening, as it applies to a rating.

Digital zoom sits here as well, because it changes what the tool receives without changing where you stand; digital zoom or moving closer separates the two. Strength: strong. These are the links where a protocol earns its keep, and the ones to fix first.

Link five: light

Light has three parts: direction, colour and amount.

• Direction shapes relief and shadow and is the strongest of the three; see light direction and the score.
• Colour shifts the whole image's tone, and phones try to cancel it with auto white balance, which is itself a variable; see warm vs cool light and white balance and colour cast.
• Amount affects exposure and noise; see exposure, ISO and noise, and note that brighter is not better.

If any of your light is daylight, it also varies with the hour and with the year, as time of day and seasonal light describe. Strength: direction is strong, colour is moderate and tool-dependent, amount matters mainly at the extremes.

Link six: framing and cropping

What fraction of the frame the subject fills, what surrounds it and how it is cropped are all inputs. Cropping decides both what goes in and what comes out, and framing and cropping differ in whether perspective changes. The background is part of this link, since the tool encodes the whole image rather than the part you care about; background clutter explains the consequence. Orientation is small but real. Strength: moderate, and easy to fix, which is why it is worth fixing.

Link seven: file handling

The tool receives a file, not a scene. What happens to the file between your phone and the tool is the part most people never consider. Screenshots, forwarded copies and messaging-app recompression change the image; JPEG compression and screenshots is the full account. Edits and filters change the image deliberately and make the result incomparable with any unedited one, as filters and edits sets out.

Resolution is the folklore-prone part of this link. Below a threshold it costs detail, and above it more pixels change nothing except upload time; resolution and the score draws the line. File metadata is worth a mention: the Exif standard (CIPA, Exif 3.1) is the format that lets a photo carry its capture settings with it, which is a quiet reason to submit the original file and keep a copy. Strength: compression and editing are strong when they happen; resolution is weak once you are above the threshold.

Link eight: the tool's own pipeline

After the file arrives, the tool does things you cannot see: resizing, cropping on input, encoding, scoring against a rubric, mapping the result to a scale and rounding. Some of this is covered in what normalisation is in scoring and what a rating tool actually looks at. Two properties of the pipeline matter for variables.

First, it changes over time, which is why calibration drift means last year's score sits on a different scale. Second, it has noise of its own. The identical file submitted twice shows how much, as in the reproducibility test you can run on any tool, and noise in the photo or the tool separates it from yours.

Whether the pipeline compensates for your errors is checkable and often untrue, a point the tool corrects for angle - does it? tests directly. Strength: not controllable by you, but measurable, and it sets the floor beneath all the other effects.

Strong, weak and folklore, sorted

The folklore entries share a shape: advice that makes a photo better rather than more repeatable. A better photo is not a more accurate rating, and a better photo can score lower.

Turning the map into practice

Fix the strong links first, log them, then work down. The reference-conditions protocol is the way to hold them, the pre-submission checklist is the minute-long version, and a control shot separates tool drift from your own. When two sessions need comparing, that procedure is its own post.

Some variables cannot be fixed, such as ambient light on an overcast week or the state of the subject, and the honest response is more samples and wider error bars; what you cannot standardise covers it. Several repeats give a spread, and the spread is the yardstick every other effect on this page should be judged against.

What lives elsewhere

How a model turns an image into a number is part of how these tools work, and it is why every link above reaches the score at all. Equipment choices that make sense for measuring things, rather than rating them, are a gear conversation that this site does not enter. Where a rating tool sits in a business, and what its owner is optimising, is its own commercial topic and affects the pipeline link only indirectly.

Rate Cock is one instance of the category that documents its accepted inputs, which is the checkable property to look for in any tool: a tool that says what it will take is a tool whose chain you can start to hold constant. A tool that does not say is a tool for which every link above is a guess, and the first test of any of them is whether the same file, sent twice, comes back as the same number.