Which variable to fix first

Distance and camera angle move a rating score the most, followed by lens choice, then light, crop and edits, with background and resolution last. A protocol checklist gives every variable the same one line, which flattens that difference; if you are starting from nothing, the order matters more than the completeness.

This is a directional ranking, not a set of numbers - nobody has published magnitudes for rating tools, and a specific figure for them here would be invented. What follows is the order the mechanisms suggest, roughly largest effect to smallest.

The variables that move the most

Distance. Phone lenses are wide-angle, and anything near the lens is enlarged relative to anything further back in the same frame, so a change in distance changes the proportions the tool reads directly. This is usually the single largest source of unintentional score movement, because it is also the easiest variable to drift on without noticing. The underlying geometry has been measured for faces: Ward and colleagues (2018, JAMA Facial Plastic Surgery) calculated that a photo taken from 12 inches makes the nose look 30% larger in men and 29% in women than an undistorted projection, with essentially no difference at 5 feet.

Angle. A downward tilt foreshortens, and the amount depends on exactly how far down the tilt went - not something you can judge by feel session to session. Camera angle is the second big lever, for the same geometric reason as distance: it changes what the lens sees before anything about the subject enters into it.

Lens or camera path. Switching between a phone's rear lenses, or between the rear camera and the front camera, changes focal length and field of view outright. Focal length is a hidden variable precisely because most people do not realise a phone has more than one lens behind the same shutter button.

The variables with a moderate, real effect

Light direction and quality. Where light comes from decides what the image shows in relief, and flash versus window light is not a cosmetic choice - it changes what detail survives into the file. Light direction is a score variable in its own right, sitting a rung below distance and angle but well above cosmetic choices like background colour.

Crop and framing. What fraction of the frame the subject fills, and what surrounds it, are both inputs a tool reads whether or not you meant them as data. Cropping puts things in and takes things out of the score at a smaller scale than distance, because a crop only changes the frame, not the geometry the lens already captured.

Filters and edits. Any smoothing pass, colour grade or beautification filter changes the file the tool actually receives. The subject does not change and the score still does, and because the change is deliberate it is also the easiest of these to simply not do.

The variables with a small but real effect

Background and clutter. A busy background is not ignored by the model - it is scored alongside everything else in the frame - but its effect is smaller than any variable that changes the subject's own proportions or texture.

Resolution, within reason. Below a threshold, resolution costs real detail; above it, more pixels do nothing except slow the upload. Most modern phone cameras clear that threshold by default, which is why this sits near the bottom of the list rather than off it entirely. Compression belongs down here too: testing image classifiers against five distortions, Dodge and Karam (2016) found them particularly vulnerable to blur and noise, with JPEG and JPEG2000 compression among the effects they tested.

Why the order matters

If you can only fix two things before your next submission, distance and angle are the two - they are also the two the tool has the least ability to correct for on its own, unlike resolution or compression, where a tool's preprocessing may already smooth some of the difference out. This list is about photographs specifically; none of it applies to a measurement taken with a tape, where distance and angle to a lens are not part of the method at all.

None of this replaces the whole protocol. The reference conditions worth fixing, taken together, still cover every item on this list - this is only a guide to which one to tackle first when you are building the habit from nothing, and a reminder that a person judging directly discounts most of these instinctively in a way a fixed model cannot. A tool that shows a breakdown, the way Rate Cock does on public entries, is also the easiest kind to use for checking this list yourself: change one variable, rescore, and see which axis moved.