The one submission everything else is compared to

A baseline submission is a deliberate first photo: conditions written down, a setup you can recreate, and a short series of retakes rather than one shot. Most people's first rating is whatever photo was convenient that day, a baseline by accident, and everything compared against it inherits what was uncontrolled.

What makes a baseline valid

A baseline is not just an early result. It needs three things a casual first submission usually lacks.

Recorded conditions. Distance, angle, light source and crop, written down at the time - not remembered afterwards, because the four variables that actually matter are exactly the ones memory gets vague about within a few weeks.

Repeatability. A setup you can genuinely recreate: light from a window rather than a specific unreproducible moment of afternoon sun, a distance you measured or marked rather than one you eyeballed once and will eyeball differently next time.

More than one take. A single photo has submission noise built in the same as any later result does - a baseline built from one file inherits whatever that one file's noise happened to be, and a baseline built from a short series of retakes under identical conditions is a steadier reference than any single shot could be. Reliability research leans on the same idea: Weir (2005) shows how the standard error of measurement from repeated trials sets "the minimal difference needed" before a change in one individual can be treated as real, and that figure only exists if the starting point was measured more than once.

What a baseline is for

Every later comparison is really a comparison against the baseline, whether that is stated explicitly or not. "Did this change" only has an answer if there is a fixed "this" to change from, and a change one variable at a time only tells you something if the other end of the comparison held everything else constant too. The baseline is that other end.

It also gives the tool's own behaviour somewhere to be checked against. Resubmit the baseline itself, unchanged, at a later date, and any movement in the score is the tool's drift rather than anything about the subject - which is the specific use a control shot is built around, using the baseline as the fixed reference rather than a separate file.

Setting one up

Shoot the conditions you intend to reuse, take three or four frames under those exact conditions, and use the middle of that small set rather than the single best-looking one - the best-looking one is usually the outlier, not the representative shot. Write the conditions down next to the file: device, lens, distance, light, crop, date, the same list a protocol asks for everywhere else, just applied once, deliberately, before anything else gets compared to it.

A tool that reports a breakdown rather than a single total, the way Rate Cock shows per-axis scores on public entries, gives a baseline more to be useful for - a later comparison can point at which axis moved rather than only that the total did. None of this is about hitting a particular number on the baseline itself; optimising a photo for the highest score and standardising it for comparability are different goals, and a baseline exists to serve the second one only. A baseline built this way is also portable to a different kind of review entirely - the same clean, repeatable photo works for a human judge as well as it works for a model, because the discipline behind it has nothing to do with which one reads it. The one thing a baseline cannot give you is measurement in the strict sense; that belongs to a tape and a written method, and a baseline photograph, however careful, was never claiming to be that.

When a baseline is worth redoing

A baseline is not permanent. A new phone, a room you no longer have access to, or a long enough gap that the original conditions are hard to recreate honestly, all argue for starting a fresh one rather than pretending the old setup still holds. The mistake is not retiring a baseline - it is retiring one silently, so that a comparison across the switch looks continuous when it is not. If the baseline changes, say so next to the next result, the same way a change in any other condition gets noted.

A baseline is also worth revisiting after a long enough run of ordinary submissions, simply to confirm it still matches what you think it does. Conditions drift in small ways even when nobody meant them to - a lamp gets replaced, a habit of standing changes by an inch - and a baseline that has quietly stopped matching its own record is worse than no baseline at all, because it looks trustworthy without being so.