Home The Same-Light Study
The Same-Light Study
What this is
An open measurement of a single question: how far does a skin reading move when only the lighting changes? Participants photograph themselves under three lights, three times each, within one minute. We publish the spread — not accuracy, not validation. Collection is running; there are no numbers on this page yet.
Status: open, no results. This page is deliberately published before the data. If the numbers appear later and the method has quietly changed to fit them, you will be able to tell, because this version is dated and in public.
Why bother
Every product in this category, including ours, asks you to believe a number derived from a photograph. None of them publishes how much that number wobbles.
That is an awkward gap, because the wobble is knowable and it is cheap to measure. Two sources of movement matter:
- Repeat spread. Photograph the same face three times in one minute under identical light. Any difference between those readings is the measurement talking to itself — it is the floor below which no real change can be detected.
- Lighting spread. Photograph the same face under three different lights within the same minute, so the skin genuinely has not changed. Any difference is the light being read as skin.
If the lighting spread is larger than the change a month of skincare produces — which is entirely possible — then "same window, same time" is not a nice-to-have piece of advice. It is the difference between a measurement and a mood ring, and we would rather know.
Method
Recruit 20–50 participants
Friends, volunteers, and people who reply to a post. This is a convenience sample, not a representative one, and that limitation gets stated everywhere the results appear — not in a footnote.
Three lights, three shots each, inside one minute
Indirect daylight facing a window · overhead bathroom or ceiling light · one warm indoor lamp. Three photographs in each, nine in total, all within about a minute so that skin itself has not had time to change. That constraint is the whole design: it is what lets us attribute any difference to the light.
Same device, no filters, neutral face
One phone per participant throughout, beauty modes off, no makeup applied or removed mid-session, neutral expression, camera at eye level.
Analyse each photo with the same model and prompt
Every image goes through the identical pipeline the app uses, with the same fixed phrase pool and the same JSON contract, so the only difference between two readings is the image.
Delete the images, keep the numbers
Photographs are deleted after analysis. What remains is five scores, a lighting label, a repeat index and a participant number.
Publish spread, dataset and method
Standard deviation and range for repeats and across lighting, per dimension, plus a breakdown by self-reported skin tone if the sample supports it. CSV released alongside.
What we will report — and the words we will not use
| We will report | We will not report |
|---|---|
| Repeat spread: how far the same photo situation moves with nothing changed | Any figure described as accuracy |
| Lighting spread: how far a reading moves across three lights in one minute | Any claim of clinical or dermatological validation |
| Which of the five dimensions is most and least affected by light | Any comparison of our numbers against a professional assessment |
| A breakdown by self-reported skin tone, if the sample supports one | A tone breakdown from a sample too small to support one |
| The full CSV and this method | Selected figures without the dataset behind them |
The vocabulary matters enough to fix in writing. This study measures consistency. It cannot measure accuracy, because there is no reference standard in the design to be accurate against. Anyone — including us — who later describes this as showing that SkinLately is accurate will be saying something this design cannot support.
The five dimensions being measured
- Hydration — how comfortable and plump the skin looks, rather than tight or thirsty.
- Texture — how even the surface reads — smooth versus bumpy or rough in patches.
- Calmness — how even the tone is. Less pink and blotchy reads as calmer.
- Clarity — how clear the surface is, without naming or judging anything on it.
- Under-eyes — how bright or shadowed the under-eye area looks today.
Limitations, stated before the results exist
- Convenience sample. Volunteers who know the person running the study are not a random sample of anyone.
- Small N. At 20–50, a skin-tone breakdown may not be supportable. If it is not, we will say so rather than publishing an underpowered split.
- Mixed devices. Each participant uses one phone throughout, but phones differ between participants. That is a limitation for absolute comparison and deliberately not one for within-participant spread, which is what we are measuring.
- Three lights is not all lights. The three chosen are the common household cases, not a lighting taxonomy.
- We built the thing being measured. The obvious conflict of interest. The mitigations are publishing the method first, publishing the raw data, and fixing the reporting vocabulary in advance — which is what this page is.
Consent and photographs
One page, plain language, signed before anything is taken. It says what is collected, that photographs are deleted after analysis, that only aggregate numbers and an anonymised CSV are published, that no face is ever published, and that a participant can withdraw at any point before publication with no reason required.
No face leaves the study. Not in the paper, not in a talk, not as an illustration, not blurred, not cropped. If we ever want an example image we will take one of the founder's own face and say so.
Frequently asked
Are there results yet?
No. Collection is running. This page exists before the data because the method should be public before the numbers are — a study whose protocol appears after its findings is a study with a free hand.
Will you publish it if the numbers are bad for you?
Yes, and that is the only reason to pre-register the method. If it turns out a bathroom light moves a dimension by fifteen points, that number goes on this page, into the FAQ, and into how the app talks about its own readings.
Is this a validation study?
No, and we will not describe it as one. It measures consistency — how much a reading moves when only the light changes, and how much it moves when nothing changes at all. It says nothing about whether a reading corresponds to anything a professional would observe. That would be a different study with a different design and a much larger budget.
What happens to participants' photos?
Each photo is analysed, the five numbers and the lighting condition are kept, and the image is deleted. No face is published, no face is stored after analysis, and the released dataset contains only numbers. Consent is written and in plain language, and you can withdraw before publication.
Can I take part?
Yes — it needs about ten minutes and three lights you already have. Email hello@skinlately.com and you will get the consent form and the capture checklist back. There is no payment and no product attached; taking part does not sign you up to anything.
Will you release the raw data?
Yes — a CSV of the per-photo scores with participants identified only by a number, so anyone can recompute the figures. Open data is also, frankly, the reason a study like this gets linked to.