Find out if your skincare is actually doing anything.
And stop paying for the parts that aren’t.
Kitchen, 08:00
Warm bulb · 2700 K
42
naive redness
Bathroom, 08:20
Window · ~6500 K
31
naive redness
−26%
42 → 31
One photograph, cut down the middle and developed at two white balances — the operation your phone performs on you between two rooms. The face runs straight through the seam because it is the same face, twenty minutes apart. Only the naive redness metric moved.
Illustrative example. The face is generated — not a participant, not a result — and the readings are sample data.
Words you won’t see here
- detoxify
- glow-boosting
- clinically proven
- skin age
- flawless
- problem areas
- miracle
- anti-ageing
- reveal your best skin
- 97% accurate
What you actually get
Three screens. That’s the whole product.
No score, no grade, no percentage a model made up. Measurements against your own baseline, and what follows from them.
- 01
The reading
Every two weeks, the same regions, each with the band we can't see into. Here the forehead moved and the cheeks didn't — and it says exactly that.
- 02
The routine
Layering order, the sunscreen an active makes non-optional, what it costs a month — and what a trial retired, with the money back.
- 03
The picks
Ranked by ingredient fit, then price per ml. The cheapest one that fits goes first, even when it earns us nothing.
Illustrative example. The face is generated — not a participant, not a result — and the readings are sample data.
The problem nobody mentions
Most skin apps are measuring your bathroom.
The photograph at the top of this page was one frame. So is this one — a second face, the same two rooms, the same twenty minutes.
Kitchen, 08:00
Warm bulb · 2700 K
38
naive redness
Bathroom, 08:20
Window · ~6500 K
25
naive redness
−26%
42 → 31
the face at the top
−34%
38 → 25
this face
Same operation, same two white balances, two people. The deeper skin loses more: less of the redness signal survives to be read there in the first place. That is why the app widens the band it can’t see into on deeper skin rather than pretending it is as sure as it was.
Illustrative example. The face is generated — not a participant, not a result — and the readings are sample data.
Nothing about that skin changed. The light did.
Between two sessions a phone camera varies by far more than any cream does in eight weeks — colour temperature, auto white balance, exposure, angle, distance, and whatever smoothing it applied without telling you. The nuisance is one to two orders of magnitude bigger than the signal. So the progress chart in an uncontrolled selfie app isn’t tracking your skin; it’s tracking where you stood.
So we didn’t start with the AI. We started with the camera.
The rule that changes everything
We measure our own wobble, then refuse to report anything smaller.
When you sign up you take two photos twenty seconds apart. Nothing about you changed in twenty seconds — so every difference between them is our error, not yours. That gives us the smallest change we can honestly claim to detect on your skin. Anything below it, we call what it is.
The shaded band below is the zone we can’t see into. It’s on every chart in the app, and there’s no setting to hide it.
Down 2.4 points. Our measurement wobbles by about ±1.9 on your skin, so this is a real change — just.
Beyond measurement noiseDown 0.8 points. Our measurement wobbles by about ±1.9, so we can't tell this apart from noise.
No detectable changeWhy this is the whole product
Every other app in this category needs the answer to be yes.
Trial complete · 10 weeks · azelaic acid 10%
No detectable change.
That’s a real answer, and it’s the one nobody else will give you. Cheek redness moved 0.4 points; we’d need at least 1.8 to call it.
€47/yr
At €39 a bottle, that’s what you can put somewhere else.
Their score rises when you use the app. Their recommendation is an ad. Their revenue depends on the product you’re testing working out.
We rank by ingredient fit and price, and commission is not an input — not as a policy, but because the ranking function isn’t given the commercial terms in the first place. If the €7 one contains the same active, the €7 one goes first.
“It did nothing” costs us money to say. That’s exactly why it’s worth something when we say it.
Specification
The boring page, up front.
Most of this normally lives in a methods section nobody publishes. It's the first thing a sceptic looks for, so here it is.
- Reading cadence
- every 14 days— shorter is below the noise floor
- Regions measured
- 11— identical every session
- Detectable change
- measured per person— your own number, on every chart
- Trial length
- 8–12 weeks— shorter measures irritation, not benefit
- Photo storage
- off by default— opt-in, encrypted, expires at 12 months
- What the model does
- writes the summary— it produces no numbers
- What we can't measure
- hydration, sebum, pore size— not from a phone camera
And what we won’t build.
Each of these would make the product easier to sell. They’re written down so they’re harder to quietly add later.
A 0–100 skin score
It moves when your lighting moves, and it grades your face.
Mole and lesion checks
Being wrong is unforgivable. We point you to a doctor.
Confidence percentages on a face
Those numbers come out of a language model.
Streaks and daily reminders
Daily nudging is anxiety with a progress bar.
Selling or sharing your data
Not at any price. Photos are opt-in and expire.
Sponsored placement
The ranking code never sees commission.
The ambition
We’d like you to scan less.
Skin changes over weeks, not days. So we measure every two weeks, we say “no change” when there’s no change, and when something finally works we tell you to stop checking it. The goal isn’t a shelf full of the right products — it’s a shelf with fewer things on it.
Be there when it’s honest enough to ship. We’ll write once — when there’s something real to try.