איך מודדים

GRRAS — the fifteen-item standard for REPORTING a reliability study: the device, the subjects, how the sample size was chosen, the statistics, and the intervals. MIRRA is audited against all fifteen of them, and the audit is published with the row that falls short still in it.

Minimal detectable change and minimal IMPORTANT difference are different quantities and one may not stand in for the other. MIRRA's threshold is MDC95: it says a change is bigger than our measurement error. It does NOT say the change matters to you or that it is visible in a mirror. Nothing in the product may phrase a passed threshold as importance.

A smartphone colorimeter reached ICC 0.85–0.95 against VISIA and DermaLab Combo. This is the benchmark MIRRA measures itself against, and it is the reason an ICC below ~0.85 in our own calibration is reported as a limitation rather than rounded up.

The ICC reliability bands — poor <0.5, moderate 0.5–0.75, good 0.75–0.9, excellent >0.9 — and the instruction to judge on the CONFIDENCE INTERVAL, not the point estimate. MIRRA grades every metric on the lower bound of its bootstrap interval for this reason.

Comparison of instrument redness measures against reference devices.

Selfie-based algorithmic grading agreed with 50 dermatologists across ages, ancestries and phototypes (r ≥ 0.75 on five of seven signs) — the premise of the category. It also found the two places it does NOT hold, and both match what our own calibration found independently: cheek PORES only moderate (r = 0.63), and PIGMENTATION weak on the darkest skin tones (r = 0.40). Pores is the weakest metric in our set too (ICC 0.65), and the tone dependence is why weights are discounted by |r(ITA)|.

Anatomical REGION was a larger source of variance than the device itself (η² 0.18 vs 0.12); calibrating per region lifted ITA reliability from ICC 0.78 to 0.93 and cut tone misclassification from ~42% to ~24%. Directly supports measuring each metric in its own facial zone. Its stated limit — a Korean-only cohort, no validation on lighter or darker skin — is precisely the gap our own 102-identity multi-ethnic calibration set exists to cover.

A validated photonumeric scale for infraorbital dark circles, quantified by L* and ITA against a CHEEK control area — the same construction MIRRA uses. ΔE tracked clinical grade at r = 0.99; intergrader r = 0.96; intragrader 90–94% identical one hour apart.

The ESCD Standardization Group on controlling the individual, temporal and ENVIRONMENTAL variables that move a skin colour reading. We meet it partly: pose, expression, capture quality and the weather are recorded and the capture coach enforces framing before the shutter — but a clinic fixes its illuminant and a phone in a bathroom does not.

102 adults photographed neutral and smiling under one controlled rig, with age, gender and ethnicity recorded. It is the dataset every range, every noise estimate and the skin-age regression in this product are fitted and validated against — and its limits are the product's limits: a university subject pool, 18–54, sd 7.1 years, studio-lit rather than selfie-lit.

A 468-point facial surface mesh recovered from a single camera in real time on a phone. Every zone boundary, the head pose the guided scan coaches against, and the per-zone coverage that decides when a scan is finished are read off these landmarks — so its accuracy is the floor under all of them.

Automated lesion counting from facial photographs reached r = 0.90 against ground truth for INFLAMMATORY lesions but only r = 0.72 for non-inflammatory ones. The gap is the honest boundary of a blemish count from a photograph: an inflamed spot has colour contrast to find, a closed comedone largely does not.

CIELAB is the working colour space for quantitative skin measurement.

The 24-patch chart that cameras and colour software are checked against. Our colour measurement reproduces its published values to within less than the smallest difference a person can see, on the skin tones. That is a statement about being CORRECT rather than merely consistent — and consistency alone could never have shown it, because an error that repeats itself perfectly looks like a good measurement.

Faces photographed after 31 hours awake were rated less healthy and more tired than the same faces well rested. People read sleep off a face — which is why a scan taken after a bad night is a real reading of a real state, and why the app asks before it blames a product.

The REGULARITY of skin texture in an image correlates with transepidermal water loss — image texture carries real information about barrier function, which is what a dryness reading from a photograph rests on. Its limits are stated and they are ours: microscopic images, not phone cameras, and R² = 0.524 — a real signal, not a substitute for measuring TEWL. This is why dryness is reported as a visible-texture finding and never as hydration.

Psychological stress measurably perturbs epidermal permeability barrier homeostasis — a controlled finding, not folklore.

Interview stress, SLEEP DEPRIVATION and exercise each changed transepidermal water loss and barrier recovery after tape stripping in healthy women. Sleep is not a lifestyle aside here — it is a measured input to the same barrier our dryness metric looks at.

Automated pore size from standard facial photography correlated with visual grading at r = 0.86 — under CONTROLLED Visia-CR capture. Ours is a phone in a bathroom, and pores is both the least reliable metric in our calibration (ICC 0.65) and the one most limited by the camera rather than by the face (79%). The gap between that study and our number is the cost of the photograph being convenient.

Skin colour distribution drives judgements of health and attractiveness in female faces.

Reflectance separation of haemoglobin and melanin — the basis of the Mexameter and DermaSpectrometer, and of how erythema is separated from tone here.

Image-based erythema scoring agrees with clinical assessment.

The European standard for measuring skin COLOUR. It endorses the exact route taken here — CIE colorimetry, reflectance-based separation, image analysis by colour camera — and it states the limitation our own fairness work found independently: erythema is barely visible in deeply melanized skin. Finding our measured weakness already written into the standard is the strongest evidence that this is the right standard to be held to.

Validated photonumeric five-point scales for the lower face including the JAWLINE, with substantial to almost perfect interrater and high intrarater reliability — sagging along the jaw is a gradeable quantity from a photograph. MIRRA measures it and shows it, but keeps it out of the overall score: it has within-person stability and no agreed population scale across capture regimes.

Skin oiliness and shine can be quantified from images by separating the SPECULAR from the diffuse component with differential polarization, calibrated against twelve graded sebum applications and correlating strongly with clinical assessment. Specular separation is the same principle MIRRA uses for shine — and the reason shine is capped in the weights is that it measures a momentary state, not a property of skin.

Colour homogeneity of facial skin alone shifts perceived age by up to 20 years, independent of topography.

Combination timeline for brightening and anti-ageing actives.

The same effect holds in male faces — the cue is skin, not gender presentation.

Wrinkle count, length, width, area and relative depth extracted from photographs under raking light tracked clinical grading at r ≈ 0.8 across all parameters. Wrinkles are legible from an image because a wrinkle casts a shadow — which is also why the app cares so much about the direction of her light.

Dose–response and time course for retinol.

Measured timeline for visible change under retinol with vitamin C.