What CRI and Kelvin actually measure

Both terms get used as if they were quality scores. Neither is.

CRI — colour rendering index. A score out of 100 describing how faithfully a light source reproduces a set of test colour samples compared with a reference source of the same colour temperature. The general form, Ra, averages eight samples, R1 through R8.

Kelvin — correlated colour temperature. The colour a black-body radiator would have at that temperature, in kelvin. 2700 K is warm and amber; 4000 K is neutral; 6000 K is cool. It says nothing about brightness, nothing about colour quality, and nothing about whether the light is any good.

What neither number describes is a face. CRI is measured on colour patches inside an integrating sphere. Kelvin is derived from the spectrum. A human face is a curved, red-dominant, three-dimensional surface lit at a glancing angle, and the thing that decides whether it looks healthy in a mirror is how much light arrives on a vertical plane at that surface — not how much light leaves the fitting.

That gap is where most mirror specifications fail. Not because the numbers were wrong, but because the numbers were never about the thing being judged.

The colour CRI leaves out

Ra averages R1 to R8: eight pastel, low-saturation samples — light skin, sky blue, leaf green, muted red, yellow, purple, two blues. Saturated red, R9, is deliberately excluded from that average. The original reason was practical: early fluorescent sources scored badly on R9, and leaving it out produced a flattering number.

Human skin is a red-dominant surface. Blood under a thin scattering layer. Leave red out of the score and you have excluded precisely the part of the spectrum that decides whether the person in the mirror looks well.

The practical symptom is easy to describe and impossible to unsee once you know it. A panel claiming CRI 95 with R9 of 10 renders blue and green beautifully and turns lips greyish-brown, and it does it with a straight face on the datasheet. On a hotel vanity, in front of a guest who has just flown eleven hours, that reads as tired.

Two ways to close the hole. Ask for R9 separately, at 50 or above, and 90 or above for makeup zones. Or move to TM-30, which reports Rf (fidelity, comparable in intent to Ra but computed differently) and Rg (gamut), and does include saturated reds. Either works. What does not work is accepting an Ra figure that has never been asked what it left out.

Which Kelvin, and why 3000 K keeps winning

3000 K dominates hospitality bathrooms for a reason that has nothing to do with fashion. At 3000 K the spectral balance is close enough to incandescent that skin keeps its warmth, and close enough to neutral that white tiles still read as white rather than cream. It is the temperature that offends the fewest people in the fewest rooms.

4000 K is the correct answer in a narrower set of cases, and it is worth being precise about which: grooming stations where makeup is applied and checked, clinical and medical-adjacent facilities, and North American projects where the whole floor is specified at 3500–4000 K and a 3000 K mirror would be the odd item out. In those rooms 4000 K is not harsh — it is consistent.

Consistency is the actual rule. A 4000 K mirror in a 2700 K room does not read as modern. It reads as a mistake, and usually as "the light is too bright", because the eye attributes the mismatch to intensity rather than to colour.

Two details worth writing into the schedule. First, give the tolerance: a nominal 3000 K that ships anywhere between 2800 and 3200 K will produce rooms that photograph differently from each other. Second, name the binning — see the section below — because tolerance on the datasheet and visible difference on the wall are two different claims.

Why light position beats light quantity

Here is the part that no datasheet covers, and it decides more outcomes than CRI and Kelvin combined.

A mirror is a vertical plane. The face in front of it is also roughly vertical. Light arriving from directly above the head grazes down the forehead and casts the brow ridge and the nose into shadow — the eye-socket look that makes every hotel bathroom photograph resemble a passport photo. Light arriving from behind the glass puts the face in silhouette and adds a bright halo the camera loves and the guest cannot use.

Light arriving from the sides, or from a diffuser running down the front face of the mirror, arrives at a near-horizontal angle to the face. That is the geometry that fills in under the eyes and under the chin. It is also the geometry that most mirror designs do not have, because a backlit panel is cheaper to make and photographs better in a catalogue.

One consequence worth carrying into the drawings: if the mirror is the light source, the mirror is now an electrical item, and its position is fixed by the lighting design rather than by the tile set-out. That reordering happens late on most projects, which is why the mirror ends up centred on the wall and the light ends up wherever the ceiling grid allowed.

It also means the mirror position and the vanity height have to be dimensioned from the same floor datum. If the vanity grows 60 mm at handover, the mirror has already been drilled at the height the old figure implied.

How much light, measured where

Bathroom lighting is normally designed to a horizontal illuminance target on a working plane. That is the right approach for a floor and the wrong one for a face.

European practice for interiors is set out in EN 12464-1, which gives a maintained illuminance for bathrooms and washrooms of the order of 200 lx on the relevant task area Standard. Read that as a floor, not a target: it is the level below which a room is considered badly lit, and it was written for a room, not for a grooming task.

For a face, the useful working figure is 300–500 lx measured vertically, at head height, at the standing position Estimate — verify. The range is wide because the task varies: shaving tolerates the bottom of it, applying makeup and checking colour needs the top of it and then some.

The measurement is easy and almost never done. Put a lux meter, or a phone with a lux app, flat on the countertop. Note the number. Now hold it vertically at 1.6 m, facing the mirror, at the position where a person's face would be. Note that number. In a typical ceiling-lit bathroom the second figure is a fraction of the first — often a third or less, because the light is arriving at a steep angle and vertical illuminance falls off with the cosine of that angle fast.

This is also the honest way to evaluate a backlit mirror. The halo around the glass can be spectacular and the vertical lux at the face can be under 100. The guest then stands closer to the mirror, which increases the angle, which makes it worse.

Whether a mirror may sit in that position at all is a separate question with its own answer — the zone rules and the IP ratings that follow from them are set out in What IP Rating Does a Bathroom Mirror Need? Light quality and ingress protection are decided independently, and both have to pass.

Flicker, and why the phone camera notices first

LEDs do not flicker because of the LED. They flicker because the driver chops the mains and the output is not smoothed enough, and because dimming is often implemented by switching the output on and off faster than the eye can resolve.

At high enough frequency this is invisible to a person standing at the mirror and completely obvious to a phone camera, which samples at a different rate and records dark bands rolling across the image. Guests photograph bathrooms. A mirror that produces banding in every photograph reads as a fault, and no amount of explaining about driver topology will change that.

The recommended practice most often cited is IEEE 1789, which sets modulation limits against frequency and introduces a stroboscopic measure Standard. It is a recommended practice rather than a regulation, which means your supplier can comply with the electrical safety standards and still not comply with this.

The specification that actually works is short: constant-current driver, output ripple under about 5%, and no PWM dimming below roughly 1.25 kHz — or dimming by amplitude rather than by chopping Estimate — verify. Ask for the driver datasheet, not the fitting datasheet. The fitting sheet will say "flicker free" in a marketing box; the driver sheet has a ripple figure in it.

There is a second, less discussed reason to care. Drivers are the component that fails. A mirror is usually removed from a wall to replace a driver, and the failure is frequently attributed to the mirror rather than to the €4 component behind it. Specify the driver as a replaceable item with a part number, and put a spare in the O&M pack.

Two mirrors in one room

A nominal colour temperature is a range, not a point. LEDs are sorted after production into bins by measured colour point, and two fittings binned at opposite ends of the same nominal 3000 K can be visibly different when mounted side by side.

The unit is the MacAdam ellipse, expressed as SDCM — standard deviation of colour matching. One step is roughly the threshold at which a trained eye notices a difference on adjacent surfaces; three steps is where an untrained eye notices.

So: specify SDCM ≤ 3, and ≤ 2 for any room with more than one mirror Standard. Then add the sentence that actually enforces it: all mirrors in a single room to be supplied from one production batch, with the batch reference on the packing list.

Twin-basin bathrooms are where this bites. Two mirrors, same order, same model number, shipped three weeks apart because the second was damaged in transit and remade. Nothing on either datasheet is wrong. The room still looks wrong, and the only remedy is replacing the pair.

The same logic governs the frame finish, which is why PVD, powder coat and chrome cannot be matched from a later production run either. Colour is a batch property in both cases, and both need a retained physical sample rather than a photograph.

How to write the requirement

The clause below is deliberately short. Every line exists because its absence has caused a replacement.

The one-line test. Ask the supplier: "What is R9, and what is the vertical lux at 1.6 m?" If both answers arrive with a number and a test reference, the rest of the datasheet is probably trustworthy. If either answer is a paragraph about quality, the datasheet is decoration.

How we build and document it

We make LED mirrors and mirror cabinets, which means the light source, the diffuser and the glass are one assembly rather than three items meeting on a wall.

• 3000 K and 4000 K as standard, with R9 reported. The R9 figure goes on the batch documentation, not only on the catalogue page, because a nominal CRI can shift between LED reels. Bergoto detail — verify
• Binning declared per order. Rooms with more than one mirror are supplied from one batch, and the batch reference travels on the packing list so a replacement can be matched later.
• Front-facing diffuser, not edge halo. Our illuminated mirrors are built with the light band on the face of the unit rather than escaping around the back, because a halo lights the wall and a front band lights the person.
• Driver as a line item. Replaceable from the front, with the part number in the documentation and a spare in the O&M pack for project orders.

Bergoto is an OEM and ODM manufacturer of bathroom and kitchen products, operating to ISO 9001 quality management and BSCI social compliance, with around thirty years in the industry, ten production lines and annual capacity of roughly 500,000 pieces. Fifteen designers work on product development and project documentation in-house. What that means on a mirror package is that the photometric data, the driver specification and the drawings come from the same team that builds the unit, rather than from three suppliers who have never met.

If you are specifying mirrors for a project and have the room layout, send it with the ceiling lighting plan. We will tell you where the vertical illuminance will land before anyone orders glass — including on the layouts where the answer is that the ceiling layout needs to change first.

About these figures

Every number above is either a value from a named standard or a field figure that needs your own measurement. The distinction is stated so you can defend each one in a meeting.

• Standard — CRI, Ra and the R1–R8 sample set are defined in CIE 13.3; the exclusion of R9 from the Ra average follows from that definition. TM-30 Rf and Rg are defined by IES TM-30. Check the edition current in your market before quoting a clause number.
• Standard — the 200 lx figure for bathrooms and washrooms is a maintained illuminance in EN 12464-1. Editions differ, and the task area definition matters. Read the table in the current text.
• Standard — SDCM and the MacAdam ellipse are colour-matching definitions; the ≤ 3 recommendation is industry practice rather than a legal limit. IEEE 1789 is a recommended practice for flicker, not a regulation.
• Estimate — verify — the 300–500 lx vertical range, the 5% ripple figure and the 1.25 kHz PWM threshold are engineering indications drawn from practice. Measure the illuminance on your own completed installation; ask the driver manufacturer for the ripple figure rather than taking ours.
• Bergoto detail — verify — R9 reporting, batch binning and front-band construction describe our standard practice. Confirm against the specification sheet issued for a specific model before quoting them in a contract.

Nothing here replaces a lighting design. The illuminance on a face depends on room geometry, ceiling height, reflectance and the interaction of every fitting in the room — which is why the last word belongs to a calculation, not to a datasheet.

Frequently asked questions