Colour temperature and CRI at home

What Kelvin actually measures#

Correlated colour temperature is the temperature, in kelvin, at which a theoretical black body radiator glows the same colour as the lamp. Heat iron and it goes dull red, orange, yellow-white, blue-white. The scale is literal for filaments, which are hot metal, and approximate for LEDs, which are not.

Two consequences that catch people out:

• Higher numbers are cooler in appearance. 2700K looks warm, 5000K looks cold. The physics runs opposite to the language.
• Equal Kelvin steps are not equal visual steps. Colour difference tracks reciprocal temperature, measured in mired (one million divided by kelvin). 2700K is 370 mired and 3000K is 333, a gap of 37; 5000K is 200 and 5300K is 189, a gap of 11. The same 300K step is roughly three times more visible at the warm end. That is why 2700K and 3000K lamps side by side look obviously mismatched while two daylight lamps 300K apart look fine, and why the domestic range is crowded into 2700 to 4000K.

The domestic band, room by room#

Space	Kelvin	CRI target	R9 target	Reasoning
Living room	2700K	90+	50+	Wood, wool, brass and skin
Bedroom	2700K, dim-to-warm	90+	50+	Evening only, warmth matters more than crispness
Dining room	2700K, dim-to-warm	90+	50+	Food and faces, both red-dependent
Hallway and stair	2700K	80+	not critical	Continuity with the rooms it connects
Kitchen, general	3000K	90+	50+	Crisper without breaking with the living space
Bathroom vanity	3000K	95 preferred	50+	Skin tone judgement, make-up, shaving
Home office	3500 to 4000K	90+	not critical	Alertness during working hours
Wardrobe and closet	4000K	90+	50+	Judging garment colour before you leave
Garage and workshop	4000 to 5000K	80+	not critical	Contrast on small parts
Candle-effect and decorative	1800 to 2200K	not applicable	not applicable	Atmosphere, near zero task value

2700K is the default because it is what incandescent lamps produced, and a century of domestic interiors was furnished under it: every warm timber, terracotta and brass finish in a normal house was chosen, consciously or not, to look right at that colour. 3000K gives a kitchen a little clarity without visibly departing from the room next door. Beyond 3500K you are lighting a workplace.

The mixed-Kelvin failure#

Human vision chromatically adapts. Sit in a 2700K room for two minutes and white paper looks white; sit in a 4000K room and it looks white there too. What the eye cannot do is adapt to two white points simultaneously, in one field of view. That is the whole mechanism: a single 3000K downlight in a run of 2700K downlights looks broken, because the adapted white point is set by the majority and the outlier is judged against it.

The usual causes: replacing one failed bulb with whatever was in the drawer; buying downlights and undercabinet strips from different brands; a decorative pendant with an integrated LED at a different Kelvin; a colour-selectable fixture whose switch was set differently on one unit. Check selector switches on every unit before the ceiling closes.

Dim-to-warm, and why it reads as incandescent#

A tungsten filament dimmed to 20 per cent does not just get darker; it gets cooler as metal, which means warmer as light, dropping from about 2700K to roughly 2000K. Every association most people have with "cosy" is that curve. A standard LED dimmed to 20 per cent stays at 2700K and simply gets dimmer, which reads as grey rather than intimate.

Dim-to-warm LEDs follow the tungsten curve, typically 2700K at full output falling to 1800K at the bottom. They cost more, need a compatible dimmer, and earn their keep in exactly three places: over a dining table, beside a bed, and in any visible decorative fixture. In a kitchen or bathroom they are wasted, because those rooms are rarely used at 10 per cent output.

Watch for the opposite defect: some LEDs shift blue as they dim, because the change in driver current alters the phosphor balance. Test a sample before committing to twenty fixtures.

CRI, R9, and the terracotta problem#

CRI (Ra) is the average of a light source's fidelity across eight test colours, R1 to R8. All eight are desaturated pastels: soft pink, soft yellow-green, soft blue, and so on. The saturated colours are numbered R9 to R14, with R9 being strong red, and none of them are included in the Ra average.

That single fact explains most of the disappointment people feel about "high CRI" LEDs. A lamp can post Ra 82 while scoring R9 at zero or negative. Saturated red is exactly the region occupied by terracotta, brick, mahogany, red wine, tomatoes and the haemoglobin under skin. Under a lamp with no red, a terracotta floor reads grey-brown, cherry wood reads flat, and everyone looks slightly ill.

Metric	What it covers	Good domestic target
Ra (CRI)	Average fidelity across 8 pastels, R1 to R8	90 or better
R9	Saturated red, excluded from Ra	above 50, above 80 is excellent
Rf (TM-30)	Average fidelity across 99 real-world samples	85 or better
Rg (TM-30)	Gamut: whether colours are muted or boosted	95 to 110
SDCM	Unit-to-unit colour consistency	3-step or tighter

TM-30 is the better modern answer. Its fidelity index Rf works like CRI but over 99 evenly distributed real-world samples rather than eight pastels, so a single catastrophic failure cannot hide inside the average. Its gamut index Rg describes something else: whether the source makes colours more saturated (above 100) or duller (below 100) than a reference. Rf 88 with Rg 104 looks slightly more vivid than daylight, which many people prefer in living spaces; Rg 92 looks faintly washed. CRI cannot express that distinction at all. Where TM-30 data exists, use it; where it does not, CRI 90 plus R9 above 50 is a reasonable proxy.

Flicker#

All mains-powered LEDs modulate. The question is how much and at what frequency.

Percent flicker (modulation depth) is the peak-to-trough variation as a percentage of the average. IEEE 1789-2015 recommends keeping modulation below roughly 0.08 x frequency for low risk and below 0.0333 x frequency for no observable effect. At 120 Hz that is about 10 per cent for low risk, 4 per cent for effectively invisible. In the EU, Regulation 2019/2020 limits the short-term flicker metric Pst LM and the stroboscopic visibility measure SVM.

Practical rules:

• Flicker invisible at full output often appears in the bottom 20 per cent of the dimming range. Test there.
• Pairing a driver with an untested dimmer is the leading cause of residential flicker, buzz and dropout. Use the manufacturer's compatibility list.
• Very cheap integrated fixtures often run at 100 per cent modulation at 100 or 120 Hz. Most people cannot consciously see it, but it shows on a phone camera in slow motion and is associated with eye strain over long exposure.
• Anything near rotating machinery should be low-flicker for the stroboscopic risk alone.

How colour temperature meets paint and daylight#

Paint is a subtractive filter. It shows only what the source contains, which is why a colour chosen in a shop under 4000K fluorescent can be unrecognisable at home under 2700K LED.

• North-facing rooms in the northern hemisphere receive indirect skylight at roughly 6500 to 7500K: cool, blue, steady. Cool greys turn steely; warm whites hold up well.
• South-facing rooms receive direct and reflected sun that runs warmer and swings through the day, pushing warm white paint towards yellow by late afternoon.
• After dark, orientation stops mattering. Every room is lit only by the lamps you chose.

That last point is worth acting on. The common advice to "use warm lamps in a north-facing room to compensate" is confused: the lamps are only on when the north light has gone. Choose lamp colour for the evening, and test paint in the room under both daylight and the lamps you will actually install, at the dimming level you will actually use.

Watch for metamerism: two finishes that match under daylight can diverge visibly under a narrow-spectrum LED, most often between a painted panel and a factory-finished cabinet door. A high Rf source reduces the risk without eliminating it. It is also why a bathroom vanity deserves the highest CRI in the house, a point developed in mirror and lighting heights.

Checking what you already own, in ten minutes#

1. Take one bulb out of every fixture and read the small print: Kelvin, CRI, lumens. Mismatched Kelvin in one room is the most common reason a room "feels wrong" despite adequate light.
2. Photograph the room with the white balance locked. The camera does not chromatically adapt, so it shows the mixing your eye is politely hiding.
3. Hold something strongly red next to each source: a terracotta pot, the back of your hand. If it goes grey-brown, that lamp has no R9.
4. Film each fixture in slow motion. Rolling bands mean high modulation flicker.
5. Dim everything to 10 per cent and look for colour shift, flicker and buzz. That is the state you use the room in most evenings.
6. Write the winning specification on a card and keep it with the spares: Kelvin, CRI, R9, lumens, beam angle, brand, series.

Colour is chosen after quantity is settled, so start with how many lumens a room needs, fix positions with recessed lighting spacing, then apply one Kelvin across every layer of the layered lighting plan.