What Color Temperature Does and Does Not Tell You

Why this matters

Correlated color temperature is the only lighting number most customers know by heart, and it is the number specifications lean on hardest, because it is easy to write and easy to argue about. It is also a one-dimensional label sitting on a two-dimensional problem. A job can be delivered with every luminaire correctly labeled the same color temperature and still look wrong from the doorway, and when it does, the specification gives you nothing to push back with, because everything shipped exactly as written.

What the number actually is, and why "correlated" is load-bearing

Heat a theoretical blackbody radiator and it glows, shifting through red, orange, white and finally blue as its temperature climbs. Plot those colors on a chromaticity diagram and they trace a curve, the Planckian locus. A source whose color sits on that curve can be labeled with the radiator temperature that matches it, in kelvin, and that is a color temperature.

Almost no practical light source sits on the curve. Fluorescent, LED and metal halide sources sit near it, off to one side. Correlated color temperature is the temperature of the point on the locus that is closest to where the source actually sits. It is a projection. It throws away the perpendicular distance, and that discarded distance has a name: Duv, the signed offset from the locus. Positive Duv sits above the locus and reads green; negative Duv sits below it and reads pink or magenta.

So a CCT label answers one question, where along the locus, and refuses the other, how far off it. Two sources with identical CCT can sit on opposite sides of the curve, and side by side they will not look like the same light.

A nominal CCT on a product sheet is looser still. It names a region of the chromaticity diagram, not a point. In the nominal-CCT scheme most manufacturers publish against, 4000K nominal is centered near 3,985 K with a tolerance of roughly plus or minus 275 K, so a product legitimately labeled 4000K may measure anywhere from about 3,710 K to about 4,260 K. That scheme binds through your specification, a rebate program's rules or a product listing, in the edition each of those names, and never on its own.

The intuition runs backwards, and it is worth saying out loud

Higher color temperature looks cooler. 5000K reads blue-white and 2700K reads orange-white, which is the opposite of how anyone talks about hot and cold. The reason is that the number is the temperature of the radiator, not of the room: a hotter blackbody emits proportionally more short-wavelength energy. Customers get this backwards constantly, and correcting it in the first thirty seconds of a lamp conversation saves the whole rest of it.

What the label does tell you

It tells you the warm-to-cool appearance of the light and, by extension, how white surfaces in the room will read. That is a real and useful thing, and it is the axis people have vocabulary for, which is why it survives as the headline. Within a single product family it also tends to track efficacy, with higher color temperatures publishing more lumens per watt than lower ones at the same drive, but that is a family tendency to verify on the test report rather than a property of the number.

Four things the label leaves entirely free

1. Spectral content. CCT is computed from three chromaticity values, and an enormous number of different spectral distributions collapse to the same three. Two sources at 3500K can put out light built from completely different wavelength mixes.
2. Color rendering. Nothing in a CCT figure constrains how objects will look under it. The rendering card owns what that is measured with and what the metric hides.
3. Tint. The Duv axis, discarded by the projection, and the axis where most delivered-job complaints actually live.
4. Effectiveness at low light levels. At the light levels of parking areas and outdoor walkways the eye's spectral sensitivity shifts, so two sources delivering equal photopic illuminance are not equally effective for peripheral detection. The multipliers that describe that come from the recommended practice covering mesopic conditions, in the edition your specification names, not from the CCT label.

Worked example: three luminaires, all correctly 4000K

A retail fit-out specified "4000K throughout." Three luminaire types from three manufacturers. On the walkthrough the wall wash reads visibly green against the downlights, and the track heads look fine. Nobody shipped the wrong product.

Measured values, taken from each manufacturer's test report for the ordering codes supplied rather than by opening an energized housing:

The axis the specification controlled. The permitted nominal window runs 3,710 K to 4,260 K, a span of 550 K. The delivered spread is 4,060 - 3,920 = 140 K, which is 25 percent of the permitted window. The three products are tightly grouped on the only axis anyone wrote down.

The axis the specification did not control. Duv runs from -0.001 to +0.005, a spread of 0.006. The wall wash sits 0.006 above the downlight, on the green side of the locus, and the downlight sits marginally below it. That is the entire complaint. Nothing about it is a defect, a failure, or a wrong shipment.

What the numbers rule out. The complaint cannot be a color temperature problem, because a 140 K spread across three products is inside the tolerance of a single product line and well below what a person distinguishes on the warm-to-cool axis in a lit room. It cannot be a rendering problem either, unless the rendering figures also differ, which is a separate question with a separate metric and a separate card. The one measured quantity that varies materially is Duv, and the one visible symptom is a tint difference, and those two point the same way.

What flips this. If the three had measured 3,750 K, 4,000 K and 4,240 K, all still inside the nominal window, the spread would be 490 K, or 89 percent of the permitted window, and the warm-to-cool difference would be doing real work in the complaint. Same specification, same compliance, completely different diagnosis. So run the two spreads before deciding which axis you are arguing about.

What should actually drive the choice

Since the label constrains less than people assume, the selection is not a technical optimization and should not be sold as one. Three things carry it, in this order for most jobs.

Adjacency. The eye adapts to whatever it has been looking at, so an absolute color temperature matters far less than a difference across a threshold. A space that reads perfectly on its own will read yellow if the corridor feeding it is two steps cooler. Walk the route the occupant walks before picking anything, and match across the openings that are in one field of view.

Surface finishes. Warm woods and warm-toned paint go muddy under high color temperatures; cool greys and white laminate go clinical under low ones. This is the interaction customers notice and cannot name, and it is settled by a mock-up in the actual space with the actual finishes, not by a number on a page.

The occupant's expectation for that building type. Residential-feeling spaces skew warm, clinical and inspection spaces skew cool, and fighting the convention costs an argument you will not win.

There is a long-standing rule of thumb that warm sources suit low illuminances and cool sources suit high ones. Its evidence base is weaker than its popularity, so treat it as a reason to build a mock-up rather than as a specification basis, and let the mock-up settle it.

What to write instead

A CCT figure alone is not a specification for a job with more than one luminaire type in the same field of view. Add two things, both of which the manufacturer can answer from test data:

• A Duv limit, so tint is bounded rather than left free.
• A chromaticity tolerance stated in the step notation your specification uses, which bounds how far any two delivered units may sit from each other rather than from a nominal point. How tight that needs to be, and how it differs from a nominal-CCT quadrangle, belongs to the card on why two lamps at the same color temperature look different.

Both of those are ordinary things to ask for on a commercial order. Neither costs anything to write, and asking after installation costs a re-lamp.

Checking your own figures

• Compliance checked against the stated window, not asserted. Window 3,710 K to 4,260 K. Delivered values 3,920, 3,985 and 4,060 K. All three inside, so all three ship correctly.
• Both spreads printed before a cause was named. CCT spread 140 K against a 550 K window, 25 percent. Duv spread 0.006. The larger relative variation is on the uncontrolled axis, which is what makes the diagnosis rather than intuition.
• The count matches the list. Four things the label leaves free were promised and four are enumerated: spectral content, rendering, tint, and low-light effectiveness.
• Direction of the tint stated, not implied. Positive Duv is above the locus and reads green. The wall wash at +0.005 is the green one, and it is the one that drew the complaint.
• Nothing borrowed from a sibling card was re-derived. Rendering is named and handed off in one clause; chromaticity step tolerance is named and handed off in one clause; both figures above are CCT and Duv only.
• The nominal scheme carried its adoption path. The 3,985 K center and plus or minus 275 K tolerance bind through the specification, rebate rule or listing that names them, in that document's edition.

References

• ANSI C78.377, the chromaticity specification for solid-state lighting products, in the edition named by your specification, rebate program or product listing, which is what makes a nominal CCT enforceable
• Manufacturer photometric and electrical test report for the exact ordering code, which is the only source for measured CCT and Duv
• IES recommended practice covering mesopic light levels, in the edition your specification names, for low-light spectral effectiveness
• See related: What Color Rendering Actually Measures; Why Two Lamps at the Same Color Temperature Look Different; How to Read a Luminaire Cut Sheet