Colour quality: what CRI hides and TM-30 reveals

Ra 90 is not a specification. Duv, R9, SDCM and TM-30 are where colour is actually won or lost.

Ra 90 không phải là một spec. Duv, R9, SDCM và TM-30 mới là nơi quyết định màu sắc đẹp hay hỏng.

A datasheet that says “3000 K, Ra 90” has told you almost nothing. Two products meeting that description can render a client’s marble lobby as warm cream or as slightly sour green, and only one of them will survive the site walk.

CCT is a shorthand, not an address

Correlated colour temperature reduces a two-dimensional chromaticity to one number: the temperature of the blackbody radiator that looks most similar. Any point on a line perpendicular to the Planckian locus maps to the same CCT.

Which means two 3000 K sources can sit visibly apart — one slightly above the locus (green), one below (pink). The distance is Duv, and it is the number that decides whether your white walls look clean.

Positive Duv → green/yellow cast. Skin looks ill. Marble looks dirty.
Negative Duv → pink/magenta cast. Often flattering, sometimes cloying.
Architectural acceptance window: |Duv| ≤ 0.006, tighter for hospitality and healthcare.

Specify CCT and Duv, or specify a chromaticity target with an SDCM tolerance. Never CCT alone.

SDCM: the number that keeps a run of lights looking like one product

MacAdam ellipses describe how far chromaticity can drift before an average observer notices. Tolerance is quoted in steps.

Tolerance	Where it belongs
2 SDCM	Museum, high-end retail, visible continuous runs
3 SDCM	Default for any architectural project
5 SDCM	Acceptable only where fittings are never seen together
7 SDCM	Industrial and back-of-house only

A 5-SDCM run of coves on a white ceiling shows visible banding. This is one of the most common and most expensive post-completion complaints in the industry, and it is entirely avoidable at specification stage for a small cost premium.

Ra averages the colour shift of eight pastel test samples, R1–R8. Nothing saturated is in that average. A phosphor-converted LED can be optimised to score well on eight desaturated patches while rendering deep red as brown.

That is why R9 — strong red — is the number to demand alongside Ra:

R9 ≥ 50: minimum for anything with people in it
R9 ≥ 90: retail food, healthcare, hospitality, timber and textile-heavy interiors

Skin tone, raw meat, red wine, teak, brick and every warm textile in the building depend on R9. It is routinely the weakest figure on a datasheet, and it is routinely left off the spec.

TM-30: two numbers instead of one

IES TM-30 uses 99 real-world colour samples and reports two figures:

Rf — fidelity. How accurately colours match the reference. Analogous to Ra but far more robust. Target Rf ≥ 85.
Rg — gamut. Whether the source saturates or mutes colour overall. 100 is neutral, above 100 saturates, below 100 mutes. Target Rg 95–110 for general architectural work.

The pair matters because they are independent. A source can have Rf 92 and Rg 88 — accurate but visibly flat and lifeless. Retail often deliberately wants Rg 105–115 to make merchandise pop. Rf alone cannot express that intent; Ra cannot express it at all.

TM-30 also produces a colour vector graphic showing which hue bins are shifted and in which direction. When a client says “the reds look wrong”, that graphic is how you find out whether they are right.

A specification that actually protects you

Nominal 3000 K, Duv 0.000 ± 0.004, ≤3 SDCM initial and maintained, Ra ≥ 90, R9 ≥ 50, TM-30 Rf ≥ 85 and Rg 95–110. Chromaticity maintenance ≤ 3 SDCM over the warranty period. LM-80 data with TM-21 projection at the specified drive current and Tc to be provided.

Two clauses in there do most of the work and are almost always missing. “Maintained” SDCM stops a supplier delivering a tight initial bin that drifts apart within two years. And LM-80 data at the actual drive current and case temperature stops an L70 figure measured on a cool bench being applied to a fitting sealed in an insulated ceiling void at 65 °C.

The practical check