Metrics & standards

CRI 90 is not a colour quality specification, and TM-30 explains why

What Ra actually averages, the specific colours it omits, why R9 gets quoted separately, and how TM-30's Rf, Rg and colour vector graphic answer the question a single number cannot — which colours shift, and in which direction.

Written by Long-Term New Media Limited · Last reviewed August 2026

Two luminaires both declare Ra 90. One makes a retail display of red produce look appetising; the other makes it look grey and slightly brown. Neither datasheet is lying. The number simply does not measure the thing that differs.

What Ra is an average of

The general colour rendering index, Ra, is computed by illuminating a set of standardised test colour samples with the source under test and with a reference illuminant of the same correlated colour temperature, then measuring how far each sample shifts. Ra is the arithmetic mean of the first eight of those samples.

Those eight are all moderate in saturation — muted pastels. The strongly saturated colours live in the extended set, R9 through R15, and none of them are in Ra. Two consequences follow directly:

• Averaging hides outliers. Eight numbers averaging 90 could be eight values of 90, or seven values of 95 and one of 55. The second source has a visible, specific defect and the same headline figure.
• Saturated colours are unmeasured. Deep red is the classic case, which is why R9 — strong red — is so often quoted separately. White LEDs made with a blue pump and a broad yellow phosphor characteristically have a deficit in the deep red part of the spectrum, and Ra is blind to it.

What Ra cannot express at all

A fidelity average answers one question: how far, on average, do colours move from the reference. It cannot tell you two things a specifier actually needs.

Which direction they move. A source that slightly increases saturation and one that slightly decreases it can score identically. Their perceptual effect is opposite: the first makes merchandise look vivid, the second makes it look washed out. Retail and food display live entirely in this distinction.

Which hues are affected. An average over eight samples cannot say that reds are fine and cyans are collapsing.

What TM-30 adds

IES TM-30 evaluates the source against 99 colour evaluation samples spanning real-world reflectance spectra with far better hue and saturation coverage, and reports several things instead of one.

The colour vector graphic is the part that changes conversations. It is a single picture showing, for each of sixteen hue bins, how the source moves colours relative to the reference. A specifier looking at two products with identical Rf can see immediately that one pushes reds outward and the other pulls them in.

Neither metric tells you which is better

This is where TM-30 is routinely misused. Higher Rf is not automatically better, and Rg above 100 is not automatically better either — the application decides.

The efficacy trade-off, stated honestly

Better colour rendering costs efficacy. Filling in the spectral regions that a blue-pump-plus-yellow-phosphor LED under-serves — deep red in particular — means adding emission where the eye's luminous efficiency function is low, so more radiant power buys fewer lumens.

The size of that penalty depends on the package technology and changes with every product generation, so quoting a percentage here would be obsolete within a year and specific to a manufacturer besides. The reliable approach is to ask a single supplier for the same fitting in two CRI grades and compare the lm/W on their own datasheets. That comparison is real, current, and free.

Two things that are not colour rendering

Correlated colour temperature is a separate axis. CCT says where the white sits between warm and cool. It says nothing about how well colours render. A 3000 K source and a 4000 K source can have identical Rf.

Duv — distance from the blackbody locus — is a third axis and it is the one that makes light look "wrong" in a way people cannot name. A source sitting noticeably above the locus appears green-tinted; below, pink. It is possible to have good Ra, correct CCT and still light a room that everyone dislikes, because Duv is off. It belongs in the specification alongside the others, and it appears in the note on colour consistency and SDCM.

What to write in the specification

1. R_a minimum and R9 minimum, both as hard requirements rather than scored ones.
2. Where colour appearance is commercially important, TM-30 R_f and R_g with the colour vector graphic supplied — most reputable manufacturers can now produce these on request.
3. CCT with its tolerance expressed in SDCM, plus a Duv limit.
4. Whether the figures are from the LED package datasheet or from the finished luminaire. Optics, diffusers and phosphor-loaded covers all shift the result, and the luminaire figure is the one that lights the room.