Colour rendering is not one number: why CRI on its own is no longer enough

Light and visual comfort

Two fittings with the same CRI 90 can render red in two different ways. The difference shows on the fabrics, not on the data sheet.

What CRI is, and what it is calculated on

The colour rendering index compares how a source renders a set of colour samples against a reference source. The number that appears on data sheets is an average over eight pale samples — pastels, not full colours — and an average, like every average, hides the distribution. Two sources can carry the same value and get different colours wrong.

The point that matters to a specifier is this: none of those eight samples is a saturated red. A fitting can therefore declare ninety and render badly the very tint the room depends on — the red of a wall covering, the warmth of a wood, the skin of whoever is sitting at the table.

Why it is red that goes missing

It is no accident that the colour absent from the average is red. Many white sources are built from a blue emission converted by phosphors, and the long end of the spectrum — the part that gives reds back — is the most expensive to fill properly. Cutting it barely lowers the average over eight pastels and lowers the rendering of a full red enormously.

The result in use is recognisable without instruments: warm fabrics go flat, woods drift towards grey, and people look tired. Nobody says "the red is missing": they say the room feels sad, and that is a judgement no specification had allowed for.

R9: the extra line that changes a specification

That is why contract specifications have started asking for R9 alongside CRI — the rendering of saturated red, which does not enter the average. It is one line, and it moves the comparison between two quotations more than any other figure declared on the same page.

A low R9 shows at once on warm fabrics and woods, which in hospitality are everywhere: curtains, carpets, panelling, counters. Asking for it in writing costs a line; discovering it after installation costs an argument with the client, and the client will be right.

The newer metrics separate two things CRI mixes

For some years there have been metrics that describe rendering with two numbers instead of one: one says how faithful the source is to the reference, the other how much it saturates or damps colours against it. They are different things, and mixing them into a single index is precisely what makes CRI insufficient.

In practice they answer a question one number cannot handle: is this light unfaithful, or is it chosen?. A source that saturates slightly can be perfect in a dining room and wrong in a gallery, and with a single index the two cases score the same.

A badly written threshold hands freedom to the supplier

If the specification asks only for "CRI greater than 90", it is leaving the supplier free to deliver the worst light that meets that threshold — and that is not sharp practice: it is exactly what a threshold authorises.

The form that works is different: you declare the quantity, the minimum value and the space in which it applies. The rendering needed over a table is not the one needed in a service corridor, and asking for the same number everywhere pays twice — once in excess where it is not needed, and once short where more would have helped.

The sample beats the data sheet, and is judged on the real material

Rendering is judged on the real material, under the real source. A data sheet describes the source; it does not describe the meeting between that source and that fabric, which is the only thing the client will look at.

It is not needed everywhere: it is needed where material and light really act on each other — fabrics, woods, brushed brasses, marbles. A test made on one piece, mounted where it will go and lit with the final source, costs less than a replacement once the installation is finished, and closes the argument before it opens.

Consistency is worth as much as the value

A specification that asks for high rendering but says nothing about consistency has solved half the problem. Two fittings from the same batch have to render alike; without a written tolerance, the difference shows when they are mounted side by side — which is always.

It is the same reasoning as colour temperature, and the same trap: the average value is easy to declare, the spread is not. Asking for it is one line, and it matters most on supplies split into phases, where there is necessarily more than one batch.

High rendering has a price, and the specification must say where it is worth it

Filling the long end of the spectrum properly costs: at equal power drawn, a very high rendering source produces fewer lumens than an ordinary one. It is not a fault of the product, it is the physical price of fidelity — and it has to be allowed for, because it means more fittings or more watts to reach the same illuminance.

It follows that asking for maximum rendering everywhere is not prudence: it is waste in every space where nobody looks at a colour. A useful specification grades it: high rendering where there are materials, faces and food; ordinary rendering in stores, service routes and back-of-house. The difference between the two approaches shows on the bill for ten years.

Where rendering really changes somebody's work

Colour rendering is not a value needed everywhere to the same degree, and treating it that way is an expensive mistake. There are spaces where high rendering changes somebody's work: the kitchen, where the cooking of a piece of meat is judged; the reception desk, where a document is read; the linen store, where two tones are matched.

And there are spaces where it only changes the experience of whoever is looking: a lobby, a dining room, a guest room. They are two different categories and they get written differently — in the first rendering is a work requirement, in the second it is a design choice.

Confusing them leads to the two symmetrical mistakes seen most often: very high rendering in a store, where nobody gains from it, and ordinary rendering in a kitchen, where somebody works eight hours a day and cannot judge what they are doing.

Daylight is always the benchmark

The client does not judge a fabric against a chart: they judge it against how they saw it in the shop or at the window. A practical rule follows that settles many arguments: if a space has daylight, the artificial source will be compared with it, every day.

In practice that means a room with large windows is more demanding on rendering than an internal room of the same function. And it is also why a test made in the evening in a windowless room can give a result that does not hold at eleven the next morning.

Tolerance between batches, written or unwritten

Sources are produced in groups, and inside each group there is a spread. On an isolated fitting nobody notices; along a row, or between the room refurbished last year and the one refurbished now, it shows at once.

Asking for a declared tolerance is one line, and it has an immediate practical effect: it obliges the supplier to select within a group rather than take what they have. Without that line the supplier delivers what they have, and they are right — because nobody asked for anything else.

Rendering is not forever

A source ages, and it does not only age by losing output: it can also shift its rendering. It is a slow, uniform and therefore invisible phenomenon, exactly like the fall in output: it only shows when a fitting is replaced and put next to the others.

A specification consequence follows that almost nobody writes: the useful guarantee is not only on remaining output, but also on colour stability over time. Asking for it is one line; discovering in four years that a refurbished room does not match the others is a problem with no economical solution.

How to look at a sample, in practice

A useful test has four elements and takes half an hour: the real material, the final source, the real distance, and the client present. If one of the four is missing it is not a test, it is a demonstration.

The fourth is the one most often skipped, and the most important. A photograph of the mock-up is an image again, and it reopens exactly the discussion the mock-up was meant to close. Whoever decides has to see it with their own eyes, and preferably sign a sheet saying they did.

What to do when two quotations declare the same number

It happens often, and it is the moment when a specification shows whether it was written well. If both declare CRI 90 and nothing else, there is no way to separate them on this quantity, and the decision slides onto price — which is exactly what a badly written threshold produces.

If instead the specification asks for R9, tolerance and a material test, the two quotations separate at once: one brings the numbers and the other says "compliant". "Compliant" is not an answer, and asking for the numbers explicitly at tender stage is the only way not to discover it once the installation is up.

How to write it, so that two quotations become comparable

The lines that make two quotations comparable on colour rendering are five, and none of them names a product:

• the minimum CRI and the minimum R9, declared separately
• the space in which each value applies, not one number for the whole building
• the tolerance between fittings of one batch and between different batches
• which materials the test will be made on, and with which source
• who attends the test, and what happens if the sample does not convince

• A bespoke piece
• The properties we have lit
• Free lighting calculation
• Lighting knowledge