Why ±50K Matters More Than the Colour Temperature Itself

What ±50K actually means

A colour temperature on a datasheet is a target, not a measurement of the item in the box. Every LED that comes off a production line lands somewhere near that target rather than exactly on it.

±50K describes how near. It says that any given production run will sit within fifty kelvin either side of the nominal figure. It says nothing about what the nominal figure is — that is your decision, and you make it based on the space.

This is why the tolerance is the more useful number of the two. Two suppliers can quote you the same nominal colour temperature and deliver product that looks obviously different, because one holds ±50K and the other holds a spread several times wider and simply never mentions it.

Most datasheets state the target. Fewer state the tolerance. When a tolerance is absent, it is absent for a reason.

Why LEDs vary at all

An LED produces its light through a semiconductor process and, for white LEDs, a phosphor coating on top of it. Both stages vary slightly across a production wafer and across the phosphor deposition. The result is that a single batch produces LEDs across a small spread of colours rather than one identical colour.

Manufacturers deal with this by binning: measuring the finished LEDs and sorting them into groups. A tight bin is a narrow group. A loose bin is a wide one. Tight bins are more expensive because sorting more finely means more of the output falls outside the bin you are selling and has to be sold as something else.

That cost difference is where colour consistency is quietly lost. It never appears as a line item. It appears eighteen months later as a cove that does not match.

Why the eye is brutal about this

Human colour vision is poor in absolute terms and extremely good at comparison. Nobody walks into a room and identifies a colour temperature. Everybody walks into a room and sees that two adjacent lengths of the same detail are not the same.

That is the entire problem. Colour differences that are invisible in isolation become obvious the moment two pieces sit side by side in the same plane, lighting the same surface, meeting at a joint.

A continuous architectural detail is the worst possible test conditions you could design. It puts the comparison directly in front of the viewer with no visual break, at exactly the scale the eye is best at judging.

SDCM, in plain terms

SDCM stands for Standard Deviation of Colour Matching, and it is measured in steps of a MacAdam ellipse. The underlying idea is simple: draw a region on the colour chart within which a human observer cannot reliably tell two colours apart. That region is one step.

The scale runs roughly as follows.

SDCM 1 — differences are below the threshold of perception for essentially all observers.

SDCM 2 — imperceptible to most people in most conditions. This is the standard for architectural work where continuity matters.

SDCM 3 — perceptible on close side-by-side comparison. Acceptable in many general applications, and a common commercial-grade figure.

SDCM 4 and above — visible to an ordinary person without prompting. Fine for an industrial warehouse. Not fine for a hotel lobby.

The step is a boundary of perceptibility, not a linear unit of difference, so the jump from 2 to 4 is far more visible than the number suggests.

Together, SDCM and the ±50K tolerance describe two different things. SDCM covers the spread across the whole colour chart, including the green–magenta axis that colour temperature alone does not capture. ±50K covers drift along the warm-to-cool axis. A product can be tight on one and loose on the other, which is why both belong in a specification.

Ra answers a different question, and both belong in the spec

Colour rendering and colour consistency get written into the same line of a specification and they are not the same thing.

Ra measures how honestly the light renders the surfaces it falls on. It compares the appearance of a set of reference colours under the light against their appearance under a reference source, and scores the result. Low Ra flattens materials — timber goes grey, fabric loses its depth, skin looks unwell. It is a property of a single piece of light, measured on its own.

SDCM and ±50K measure whether two pieces of light agree with each other. They say nothing about how well either one renders a material. A product could render beautifully and still be inconsistent from reel to reel.

The distinction matters because the two failures look completely different on site and get diagnosed wrongly. A cove that makes the stone look dead is a rendering problem, and no amount of batch control fixes it. A cove where one half looks warmer than the other is a consistency problem, and a high Ra figure is irrelevant to it.

We hold Ra ≥ 90, which is the level at which materials read as themselves in an architectural interior. Below the mid-eighties, timber and stone start to lose their character, and in retail and hospitality that shows up directly in how the product on the shelf looks.

Specify all three. They cost nothing to write and they are the difference between a specification that constrains a supplier and one that only sounds like it does.

The reorder problem

Colour consistency within a single delivery is the easy half. Any competent supplier can send you material from one production run that matches itself.

The problem arrives later.

A tenancy is extended. A damaged section is replaced. Phase two is built. Six months have passed, a new production run has happened, and the new material sits directly against the old.

If the product is held to a tight tolerance, the new run lands in the same place as the old one, and the join is invisible. If it is not, the join is the first thing anybody sees.

This is the failure mode that generates the two complaints we hear most often, usually from the same project:

One person says the new section is too warm. Another says it is too cool.

Both of them are right. They are looking at the same join from different angles, and the eye judges by comparison.

Nothing here is a defect. Every metre performs to its datasheet. The specification simply never pinned down the one number that mattered.

Where it shows up worst

Continuous coves and slots. The join has no visual break to hide behind.

Mitred and internal corners. Two lengths meet at an angle, lighting the same surface from two directions. Any difference is doubled.

Glossy and mirrored surfaces. A polished stone or mirrored reveal shows the source directly rather than the reflected wash, so nothing is diffused away.

White and near-white surfaces. A white wall is a measuring instrument for colour temperature. Warm timber and dark stone are far more forgiving.

Joins between reels within one order. Even in a single delivery, plan where reel joins land. Put them at a corner or a break in the detail rather than in the middle of a clean run.

What to specify

Write the tolerances, not just the target. Five lines are enough:

• Nominal colour temperature — your design decision
• Colour rendering — state a minimum Ra
• Colour temperature tolerance — state a number, in kelvin
• SDCM — state a maximum step count
• Batch consistency for reorders — state that later deliveries must hold the same tolerances against the original

That last line is the one everybody omits, and it is the one that decides whether phase two matches phase one.

If a supplier cannot answer those five points from a datasheet without going away to ask a factory, that is your answer about how tightly the product is controlled.

What we hold

Our LED strip is supplied to Ra ≥ 90, ±50K and SDCM ≤ 2, and those figures apply to reorders, not only to the first delivery. They are published rather than quoted, alongside the per-metre rates, which is the point of the whole exercise: the number is fixed before you ask, so there is nothing to negotiate and nothing to discover later.

Rates are HKD 72 per metre for Bare LED Strip, HKD 96 per metre for COB LED Strip and HKD 126 per metre for Neon Flex 1010. No minimum order, no quotation process, and the figures above do not change with the size of the order.