Why Two Lamps at the Same Color Temperature Look Different

Why this matters

Three fixtures get replaced under warranty and the corridor now has a stripe in it. Everything shipped to specification, every label matches, and the customer is standing there pointing at it. This is one of the few lighting complaints where the shop is genuinely exposed, because the mismatch is permanent, visible from the door, and cannot be adjusted out. It is also almost entirely preventable with one line on the purchase order, provided you know which of four causes you are actually looking at before you order anything.

The call

An office corridor, eighteen recessed downlights, nominal 3500K, installed as one lot. Six months ago three failed and were replaced under warranty with the same catalog number. The three replacements read cooler and slightly pink against the fifteen originals. The customer wants the three changed again. The shop wants to know whether changing them will fix anything.

Readings were taken from floor level with a handheld spectrometer aimed at the aperture, which avoids opening an energized housing entirely; where a label had to be confirmed, the branch circuit was opened at the panel and locked out under 29 CFR 1910.333(b)(2), the fixture proved dead using the live-dead-live sequence in NFPA 70E-2021, 120.5, in the edition the employer's electrical safety program adopts, and the housing reached from a fitted platform rather than the top step of a stepladder.

Eliminated first, because they are cheap and reversible

Dimming level. Some drivers and modules shift chromaticity as they dim, and some products do it deliberately to imitate an incandescent dimming curve. If two groups of fixtures sit at different output levels, the mismatch is a control problem and it costs nothing to fix. Checked at the control station: one zone, all eighteen at full output. Ruled out.

Ordering code. A family often carries several color temperatures and rendering grades under near-identical numbers, and a warranty department substituting the closest available code is a routine event. Labels on two replacements and two originals matched field for field. Ruled out.

Those two account for a large share of complaints of this shape, and both are answered before anyone climbs anything twice.

The measurement that separated the remaining two causes

The two survivors are age (the originals have run about 6,000 hours, the replacements about 1,100, and chromaticity drifts with operating hours) and production lot (two units of the same code from different lots can sit in different places inside the same nominal color temperature region).

They look identical from the corridor. They separate on one measurement: how tightly the fifteen originals agree with each other.

The originals spread 12 K and 0.0003 in Duv across the length of the corridor. The gap from the originals to the replacements is 65 K and 0.0040 in Duv, and the Duv gap is 13 times the spread among the originals.

That ordering is the finding. If age were driving it, fifteen units with the same hours would have drifted together, which they have, and the replacements would sit near where the originals began. Pulling the family's published color maintenance data for the originals' operating hours bounded the expected chromaticity shift at 0.0010 in Duv. So age can account for at most 0.0010 of the observed 0.0040, which is 25 percent, and the remaining 0.0030, or 75 percent, is production spread between two lots.

Both groups are correctly labeled 3500K. The nominal window for that designation runs roughly 3,220 K to 3,710 K, a span of 490 K, and the 65 K difference between the two groups uses 13 percent of it. The complaint is on the tint axis, not the warm-to-cool axis, and the color temperature card owns what that axis is and why the label discards it.

Why the specification allowed this

Two different tolerance schemes get used on the same product and they are not the same size.

• A nominal color temperature designation names a large four-sided region of the chromaticity diagram. Anything inside it ships as that color temperature. This is what a specification reading only "3500K" buys you.
• A chromaticity tolerance in steps names a much smaller ellipse around a target point, sized in multiples of a standard deviation of color matching. One step is roughly a just-noticeable difference under ideal side-by-side viewing, and architectural specifications commonly call three steps or tighter for luminaires that share a field of view. How many steps yours requires is a number that belongs in your specification document, not to any default.

There is a trap inside the second scheme worth stating explicitly: a step tolerance bounds each unit's distance from a target, not from another unit. Two units both sitting legitimately inside a 3-step ellipse can be on opposite edges of it, roughly six steps apart from each other. If unit-to-unit match is what the job needs, the specification has to say that, not just name a step count.

What was done, and the arithmetic behind the choice

Three options, compared in fixture count and lift setups rather than price.

1. Re-order three replacements bin-matched to the original lot. Three units, one lift setup, and it depends entirely on whether the manufacturer can still supply from a matched bin six months on. They could not.
2. Replace all eighteen from a single current lot. Eighteen units, one lift setup, and it guarantees the result because everything in the corridor comes from one production run. Six times the product for the same labor mobilization.
3. Relocate the three mismatched units to a back-of-house corridor with no visual adjacency and install three current-lot units in the office corridor. Six fixture swaps, one lift setup, three new units. Leaves a mismatch nobody sees, provided the receiving space genuinely has no shared sight line, which has to be walked rather than assumed.

Option three was taken because the receiving corridor was a single-fixture-type space with no adjacency. Had it not been, option two is the only one that ends the complaint, and its cost is the entire lesson: an unbinned warranty replacement can convert a three-unit failure into an eighteen-unit re-lamp.

What to specify so this does not recur

On any job where two luminaires will be seen at once, add three things to the order, all of which a manufacturer can answer:

• A chromaticity tolerance in steps, with the target point named.
• A statement that the tolerance applies unit to unit, not only to the target.
• A binning commitment for warranty and spare units, so replacements come from a matched bin for a stated period.

Then add the operational half, which costs nothing at install time and everything later: buy spare units from the installed lot on the original order and store them on site. A spare pulled from the same production run is a guaranteed match, and it converts every future single-fixture failure from a chromaticity negotiation into a swap. Two or three per luminaire type on any job with more than a dozen units in one field of view is the usual call, and the storage location goes in the closeout documents so the next person can find them.

Rendering is a separate specification with separate floors and belongs to the color rendering card. Chromaticity tolerance does not constrain it, and a tight color match tells you nothing about how a red object will look under either unit.

Checking your own figures

• Cheap causes eliminated before expensive ones, and both printed. Dim level: one zone, all eighteen at full. Ordering code: four labels checked, all matching.
• The within-group spread was measured before the between-group gap was interpreted. Originals: 12 K and 0.0003 Duv across three positions. Between groups: 65 K and 0.0040 Duv. Ratio 13 to 1.
• Age was bounded with a document, not with an opinion. Published color maintenance bound 0.0010 Duv at the originals' hours, against an observed 0.0040, so age is 25 percent of the gap and lot variation is 75 percent. The two add to the whole, so no third cause is required.
• Both groups checked against the nominal window rather than assumed compliant. Window 3,220 K to 3,710 K; measured 3,540, 3,545, 3,552 and 3,610 K. All inside, so the warranty department shipped correctly and there is no claim to make.
• The axis of the complaint matches the axis of the measurement. Duv separation 0.0040, symptom described as pink against the originals, and the replacements carry the negative Duv, which is the pink side. Direction stated, not assumed.
• The step-tolerance caveat carried its number. Two units inside a 3-step ellipse can sit up to 6 steps apart from each other, which is why the unit-to-unit clause is separate from the step count.
• No rendering figure entered this diagnosis, because no rendering figure would have changed it.

References

• ANSI C78.377, the chromaticity specification for solid-state lighting products, in the edition named by your specification or the product listing, which is what a nominal color temperature designation binds through
• Manufacturer chromaticity binning structure and color maintenance data for the specific family, which own the tolerance actually shipped and the expected shift with operating hours
• 29 CFR 1910.333(b)(2) for de-energizing before work on a fixture, and NFPA 70E-2021, 120.5, for the live-dead-live verification sequence, in the edition your employer's electrical safety program adopts
• See related: What Color Temperature Does and Does Not Tell You; What Color Rendering Actually Measures