The Percevra Guide to How Colors Change Under D65, 2700K, 4000K and Shade

The four lights

These are not arbitrary picks. D65 is the CIE's noon-daylight reference — the light hex codes implicitly assume. 2700K and 4000K bracket almost every artificial interior, and open shade covers the cool end of natural light. Together they span the range a real palette actually lives in.

Nine colors, four lights, one table

Each row is one color; each column is the unadapted stimulus — the raw color the surface reflects, the moment you walk into the room before your eyes adjust. The ΔE under each swatch is its OKLab distance from the daylight original (roughly 4 is a just-noticeable difference in context, 10 is comfortably distinct).

D65 is identity by construction — the engine treats daylight as a true no-op, so the first column is your original color, never a restyled version of it.

Under 2700K — the great rearrangement

Warm tungsten light is the strongest force on this page, and it does not shift every color equally — it rearranges the palette's internal order. Reds, oranges, and yellows are the safe zone: the lamp feeds them, they pick up a little chroma, and they move least. Everything else pays for it.

Blue is the casualty. A confident mid-blue loses most of its chroma and lands as a warm gray — not a different blue, but effectively no longer blue at all. Violet swings toward wine, its hue rotating some 60–70° warm. Greens and teals slide toward olive. And whites and near-neutrals take the full cast: with no chroma of their own to defend, they inherit the lamp's amber wholesale, which is why they post the largest ΔE in the table above.

• Violet · ΔE 14.7Biggest mover
• Yellow · ΔE 5.6Steadiest
• ΔE 10.2Mean shift, nine swatches

Practical reading: under warm light, palettes that depend on a blue-versus-neutral or blue-versus-violet distinction are the ones that quietly merge. Warm-anchored palettes barely notice the lamp.

Under 4000K — the same story at half volume

Neutral-white LED sits between the bulb and the sky, and its effect is exactly that: the same warm pull as 2700K at roughly half the magnitude. Blues still lose chroma — dimmed rather than erased. Violet drifts toward plum instead of swinging to wine. Whites go cream rather than amber.

The trap with 4000K is that it reads as white in isolation. You only see the cast when a daylight reference sits next to it — which is precisely what the matrix above provides, and what a store display or an evening room never does.

• White · ΔE 6.4Biggest mover
• Red · ΔE 3.0Steadiest
• ΔE 4.7Mean shift, nine swatches

Under blue-sky shade — the quiet inversion

Shade lit only by blue sky runs the whole mechanism in reverse, gently. Cool skylight at around 10000K gives blues and violets a little more chroma — the only light here that flatters them — while warm colors dull slightly. The most interesting move is in the neutrals: a warm greige loses its warmth and lands as a plain gray, and white picks up a faint blue cast.

The 10000K figure is a declared proxy for shade illuminated only by clear sky (real measurements run ~9000–12000K), reconstructed on the CIE daylight locus — an estimate, not a measured sky spectrum, and not a universal shade value. The scoring diagnostic further down keeps the CIE D75 standard.

The magnitudes are an order gentler than 2700K — mean shift around ΔE 3.2 against 10.2 under the warm bulb. Shade rarely breaks a palette; what it does is strip the coziness out of warm undertones you chose on purpose.

Why the room doesn't look this bad

If a 2700K room really looked like the matrix above, every evening interior would be unbearable. It doesn't, because your visual system discounts the color of the light — chromatic adaptation. After a few minutes, most of the amber cast is gone. The strip below shows the same nine colors under 2700K twice: at first glance (the raw stimulus) and settled, after the engine applies CAT16 adaptation at D = 0.85, modelling a viewer whose eyes have mostly adjusted. The settled column is exactly what the product's palette previews render; the scoring diagnostic deliberately uses a harsher, partly-adapted degree — that split is documented on the Method page.

Adaptation recovers most of the appearance — but not the relationships that collapsed. A blue that merged into the neutrals at first glance does not fully un-merge, which is why Percevra scores structure (do distinct colors stay distinct?) rather than absolute appearance. The full two-number scoring story is on the Method page.

One palette, four ways

Here is a five-color interior palette — warm white, sage, slate blue, navy, terracotta — run through the real validation pipeline. Mean shift describes how strong each cast is; retention is the score that matters: how much of the palette's weakest color relationship survives.

This section runs the scoring diagnostic (the tint engine, documented on /method), so its lamp identities are the diagnostic's: Shade here is the CIE D75 standard, not the preview's blue-sky proxy. The strips are first-glance stimuli; the retention figures are computed at the diagnostic's settled view, the same way the product scores every palette.

What to do about it

• Judge a palette under the light it will live in. A bedroom palette spends its evenings at 2700K; a kitchen mostly lives at 4000K. The daylight version on your screen is one appearance out of several, not the truth.
• Distrust blue and violet under warm light. They lose chroma fastest, and distinctions that depend on them — blue versus gray, violet versus mauve — are the first to merge.
• Choose undertones for neutrals deliberately. Whites and greiges carry the cast of whatever light hits them: amber at 2700K, flat gray in shade. A neutral is a decision about which light gets to win.
• Lean warm if the room is warm. Reds, oranges, and earth tones are the steadiest swatches under 2700K and 4000K — they move least and often improve.
• Test structure, not single swatches. The dangerous failures are pairs collapsing, not one color drifting. That is measurable, and it is exactly what the lab below measures.

Method·Light-Aware Palette Lab·Color accuracy