Inconstancy: the full matrix

The detail behind Colour Inconstancy. Every colour in the library, computed under each of the eleven standard lights and both standard observers, against D50 at 2°.

ΔE₀₀ against D50 / 2°, M2, all 4253 samples. Computed from the spectra with the library’s own tables, so every row is reproducible from the shipped files.

Four things this table says

Tungsten is the worst room in the building. Illuminant A moves the median colour 3.38 and the worst 17.60. Anything judged under warm domestic lighting is being judged somewhere very far from a booth.

F2 has the worst median, F11 the worst tail. A cool white fluorescent moves nearly everything a noticeable amount; a narrow-band tube leaves most colours alone and destroys a few. Two different kinds of problem, and the second is the one that surprises people.

The observer change is not a small correction. Look at D50 / 10°: median 1.98 with no change of light at all. Several lights move colours less than swapping the observer does.

Illuminant B is a red herring. Its 0.44 median is the smallest in the table, and it means nothing useful — B is close to D50 by construction and is obsolete as a standard. It is in the list because the old browser offered it.

The same handful of samples keeps appearing

7V4.28, 7V5.23, 7V3.37, 6B2.28, 8V6.17, 7B4.55 — saturated violets, blues and deep cyans occupy the worst-case column of almost every row. Their spectra concentrate energy where lamps differ from one another most sharply.

Nothing warm ever appears there. The red end of the spectrum is where lamps agree, so a saturated orange is a far safer brand colour than a saturated violet — not because it is simpler, but because of where the lamps happen to put their energy.

What this is not

These figures recompute each sample under the new light without a chromatic adaptation step. A real observer adapts to the room they are standing in, and a formal CII adapts with them — so the numbers here are the right answer to “what does the instrument say under that lamp” and slightly larger than what a person would report. They rank colours correctly, which is what they are for.