How This Library Was Made

A colour library is only worth what you can check about it. This page is what there is to check: what was measured, on what, what was thrown away, and what is known to be imperfect in the data you are about to download.

4253samples published

36readings per sample, 380–730 nm

3measurement conditions: M2 measured, M1 fitted to its measurement, M0 derived

4253carry an English name

Physical samples, measured

Every colour in this library began as a printed patch, not as a number. Each was measured on an X-Rite i1Pro 3 at 10 nm intervals from 380 to 730 nm, under all three measurement conditions — M2 is measured; M1 is carried as a one-coefficient fit to its own measurement; M0 is derived from that coefficient. What that distinction means is on the Measuring and Viewing page.

The published set is a reduction from a far larger measurement campaign — many times this many patches, measured by hand, of which these are the ones that survived selection. The library you download is the result of that reduction, not the campaign.

What was corrected after the campaign

The stock under a sample is part of the measurement. A transparent ink on the reference stock cannot return more light than the stock does, and 175 samples did — by 2 % or more where the ink is transparent. The spectra separate three causes: an ink that fluoresces under visible light, an opaque white base that is simply brighter than the paper, and a brighter sheet under the ink. Only the last is a fault of the data, and it was corrected on 8 September 2026:

30 colours were transposed to the reference stock — the ink taken as a filter over its sheet, the sheet as a uniform gain over the stock (median 1.037), the curve divided by that gain. Each carries a new code, because the code is derived from the coordinates; the measured curve is kept in the archive under the old code, which is retired. Their provenance says corrected, not measured.
8 colours were withdrawn as duplicates: once on the stock they sat within 1.5 ΔE00 of a colour measured there.
One colour, 0Y2.94, was withdrawn as a sheet — the unprinted paper of another batch, not an ink.
137 colours stay as measured, with the reason recorded: fluorescent inks, opaque bases, and cases the spectra cannot separate. They are listed on Known issues.

Not every colour that beat the stock could be corrected: where the corrected coordinates land on a code already taken, or too close to a code boundary to be unambiguous, the colour is left as measured and says so.

What the reduction was for

Not to be exhaustive. A library nobody can browse is not useful, and a library with two swatches you cannot tell apart is worse than useless. So the selection had two jobs, and both can be checked against the published data:

The job	Measured on the published set
No two colours too close	Closest pair 1.50 ΔE₀₀
No colour too isolated	Widest gap to a nearest neighbour 4.03
Even, not clumped	Average distance between neighbours 2.52
A grey ramp that reads as even	36 neutrals, stepped 1.90 to 2.00 apart

That band — 1.5 to 4.0 ΔE₀₀ — is the specification. Close enough that the steps look even, wide enough that the book stays a usable size. The full distribution, and what the subsets are worth, is on Subset Statistics.

The evidence is the statistics, not the method. How the reduction was performed is not published. What is published is every consequence of it, on every sample, in numbers anyone can recompute from the spectra we ship — which is a stronger claim than an explanation would be.

What is known to be imperfect

Small, measured, and published with the data.

What	How many	Why it is left alone
Stored Lab disagrees with its own spectrum	88	By 0.05 ΔE₀₀ or more, worst 0.44 on `1Y3.79`. Campaign data. Correcting it would mean changing a published colour on the strength of arithmetic rather than a remeasurement.
Fluoresce more than the reference stock	3	`0P5.79`, `0B5.92`, `0P4.90`. Either the ink carries its own fluorescent pigment or that sheet was brighter. Flagged and published as measured.

What is measured and what is computed

The distinction matters more than having both, so it is drawn plainly.

	Measured	Computed from it
Spectra	M0, M1, M2 — all three, per sample	—
Colorimetry	D50 / 2° is published as measured	every other illuminant and observer, from the spectrum
Other substrates	—	a paper with more or less brightener than ours — a feature of the desktop application
CMYK figures	—	through `ISOcoated_v2_eci`, absolute colorimetric
Ranges	—	per print condition, against the ICC profile that describes it

What is frozen, and what may change

A published colour never moves. Its code and its spectra are fixed the day it is published. What may change is its plain English name, and what may be added is a new colour — never closer than 1.5 ΔE₀₀ to anything already there.

A code is never reissued. If a colour is withdrawn, its code is retired permanently and its full record — spectra, colorimetry, name, position in every ordering — is kept. A code that stops resolving is an inconvenience. A code that resolves to a different colour is a defect that spreads silently through production files, and this library will not produce one.

The same rule governs names: a withdrawn name is never given to another colour, in any language.

The instrument, and its noise

Every number on this site sits above a noise floor, and it is worth knowing where it is. Re-reading the same patch without moving the instrument gives differences of about 0.03 ΔE₀₀. Anything smaller than that is not a difference — it is the measurement.

Fluorescence is the exception to that comfort. Two instruments that agree closely on M2 can disagree considerably on M1, because simulating the ultraviolet in D50 is difficult and there is no agreed way to compare how well two devices do it. It is one of the reasons this library codes on M2, and it is the reason an FI figure should be quoted with the instrument that produced it.

If you find something wrong — a colour that does not match its spectrum, a name that reads badly in your language, a gap where a colour should be — say so. Corrections are published, dated, and the reason is recorded alongside them.

For the thorough. The full figures behind this page are on Known issues, in full.