Colour on Tartan Dictionary

Weaving in software — making the cloth measure true

Sun, 19 Jul 2026 00:00:00 +0000

Every woven swatch in the Dictionary — the samples on the variant pages, the favicon, the background tiles, the patches on the print sheets — is woven in software, thread by thread. This is the story of how that weaving works, and of a calibration problem it turned out to have: the woven cloth measured darker than the dye it was woven from.

The loom, pulled apart

Tartan is a 2/2 twill: each warp (vertical) thread passes over two weft (horizontal) threads, then under two, and each row of weft advances the pattern by one thread. Woven at a few fat threads, with gaps left open so every crossing is visible, the software's loom looks like this:

Where the Rainbow Sits in OKLab

Tue, 14 Jul 2026 00:00:00 +0000

The dictionary does all of its colour reasoning in OKLab — classifying shades, measuring ΔTartan, drawing dye-range maps. Those maps always show colours inside a gamut. This story asks the opposite question: where is the outer edge of all physically possible colour — the locus of pure monochromatic light, the curve that draws the famous horseshoe of the CIE 1931 chromaticity diagram — when you plot it in OKLab?

The fine grid — a colour space sampled evenly

Sat, 13 Jun 2026 00:00:00 +0000

A specifier book like Pantone or Munsell samples colour space unevenly — densely where its trade cares, thinly elsewhere — because it is a record of what people actually made. The Dictionary needs the opposite: a frame in which equal distance is equal perceived difference, filled evenly, so that "how far is this shade from that one" has a single honest answer. That frame is the fine grid.

What it is

Take OKLab — the perceptual colour space the colour list already uses — and simply fill it at one spacing in every direction:

The six-colour reference palette

Sat, 13 Jun 2026 00:00:00 +0000

The Dictionary names every pattern in a simplified six-colour vocabulary — white, black, red, green, yellow, blue, "the rest are shades of these six." These are not screen primaries (tartans are not woven in #FF0000); they are representative cloth shades, chosen to maximise discrimination — to keep the colour distinctions that genuinely separate one tartan from another and throw the rest away. This post re-anchors those six onto the fine grid and re-runs the experiment that picked them.

A tartan's colour order is nearly its name

Mon, 01 Jun 2026 00:00:00 +0000

Reduce a tartan to almost nothing — six colours, and the order they run in, with every thread count thrown away — and you would expect to be left with something far too coarse to tell one tartan from another. Hundreds of designs ought to collapse together. They don't. In the Scottish Register of Tartans, that stripped-down colour sequence is very nearly a fingerprint.

What we kept, and what we threw away

We took the register-of-tartans dataset on its own — 7,393 setts — so that every match is two genuinely distinct tartans, not the same tartan re-listed across sources. Each stripe was quantised to the nearest of six base colours (blue, black, red, green, yellow, white), and the thread counts were discarded entirely. What survives is the Pattern: the bare order of coloured stripes across the half-sett, folded so a reflective sett and its mirror count as one. So B24 K8 G24 R4 and B6 K2 G6 R1 both reduce to the same four-colour Pattern, BKGR.