194. A series colour is derived from Nord, not taken from it

Date: 2026-08-23

Status

Accepted. The categorical palette for docs/charts.md’s chart widgets, measured rather than chosen.

Context

content-widgets.md §3 says charts take “categorical series colors from aurora + frost hues” and “sequential/diverging ramps interpolated in OKLCH”. Read quickly that says use the Nord hues. Measured, it cannot mean that.

Nord is a UI palette: low chroma, high lightness, designed to sit quietly behind text. Series colours do the opposite job — they carry identity, at small sizes, next to each other. Checked against the six standard categorical checks, Nord’s nine hues used literally fail five:

One measurement explains most of it. Nord’s frost family spans 23° of hue across four members, and nord9/nord10 are 5° apart — the same hue at two lightnesses. That is an elevation ramp, and it is why the toolkit uses it as one. It is not two identities.

This is the same finding ADR-0175 made about the semantic hues, where the theme’s own contrast claim was untrue in five of eight pairs. A palette’s stated purpose is not evidence that it serves a different one.

Decision

Eight slots, re-stepped from Nord’s hue angles

Each slot keeps a Nord hue’s angle and takes a chart-specific lightness and chroma: OKLCH L 0.62 light, L 0.66 dark, C 0.14 clipped into sRGB. So a series still reads as the Nord hue it came from, and also clears the floor that makes it legible as identity.

All six checks pass in both modes: worst adjacent CVD ΔE 12.4 (protanopia), worst normal-vision ΔE 21.9, every slot ≥ 3:1 on its surface.

Dark is its own steps, not a flip. L 0.62 on --gb-surface #3b4252 measures 2.57–2.9:1 — below the 3:1 mark floor — so the dark set is stepped at 0.66, which is inside the dark lightness band and clears it. Two numbers rather than one transformation, for the reason ADR-0087 gives about the semantic hues: contrast is a property of a pair, and the pair is different in each theme.

The order is searched, not chosen

Adjacent slots are what touch in a stacked bar, a grouped bar and a multi-line chart, so the sequence is the CVD-safety mechanism. All 40 320 orderings were scored against the validator in both modes and the best kept. This matters concretely: ordering by Nord’s own numbering puts nord12 orange beside nord14 green, which is ΔE 0.8 under deuteranopia — two series a reader cannot tell apart at all.

Slots are assigned in order and never cycled

A ninth series is not a generated ninth hue: under CVD it is indistinguishable from one of the eight, and generating it would break the property the search just established. Nine series folds the tail into “Other” or becomes small multiples.

nord9 gets no slot

Four frost hues yield two usable identities. Leaving nord9 out is the honest form of that, rather than shipping eight slots of which two collide.

Consequences

• A series red and a danger red are different reds, on purpose. --gb-danger means “this is bad”; slot 5 means “this is series 5”. A threshold band and a series must not be the same colour, or the chart says something it does not mean — which is design-system.md §1.2’s rule about aurora hues carrying semantic meaning, read the other way round.
• The palette is checkable in CI, like the 3:1 sweep ADR-0175 added. The values are a table, the checks are arithmetic, and a slot edited by hand fails the same way a contrast regression does.
• content-widgets.md §3’s sentence is honoured rather than contradicted. It said derived, and this is the derivation; what it did not say is that the literal hues fail, which is why this record exists.
• Eight is the ceiling and it is a real one. The dashboard-grade scope (§3’s “deliberately small”) and the palette’s ceiling agree, which is a convenient accident and not an argument: past eight the answer is a different form, not more colours.
• Nothing renders yet. This is a table and a set of measurements; the widgets that use it are M3’s remaining work.