A chart carries an ordered numeric column on a set of distinct unrelated hues — what does a reader lose?

## Two kinds of colour scale, two kinds of column The **scale** for a colour channel is the rule turning a column's values into colours, together with the legend that lets a reader read them back. Two shapes of that rule matter here: - **a colour ramp** — a continuous graded path, so that moving along the column's values moves steadily along the colour; - **a colour set** — a handful of distinct, deliberately unrelated colours, none of which is meant to be greater than any other. Columns come in two matching shapes: ones whose values have an order (a count, a price, a temperature, a rating) and ones whose values do not (a region, a device kind, a plan name). Matching them correctly is not a style question. The colour scale is a claim about the column, and getting it backwards makes the chart assert something false about the data. ## The four combinations | column | colour scale | what the reader gets | |---|---|---| | ordered | graded ramp | correct: distance in value reads as distance in colour, and direction is visible without the legend | | ordered | distinct set | **order destroyed**: no colour looks bigger, every comparison needs the legend, and a fixed number of entries silently cuts the column into bands | | unordered | graded ramp | **order invented**: the reader concludes the categories are ranked and that adjacent shades are similar things | | unordered | distinct set | correct: difference without implied rank | The two failures are not symmetric in how they get caught. Destroying order looks obviously ugly, and someone usually says so. Inventing order looks elegant — the chart is smooth and coherent — and the false ranking it implies goes unchallenged, especially when the categories happen to be sorted by something in the legend. ## What "perceptually even" does and does not buy you A ramp is **evenly perceived** when equal steps in value look like equal steps in colour. Ramps that are not even manufacture structure: a band where the colour path turns sharply produces a visible edge across a region where the data is perfectly smooth, and readers reliably interpret that edge as a boundary in the data. Older banded defaults are notorious for this, and the change to evenly perceived defaults in modern tools is a genuine improvement. But it answers only the first of the two questions. A perfectly even ramp applied to an unordered category is still asserting a rank that does not exist. So: 1. **Is the ramp ordered and evenly perceived?** A current default usually is; an inherited palette may not be. 2. **Is the column it carries actually ordered?** No tool can know this, because the ordering lives in what the values mean, not in how they are stored. The second question is the one that has to be answered by a person, and it is the one that gets skipped. ## Further things a colour scale decides - **A meaningful midpoint.** Where the column is a change against a target or a signed difference, a single-direction ramp turns the sign into a shade, and a reader cannot see which side of zero a mark is on. A scale with two arms meeting at the midpoint restores it, and only if the midpoint is actually placed at zero rather than at the middle of the observed values. - **A category set with more members than the colour set has entries.** The scale then reuses colours, or squeezes them, and two genuinely different categories become indistinguishable. Fewer categories, or a different channel, beats a crowded colour set. - **Colours separated only by hue.** A set whose entries differ in hue alone fails any reader who cannot separate those hues, whereas an ordered ramp that also varies in lightness survives being read in grey. - **Colour is a low-accuracy channel in any case.** Whether the scale is honest is a separate question from how precisely a reader can recover a number from a shade, and precise recovery is not something a colour scale can offer. ## What to do about the chart in the question Put the ordered column on a graded ramp and the direction comes back for free: the reader sees which end is larger without reading anything. If the distinct colours were there because they are the house convention for that column, that is an argument about consistency, not about correctness, and the chart is still asserting that the column has no order. If the values really need to be read precisely rather than compared, the honest answer is not a better ramp — it is to stop carrying the number on colour and put it where it can be read accurately, on a position or a length. Colour is for the comparison you make across the whole picture at a glance; the legend is what you fall back on when that glance fails, and a scale that forces the fallback on every mark has already given up the thing colour was for.