color-and-accessibility

Use when choosing colours for a chart or report, when a figure must work for colour-blind readers, in grayscale print or in email, when writing alt text or a text alternative for a figure, or when che

SKILL.md

GitHub

Colour and accessibility

Colour is a data channel, not decoration. Every reader, including the roughly 1 in 12 men with a colour-vision deficiency and anyone reading a grayscale printout, must get the same message.

Colour rules

  1. Assign a role before a colour. Categorical (distinct hues), sequential (one hue, light to dark), diverging (two hues around a stated neutral centre), highlight (one accent, rest grey).
  2. Few colours. Aim for five or fewer categorical colours in a chart; beyond that, group into "Other", use small multiples, or highlight one series and grey the rest.
  3. One colour per category, everywhere. "Google Ads" is the same colour in every figure of a report and across report issues. Keep a mapping table in the report template, not in each chart.
  4. Never colour alone. Pair colour with line style, marker, direct label, position or hatching. Red vs green as the only cue fails for the most common deficiencies; if good/bad must be shown, add a sign, label or icon.
  5. Contrast. For web/WCAG 2.2: text at least 4.5:1 against its background (3:1 for large text); graphical objects needed to understand the chart (lines, bars, markers) at least 3:1 against adjacent colours. Light pastel series on white usually fail.
  6. Grey is for context. Benchmarks, prior years and comparison series in grey; the series the title talks about in the accent colour.
  7. Semantic colours are reserved. If red means "below target" in the report, do not also use red as an ordinary category colour.
  8. Check in grayscale and with a CVD simulator. Grayscale is a useful screen, not a complete test.

A safe categorical starting point is the Okabe–Ito palette (designed for colour-vision deficiencies). Brand palettes must be checked against rules 4–5 before use.

Text alternatives

Every figure that leaves the workspace carries a text alternative. Good alt text states the message and the key numbers, not the chart's appearance and not every data point:

Line chart of weekly paid orders, 31 Aug to 4 Oct 2026. Orders rose from 41 to 47; the week of 14 Sep has no data because the order export failed.

Bad: "Chart showing orders." / "A blue line going up." / a list of all 52 weekly values.

Also provide: the data as a table or download where the medium allows, a specific source line, and the description directly below the figure in web reports.

ieeVisualizer: what it does and does not do

  • Forecasts are distinguished by dash pattern, opacity, hollow markers and hatching, not by colour alone, and the legend says "forecast".
  • Uncertainty bounds take their value series' colour (color_of), keeping one colour per entity.
  • After 10 series the colour cycle wraps (colour_cycle_wrapped); the renderer varies line style/marker so series stay distinct. Fix the figure (fewer series) rather than relying on it.
  • Not enforced: contrast ratios, CVD safety of the default 10-colour cycle, consistent category→colour mapping across figures (set style.color explicitly per layer to guarantee it), alt text (the manifest has no alt text field; supply it in the report). Automated accessibility checks are planned for agent-contributed templates (ROADMAP "Phase 2 onward").

Good / bad

Bad: six acquisition channels in six saturated hues, legend at the bottom, "Paid social" and "Organic" in red and green. Good: the two channels discussed in accent colours with direct labels, the other four grey, labelled at line ends.

Bad: heatmap where "no data" cells are white and low values are near-white. Good: missing cells in a distinct neutral with a "no data" key entry (ieeVisualizer leaves null cells blank: say so in the caption).

Checklist

  • Each colour has a role; categorical palette ≤ about 5 colours.
  • Same category → same colour across the report.
  • Every colour distinction has a second cue (style, marker, label).
  • Text ≥ 4.5:1, essential graphics ≥ 3:1 against their background.
  • Figure checked in grayscale and with a CVD simulator.
  • Alt text states the message and key numbers, plus a data table or link.
  • Missing values look different from low values.

Adapted in part from ONS Service Manual and UK Analysis Function guidance (OGL v3.0), Anthropic data-visualization (Apache-2.0) and K-Dense scientific-visualization (MIT); changed: rewritten, merged, examples replaced. Sources and licences: SOURCES.md.