palette-check

Validates a categorical chart palette before you ship the chart.

Colourblind-safety is arithmetic, not taste, so this runs the arithmetic: every pair of colours is compared in OKLab under full-colour vision and under simulated protanopia, deuteranopia and tritanopia, and every colour is checked for WCAG contrast against the surface it will actually sit on.

Input

{
  "colors": ["#1f4e79", "#e8c547", "#8b1a1a"],
  "surface": "#ffffff",
  "pairs": "adjacent",
  "min_cvd_delta": 8.0,
  "min_normal_delta": 15.0,
  "min_contrast": 3.0,
  "proof_sheet": true
}

pairs: "adjacent" compares neighbours in assignment order, which is what a reader compares when series sit next to each other. pairs: "all" compares every combination, which is the right check when a filter can leave any two series alone together.

Output

ok, a fixed-width summary table, per-colour swatches, per-pair pairs, and failures / warnings written in plain language with the fix stated. With proof_sheet: true you also get an SVG showing the palette in all four vision types.

A pair FAILs when full-colour separation drops below min_normal_delta or any simulated dichromatic separation drops below min_cvd_delta. Contrast against the surface WARNs rather than FAILs, because a direct label or a table view can carry identity where colour alone cannot.

Method

• sRGB transfer function and relative luminance: WCAG 2.1 / IEC 61966-2-1
• OKLab: Bjorn Ottosson, A perceptual color space for image processing (2020); ΔE is Euclidean distance in OKLab scaled ×100 so the numbers read like CIE ΔE
• Colour-vision-deficiency simulation: Machado, Oliveira and Fernandes (2009), severity 1.0, applied in linear light

Tests

python3 test_palette.py

No dependencies beyond the standard library for the colour maths.