Ὕμνοι · kind palettes

tol-muted

Paul Tol's muted qualitative scheme (wine, teal, olive, rose, green, indigo, pale cyan), each colour nudged ≤ 4.9 ΔE₀₀ to clear cream paper and coated stock.

okabe-ito

Okabe & Ito's colour-universal-design set (vermillion, bluish green, orange, blue, reddish purple, black, sky blue), nudged ≤ 8.1 ΔE₀₀ to fit FOGRA39 and the attention hierarchy.

baseline

The placeholder in site/palette.css: terracotta, aegean teal, ochre, lapis, murex, black-figure, olive.

Method

• Colour difference. sRGB (D65) to CIELAB, CIEDE2000 with k_L = k_C = k_H = 1; the implementation reproduces Sharma, Wu & Dalal (2005) test pairs to four decimals. Targets: ≥ 20 between any two of the seven kinds in normal vision, ≥ 10 under each simulation.
• Colour-vision deficiency. Machado, Oliveira & Fernandes (2009) matrices at severity 1.0 (full dichromacy), applied in linear RGB. The swatch strips and the little graphs above are drawn in the simulated colours.
• Text. WCAG 2 contrast against #faf6ec; 4.5:1 is the bar for the Greek labels. Where a fill falls short, or where darkening would make two labels collide, the palette names a text colour of the same hue (±12°, at least ¾ of its chroma) in its CSS comment. The second block of four columns in the summary measures the seven colours actually used for text. The baseline file names none, so its text colours are the fills darkened along constant hue until they reach 4.6:1.
• Print. littleCMS through Pillow's ImageCms: sRGB → ISO Coated v2 (basICColor), which is the FOGRA39 characterisation for coated stock → sRGB, relative colorimetric, no black-point compensation; ΔE₀₀ between the colour and its round trip. A second trip through PSO Coated NPscreen (FOGRA43) is shown beside it. A round trip under 1 means the colour is inside the press gamut; every fill of the three new palettes is (worst 1.00, Tol's pale cyan, at the edge). CMYK values are the FOGRA39 separations, for the printer. The paper colour itself would be a light tint on white stock, or the stock.
• Attention. "God chroma rank" is where the god colour falls when the seven are ordered by CIELAB chroma; "place salience" ranks the seven by ΔE₀₀ from the paper, and "quietest" means places are nearest the paper of all kinds, yet every place colour stays ≥ 22 from it so a 3 px dot is still seen. The graphs use the real 286 entities, sized by mention count as the site does (×0.62), with nine hymn discs for scale.
• How the new palettes were made. Differential evolution (SciPy) over CIELCh for all seven colours, maximising the worst of (normal ΔE / 20, each simulated ΔE / 10) with penalties for leaving the FOGRA39 gamut, for any kind out-chroma-ing the gods, and for any kind sitting nearer the paper than the places. Pigmenta is confined to hue windows for its pigments (cinnabar 28–42°, verdigris 155–190°, yellow ochre 72–86°, Egyptian blue 250–285°, murex 325–355°, green earth 100–135°, carbon black C ≤ 6); Tol and Okabe–Ito are tethered to their published colours (≤ 4.9 and ≤ 8.1 ΔE₀₀).
• Caveats. CIEDE2000 is calibrated on patches of a few degrees; a 3 px dot subtends far less, and small fields lose chroma first (blue–yellow worst), so real dot-to-dot differences are smaller than the table. Greek labels, not colour alone, identify each node on the graph, and the legend repeats the colours. Machado at severity 1.0 is the worst case; most colour-deficient readers are anomalous trichromats and see more than the simulations show.