A chart palette has to do three things: separate the classes, keep their order if there is one, and still work for readers who see colour differently.
For ordered values, from low to high. Lightness carries the order.
Lightness by class: 88, 74, 60, 45, 31, 16.
Pass: Neighbouring classes
The closest neighbours are class 5 and class 6, 11.7 apart in ΔE2000. Colourwise treats 10 as a working minimum for chart marks.
Pass: Lightness order
Lightness moves steadily, so the order can be read without a legend.
Check: Greyscale and photocopying
Classes 1 and 2 are close in greyscale. Pale greys are harder to separate, so label them directly.
Pass: Colour vision differences
All classes stay apart in each simulation. Simulations show a typical case, so individual vision will differ.
Check: Contrast with the background
2 of 6 classes fall below 3:1 against the background, which is common at the pale end of an ordered scale. Outline those marks or label them.
These are models of typical dichromacy, so they show likely confusions. Colour vision varies from person to person, and no simulation shows what any one reader sees.
Typical colour vision: no close pairs
Protanopia (simulated): no close pairs
Deuteranopia (simulated): no close pairs
Tritanopia (simulated): no close pairs
Greyscale print or display: classes 1 and 2 may look alike
Use sequential for values that run low to high, diverging for values either side of a meaningful middle, and qualitative for categories with no order. Open a palette from another tool here to check it against the same rules.
More than about eight qualitative classes rarely stay distinct, whatever the colours. At that point, direct labels and small multiples help more than a larger palette.
The simulations use a published model and approximate a typical case. Passing them does not mean every reader will separate every class, so never use colour as the only key. The greyscale check approximates print, and real printers differ.
Ordered palettes step evenly in OKLCH lightness, which is what lets people read order without a legend. Qualitative palettes spread hues and vary lightness on purpose, and candidates that sit too close are moved. Each pair is then compared with CIEDE2000, both as it is and under simulated protan, deutan and tritan vision, using a threshold of ΔE00 10 as the working limit for telling marks apart.