The equivalent NPE system is a little different to what you did as you also need a color update step. You can use the function ConvertToNodeParticleSystemSetAsync to see how your regular system would look in NPE.
The issue for the initial playground and the conversion function was that they were using a random between two colors, which can do things like produce the violet. While thinParticleSystem what it does is to lerp between both colors. This PR updates the conversion function to Lerp instead of Random, and now the colors are fine.