Ask HN: Best practices for research code?

Writing research code (in my case ML/AI) is very different to writing production code. The goals are different, and thus so are the best practices, patterns, and values.

What's your favorite resource on how to write code in research? What are the research-code-specific equivalents of Rich Hickey's talks or SPJ's posts or the many many SWE blogposts posted to HN?