Documentation on huerestics used during article distillation in Immersive Reader

Chromium's reader mode is known as DOM-Distiller, a Java program transpiled to JavaScript. Mozilla, Vivaldi, Brave, and others use Readability.js. Apple uses a heavily-customized fork of Readability. But Immersive Reader is a mystery.

To what extent, if any, is Edge's Immersive Reader based on existing open-source implementations?
If it is based on existing implementations, which are they?

Is there any way to learn about the algorithm behind its distillation process? Readability.js and DOM-Distiller are open-source, and I can see that Readability understands Microformats, Microformats2, Microdata with Schema.org vocabularies, some RDFa, etc; DOM-Distiller supports JSON-LD and Microdata with schema.org vocabs as well as microformats1. Both support some of Internet Explorer's reader mode markup. All implementations understand semantic HTML tags like `<article>` and headings. Which standards (or non-standards?) does Immersive Reader understand?

I see that dom-distiller.js is credited in edge://credits; however, the extracted content seems to be more in line with what Readability.js extracts than what DOM-Distiller does. For example, endnotes and sections with a high link-density are (thankfully) not filtered out by Immersive Reader.

What's the relationship between Edge's Immersive Reader and Azure Immersive Reader?

Some documentation on Immersive Reader's behavior/parsing for content authors would be appreciated.