Paragraphs in flow content are defined relative to what the document looks like without the a, ins, del, and map elements complicating matters, since those elements, with their hybrid content models, can straddle paragraph boundaries, as shown in the first two examples below.

Is it just these elements or any transparent element?

Let view be a view of the DOM that replaces all a, ins, del, and map elements in the document with their contents. Then, in view, for each run of sibling phrasing content nodes uninterrupted by other types of content, in an element that accepts content other than phrasing content as well as phrasing content, let first be the first node of the run, and let last be the last node of the run. For each such run that consists of at least one node that is neither embedded content nor inter-element whitespace, a paragraph exists in the original DOM from immediately before first to immediately after last. (Paragraphs can thus span across a, ins, del, and map elements.)

I can't really say that I understand this part very well either.

Let view be a view of the DOM that replaces all a, ins, del, and map elements in the document with their contents.

This part is pretty clear. We also agreed that those are the only elements that will be replaced.

Then, in view, for each run of sibling phrasing content nodes uninterrupted by other types of content, in an element that accepts content other than phrasing content as well as phrasing content,

This part is somewhat long and confusing.

phrasing content as well as phrasing content

Does this mean just flow content?

for each run of sibling phrasing content nodes uninterrupted by other types of content, in an element that accepts content other than phrasing content as well as phrasing content,

for each run of sibling phrasing content nodes uninterrupted by other types of content

Are these the two sentences you were talking about?

in an element that accepts content other than phrasing content as well as phrasing content

For each such run that consists of at least one node that is neither embedded content nor inter-element whitespace, a paragraph exists in the original DOM from immediately before first to immediately after last.

some text node
<em>some em node</em>

Does this mean [ P P ]?

And

some text node
another text node

Mean [ P ]?

run of sibling phrasing content nodes uninterrupted by other types of content

Okay, I think I understand this part.

let first be the first node of the run, and let last be the last node of the run

Can we select those on the first example?

I'm talking about this one:

<section>
  <h1>Example of paragraphs</h1>
  This is the <em>first</em> paragraph in this example.
  <p>This is the second.</p>
  <!-- This is not a paragraph. -->
</section>

a paragraph exists in the original DOM from immediately before first to immediately after last

What about that?

despite being grouped in with comments here:

Inter-element whitespace, comment nodes, and processing instruction nodes must be ignored when establishing whether an element's contents match the element's content model or not, and must be ignored when following algorithms that define document and element semantics.