The marking seems correct. p3.3 in the text that follows requires "equality-preserving" when the end result is a forward iterator.

Yeah, I'm aware. I'm just being pedantic because what I believe to be normative text requires marking

p3.3 indeed adds the additional requirement, but http://eel.is/c++draft/concepts.equality#4 says that all are equality-preserving unless it uses this comment, and doesn't give leeway to waive the requirement some other way as usually done with the phrase "unless stated otherwise".

"sometimes equality-preserving" is true for all expressions "not required to be equality-preserving". Perhaps "conditionally required to be equality-preserving" is a better fit. This seems like LWG material. Should I open an issue?