>> Is it wrong ?
> Not in my opinion! Though I think I would either supply @n attributes
> all the time or not at all -- or is your intention to signal their
> presence in the source?
For offshore operators or automatas, this rule seems to me the more robust.
When a line is outside numbering from the editor point of view (ex: a
title added by the editor, for pedagogical reason), you may get 6 <lb/>
instead of 5, and you need an intelligent decision, understanding the
edition rules, to not put @n on the title added. If there are no @n at
all, you can't guess if numbering start at each pages, or each sections.
You can render it nicely for a media, but the canonical reference is
lost. There is probably a way to put some declaration in the header to
specify a numbering scheme (like in the line numbering feature of a text
processor), but I can't see yet a great value added by this extra work.
Such rule is not the best for a TEI enabled scientific editor, but it
seems to me open to better encoding in future, with no more need to go
back to paper.