The Book shows sei being used in conjunction with lu...li'u, not instead of it. This is good, because otherwise, after a sentence marked with sei, how can you tell whether the next sentence is narration or a continuation of their utterance?
So, a quote with a speaker specified should be lu...sei...li'u, and a quote without a speaker would be that without the sei - in other words, simply quoted text. The result is a bare sumti in the sentence, perhaps observing the existence of that quote.
(If instead you choose to not put dialogue in quotes - as in the script to a play - then I assume unmarked sentences would belong to the most recently named speaker, and narration (or stage directions) would be in to'i...toi.)
(I disagree. sei in itself is merely metadiscourse--
whether or not the speaker changes, is not specified. If
a new sentence begins, it is of course a continuation of
the previous discourse, unless explicitly changed by ni'o.
This convention for a draci se ciska is not the only
one, of course, but it is workable and not un-lobykai.
However, I have to address a mere importation of the conventional
drama text conventions, which utilize italics and line spaces
to convey metadiscourse information. In Lojban, such things
ought to be made explicit. An empty line is not a unit of
information in Lojban. Italics are explicit and refer only to
a change in typeface. I realize that any other method will
seem clunky and unnatural, but that's just how it is. The
drama is not a literary form native to Lojbanistan.)
So how is this disagreeing? What's not explicit about to'i...toi?
Or if you're referring to the lu...sei...li'u form of quotes, a new sentence would certainly continue the discourse because it would still be inside the lu...li'u! But what I'm saying is that if a bare quote with no sei occurs, the speaker is zo'e and (as in English text) is most likely the person who spoke the utterance before last.