@Soda, I don't have proof, but I think how it works is that the later attributes of the pagelist tag take priority over previous ones if there's an overlap. However text labels like "-" don't have a "numeric nature" any more, so a number scheme like "roman" has no effect.

In fact, _any_ numbering scheme has this effect:

<pagelist
1to10=—
5=1
/>

Where page 5 will not appear as 1, it's still "—". On the other hand, text can still overwrite text:

<pagelist
1to10=—
5=foobar
/>

and now page 5 is labelled foobar.

<pagelist 1to10="-" 2to4="roman" />

for the element in red

we can see that the type "roman" where well applied
but for the number it stay "-" but the space &nbsp; is well applying too due to the "roman" type

so it's a problem with the set of the text part

I have asked the advice on https://fr.wikisource.org/wiki/Wikisource:Scriptorium/F%C3%A9vrier_2022#%3Cpagelist_/%3E_bug_avis
If we should modify how that work now, but everyone asked to don't do anything.
They think the overwrite is commonly use and could break something.
For them they say the romans characters are less used than numeric so it's normal for them to have this behavior

Furthermore, they say the visual editor is commonly used such case should not append in the future.

So may I close this ticket ?