Thanks, @complexpoint, indeed a search expression is the best way to demonstrate the problem.
As your example shows, direct grandchildren are impossible at the logical level but perfectly possible at the visual level. So there is the potential for highly misleading conflicts between logical and visual representation, and I am inclined to say that in those cases the outline is ill-formed.
Lines 2, 3, and 4 in the following example are all level-1 at a logical level. Meanwhile, the last line is counterintuitively lower-level (level 2) than both line 2 and line 3.
Needless to say, the visual representation is deceptive. I’d go as far as saying that the outline itself is plainly ill-formed. But in the current implementation such spurious indentations can be common artifacts of copying and pasting content (or murky thinking).
I do appreciate your point about the spirit of plain text—it is the reason we are all here—but in this case it is possible that the cost of inconsistency might outweigh the benefits of expressivity.
Again, I am not sure myself. I am only pointing out a problem which, I think, begs for a strong answer, especially if FoldingText and TaskPaper are to remain distinct and equal citizens in the long-term product lineup.
Perhaps it boils down to an architectural question: is TaskPaper fundamentally a general-purpose plaintext editor with outlining facilities, or is it a dedicated outliner in plaintext, with built-in constraints to safeguard the integrity of the outline?