Source code is text. Emoji are text. Some simple math expressions are text, but more complicated ones involving specialist layout rules are not. Architecture plans, no way.
Linearity is part of it. Source code and emoji and simple math expressions can be seen as a linear stream of "characters" (defined loosely), even on a fundamental level. An architectural plan could be transliterated into a linear form, but that would be a transformation.
Of course then you need to explain why audio doesn't count as text. Discreteness? Ease of faithful reproduction?
Architecture is a beautiful example which provokes thinking. From first view it is a non-text, but an architect Antoni Gaudí has developed an attitude with hanging models which made his architecture plans to share some abilities of text even not being written. You may tell me I am crazy but his hanging models is a WYSIWYG model of something from real world so I consider it a text despite it could not be simply copied. IDK maybe a modern architect software shares that attitude, for make these models be copied easy. So my opinion is that architecture plan may be text at least from a structure point of view.
Something closer to a definition of text is parts of their criteria for encoding symbols [1] but even that is imperfect.