The paper distinguishes two forms:
- surface-form poeticization: a prose request decorated with line breaks, metaphor, or archaic diction while preserving an extractable payload;
- generative poetic transformation: form and content become inseparable, producing multiple unresolved meanings.
Its theory of crystalline semiosis treats poetic meaning as relational and non-local rather than reducible to isolated tokens.
The paper argues that ordinary threat classifiers may face a category mismatch when confronting poetry because:
- poetic meaning is distributed across structure;
- paraphrase can destroy the relevant object;
- multiple valences can remain simultaneously active.
Its central prediction is:
> measured attack success should rise with semantic density.
That prediction remains untested in this edition and is deferred to a later measurement program.
The edition note explains that:
- v1 appeared inside an artistic “data-leak-as-poem” cluster;
- v2 establishes a standalone scholarly object;
- the earlier state remains on record;
- corrections are disclosed rather than silently overwritten.