Gödel's incompleteness theorems are usually stated as a gap between a formal system and external mathematical truth. PA can't prove Con(PA), but Con(PA) is “true” — true in the standard model, true by the lights of set theory, true from outside. The incompleteness points outward, from syntax toward some richer semantic domain the syntax can't fully capture.
Note (2026-11-08): the abstract (Gheorghiu) distinguishes derivability from a 'semantic notion of consequence given by support', defined compositionally from the inferential roles fixed by a theory, within proof-theoretic semantics, and says theories such as Q and Peano Arithmetic support their formalized consistency statement without proving it, which reframes incompleteness as a divergence between two internally determined notions of consequence. The essay says both notions are 'syntactic' and describes support as evidence for consistency that falls short of proof; the abstract calls support a semantic notion and does not describe it as evidence.
Gheorghiu (arXiv:2603.09900) reframes this as an internal divergence. Arithmetic has two notions of consequence, both defined within the system itself. Derivability: what follows from the axioms by the rules of inference. Support: what the inferential structure of the theory makes evident without requiring a completed derivation. These are distinct concepts, and they diverge.
PA cannot derive Con(PA) — that's the first incompleteness theorem. But PA supports Con(PA) — its inferential structure, the web of proofs and connections among theorems, constitutes evidence for consistency that falls short of proof. The incompleteness isn't between PA and external truth; it's between two different ways PA relates to its own statements.
The structural shift: incompleteness becomes a phenomenon about the internal geometry of formal systems rather than a failure to reach external reality. The gap isn't syntax-versus-semantics; it's derivation-versus-support, both of which are syntactic. A theory can provide evidence for statements it cannot prove, and the distinction between evidence and proof is already present within the formalism. No model theory required. No Platonic realm invoked. Just two different consequence relations, defined on the same axioms, giving different answers about the same statement.