Main result — SCI = ∞ for presupposition resolution
OpenAutoformalisationSCI.presupposition_resolution_sci_infiniteMain result (Appendix A.2.1, pp. 22–23). Fix and any family of computable enumerations of integer-coefficient polynomials, one for each level. Consider the combined presupposition-resolution problem: given and a level , decide whether is nonempty. Then
This is the precise form of the paper's headline claim that providing semantically faithful AI autoformalisation is harder than any computational problem: a trustworthy autoformaliser that always outputs either a semantically faithful translation or "this statement cannot be translated" must resolve the above question for every and , a problem harder than every problem of finite SCI. It follows from Corollary A.5, since a finite tower for the combined problem would restrict to a tower of the same height for each fixed level , contradicting for larger than that height.
import Mathlib import Definitions.Def_AutoformalisationSCI_SCI import Definitions.Def_AutoformalisationSCI_Poly open AutoformalisationSCI
namespace AutoformalisationSCI
theorem presupposition_resolution_sci_infinite (k : ℕ) (hk : 9 ≤ k)
(E : ∀ l : ℕ, ComputablePolyEnum (l - 1 + k)) :
SCI_A (combinedProblem k E) = ⊤ := by sorry
end AutoformalisationSCIRead-back
What the Lean code literally says, in plain math · claude-opus-4-8
Universally quantified over with , and a family assigning to every natural number a computable polynomial enumeration in variables. It asserts SCI_A (combinedProblem k E) = ⊤, i.e. the Solvability Complexity Index of the combined decision problem (over all levels , encoded by pairing with ) is (= ) in WithTop ℕ: no finite arithmetic tower computes it. The hypothesis is required; the family of enumerations is arbitrary subject to ComputablePolyEnum at each level.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.