19.5 Limits and evidential ladder

E\(_0\) establishes that the typed machinery behaves correctly when perturbations and tickets share a structured representation. Stronger claims require three additional empirical regimes.

1. Extraction robustness requires held-out human or model-generated paraphrases and revisions evaluated after imperfect Toulmin extraction.

2. End-to-end repair requires language-model proposals compared under guarded admission, scalar reranking, unconstrained self-revision, and abstention.

3. Preference-sensitive admission requires testing whether preferences among argument candidates admit a scalar reward, with a relational fallback when cycles or population heterogeneity remain.

The present evidence therefore supports a narrow but substantive conclusion. SCoLT validates typed perturbation, quotienting, localization, and guarded repair for structured argument tickets. It does not yet establish robust extraction from unrestricted prose, reliable generation of novel repairs, or superiority to scalar baselines in end-to-end argumentation.