One Successful Seed Forces Universality: Selection Jump and Witness-Shape Laws in Exact Witness Architectures

The central theorem of this paper is a selection jump. For a decidable source-independent stagewise verifier (n, c), let_=\e: n\, t\, ( (e, n, t) (n, (t) ) ) \. S_, then S_ is ⁰₂-complete. Thus a single successful seed already forces maximal stagewise complexity. We prove a companion one-shot calibration theorem: for every decidable D (c), the classD=\e: t\, ( (e, 0, t) D ( (t) ) ) \ either empty or ⁰₁-complete. Together these results yield a witness-shape universality principle for exact local verification: one-shot local exactness yields ⁰₁-universality, while stagewise local exactness yields ⁰₂-universality. We place this principle inside a broader barrier hierarchy. At the lower level, finite stagewise prefixes are uniformly insufficient, and undecidable co-c. e. \ witness architectures admit neither decidable one-shot positive certifiers nor exact-domain witness compilers. At the higher level, same-theory adequacy along a universal ₁ embedding yields full ₁ reflection, while arithmetic exact terminality on a truth-faithful image exists exactly when the corresponding fragment truth set is arithmetical, yielding Tarski and diagonal barriers. For fixed propositions with exact two-sided decidable witness packages, these results assemble into a two-sided bridge trichotomy: isolated extensional bridges are vacuous; the effective bridge layer splits into a positive stagewise fibre that is empty or ⁰₂-complete and a negative one-shot fibre that is empty or ⁰₁-complete; and assertion-enriched resolver layers are truth-universal.