Truth Without a Global Verifier: Correspondence, Shared Error, and Graded Realist Commitment in a Consistency-Constrained Information History

The preceding paper in this series characterized what an embedded observer can reconstruct from finite, local, record-mediated access, and stopped at target-relative correctness: it stated an external adequacy relation between the actual structure and a model class but left the content of that relation — what correspondence is, what makes a truth attribution correct, and how genuine truth differs from shared error — to a later alethic treatment. This paper supplies that treatment. It distinguishes world-fixed truth from observer-indexed truth attribution and formalizes correspondence as a typed satisfaction relation rather than an act of external comparison. It introduces an Alethic Specification that separates the claim type from the world-side target type, types truth-aptness through a standing representation architecture together with world-indexed reference status, and relativizes empirical event and record claims to an admissible record-supported quasiclassical frame. It gives seven satisfaction modules — event-occurrence, record veridicality, propositional, model-adequacy, structural-correctness, modal, and explanatory-adequacy — of which the prior paper's adequacy relation is the model-adequacy instance. It then lifts the prior paper's protocol kernel from candidate models to world candidates and defines alethic transparency: a proposition satisfies the necessary factorization condition for transcript-based determination only when its alethic status is constant across the community's observational-equivalence class; truth-value determination additionally requires a true-or-false rather than undefined status. The Collective Kernel Proposition shows that consensus cannot distinguish world candidates the collective protocol kernel does not distinguish. Two worked models — a shared calibration bias and a falsely unique causal recovery — exhibit arbitrarily precise shared error and unique-within-family falsehood. Finally, the paper develops a graded, selective profile of realist commitment, kept type-distinct from the warrant for it, while denying that truth itself comes in degrees. Truth requires no global verifier, while its detection remains local, partial, and fallible.