Ready to Build V: Qualified Admission — Binding Verdicts to the Strength of the Check in the Truth Arbitration Agent

This short paper reports what a fifth agent adds to a series on execution-neutral specifications. The first paper (https://doi.org/10.5281/zenodo.23229267) presented execution-neutral buildability; the second (https://doi.org/10.5281/zenodo.23342767) argued that a specification can fix which of its outcomes must stay distinct; the third (https://doi.org/10.5281/zenodo.23386040) reported conformance records that assert the emitted verdict together with the state; and the fourth (https://doi.org/10.5281/zenodo.23389781) reported a suite that fixes, and checks on representative pairs, the outcome of the state and trigger pairs its workflow does not enumerate. In none of them did the admissible verdict depend on the strength of the check behind it; in (https://doi.org/10.5281/zenodo.23386040), the choice between ALLOW and ALLOW_CONSTRAINED was left to the deploying operator. The Truth Arbitration Agent pack (TAA-TRUTH-ARBITRATION-001) specifies a gate that decides whether a mechanism-level claim may be used in a governed decision, not whether it is true. The Specification requires its validators to declare a class and a strength, and the verdict they permit depends on both: a heuristic conflict check below threshold cannot support a simple ALLOW, and the constrained verdict it leads to must carry the mode, strength, confidence, coverage, and residual risk of the check. The pack also declines to trust the role label it receives: it computes the boundary role itself, makes its central invariant absolute only after that validation, and treats reclassification as an action, not a verdict. The suite checks the qualified outcome on its evidence bundle and on the routes to the shared decided state; its state projection takes the trigger, not the strength of the check, as input. In the pack's NNT companion run, five paradigm-isolated reconstructions in Python, Rust, Prolog, C, and Bash ran the 27 conformance records live and each reported 26 PASS, 1 SKIP, 0 FAIL, and 0 ERROR, with no divergent record and no corrections; none of the 38 questions they recorded was classified as a real residual. In the recorded mutation run, the mutants that replaced the fail-closed default, or let a blocking trigger keep the current state, changed the status of the records that assert those outcomes; no operator targeted the qualified verdict, and no evidence-bundle assertion changed status. A supplementary artifact renders a declared 696-input projection in five further paradigms, with byte-identical text outputs and matching spreadsheet results, and records E for eight inputs, among them both values of the route from the classifying state to the constrained verdict, for which the canonical set states no selecting event or predicate. The implementations that pass the suite differ from that projection on those routes, on Boolean branches that no record asserts, and, in one baseline, on one pair the workflow does not enumerate. We read the specification as evidence that a specification can bind the verdict it permits to the strength of the check that ran, and the run as evidence that its suite can check the qualified outcome where it is reached, though not the binding itself, for this artifact; it is not a certification or a quality score. Use of AI tools. The author constructed the pack's seven canonical files, including the conformance tests, reviewed them manually, and designed and calibrated the harnesses for the reconstruction campaign and the supplementary derivations. Large language models were used openly, as tools to accelerate research and production, with manual review: they assisted the drafting of the pack's documents and the revision of this manuscript and, by design, built the reconstructions and the supplementary derivations that are the object of the demonstration. No tool is an author; the author is accountable for the content. What matters is not which tools touched the documents but whether the artifact can be verified. The claims in this paper are limited to what a reader can check with the shipped hashes, the runnable conformance suite, and the reproducible outcome vectors.