Ungrounded
Also known as: UNGROUNDED, ungrounded verdict
A first-class per-claim verdict state in Veritas: the cited source structurally cannot multi-source-corroborate the claim. Not contradicted, not verified, not unverifiable — bounded-honest disclosure that the substrate class routes to ungrounded by architectural design.
Ungrounded is the verdict state Veritas uses when the cited source structurally cannot multi-source-corroborate the claim — most commonly single-source DMS scalars where the experimental measurement appears in exactly one published dataset and no independent re-measurement exists to cross-check against. The claim is not contradicted (we have not shown it false), not verified (we cannot corroborate it), not unverifiable (we know exactly why corroboration fails); it is bounded-honestly disclosed as routing to ungrounded by architectural design.
When automated fact-checking collapses to pass/fail, ungrounded gets silently absorbed into one bucket or the other. Treating ungrounded as “fail” overstates the failure — the claim might still be true; the substrate class simply cannot corroborate. Treating ungrounded as “pass” or “not checked” understates the problem — a reader expects the citation to support the claim, and the substrate class cannot make that support multi-source-checkable. Both collapses mislead.
Most ungrounded outcomes are stable: a single-source DMS scalar is what it is, and the bounded-honest disclosure is the right output. Where a better citation can be found — a source that does corroborate — the substrate-lane can advance the claim with a new substrate-pin and Veritas re-fires the TEST. The original ungrounded outcome stays in the audit trail.
Ungrounded is the load-bearing concept that makes the four-state routing framework different from binary fact-checking. On per-row badges and methodology-page disclosure, the count of ungrounded claims surfaces alongside verified and contradicted counts, never aggregated into a single percentage that hides the distinction.
Layer-grain outcomes feed into the per-claim rollup that produces an ungrounded verdict. L1 emits an ungrounded outcome directly when its check fires on a substrate class that cannot corroborate; L3 and L4 emit layer-specific outcomes (insufficient_evidence and ambiguous respectively) that the rollup can resolve to UNGROUNDED at the per-claim grain.