nx_dr_verify.nx

nx_dr_verify.nx -- SOVEREIGN citation-entailment verification loop (DR-3). The deep-research vulnerability (Rios-Garcia et al. arXiv:2604.18805, 25k+ runs): evidence is IGNORED in 68% of traces; refutation-driven belief revision happens in only 26%. The fix is STRUCTURAL, not hope: every claim must be entailment-checked against its source evidence, and anything that fails is DROPPED from context memory (blueprint #1 citation traceability + DEER back-tracking of cited AND uncited claims, arXiv:2512.17776). This organ forces evidence consumption by construction: an unsupported claim can never enter memory. Entailment proxy = token CONTAINMENT (fraction of the claim's tokens present in the evidence), integer/permille -- the sovereign LEXICAL tier; PPMI (recall lane) and the no-float LLM are the escalation tiers (same socket, honest tiering). Imports ONLY nx_syscalls = drift-immune. No hardware writes (Rule 26). module: nishi-core.research.dr_verify depends: nx_syscalls.nx genealogy_id: textual_entailment_rte + deer_2026_backtracking + coscientist_reflection