Minimal Informational Boundaries Recovering Lineages from Accumulated History by Provenance-Closed, Verdict-Preserving Selection

The public Zenodo record description accompanying Question Lineages names the next research question: “What is the minimal informational boundary required to recover a lineage from accumulated history?” The sentence appears in the record description rather than the deposited paper PDF. This paper formulates that question as an independent provenance problem. Given a complete bounded record R, a specified lineage claim λ, a frozen target verdict V, and a declared disclosure cost μ, a Minimal Informational Boundary (MIB) is a minimum-cost provenance-closed subrecord B* such that V(B*, λ) = V(R, λ). The boundary identifies the least disclosed record sufficient to recover the same record-relative determination under the same rules. The broader formal program targets the complete lineage verdict, including transition validity, standing, abstention, mortality, and contextual classification where applicable. The bounded experiment instantiates a restricted verdict containing classification, the complete admissible chain set, and transition-level diagnostics. Preserving only the final class is insufficient: an empty record can reproduce an obstructed label while erasing the evidence and diagnostics that made the complete-record obstruction meaningful. The study evaluates an exhaustive oracle and a bounded two-phase selector whose fallback is demoted when it exceeds the verified joint-search limit. Against exhaustive enumeration, the verified selector path reached the oracle optimum on all 13 enumerable fixture rows. In a standalone bounded demonstration, a 21-event record was reduced to a 5-event packet, 24.3% of the sensitivity-weighted record was disclosed, all three designated high-sensitivity context events were excluded, and a determinate verdict was preserved. A standalone verifier recomputed packet integrity, internal provenance closure, and the stated verdict. Semantic tampering remained detectable after the packet hash was recomputed. These results establish bounded feasibility, fixture-scale optimality, and packet-level internal verification for the restricted verdict. They do not establish record-scale optimality, a complexity class, or standalone proof that a packet preserves a concealed complete-record verdict; that sufficiency relation remains generator-certified.