Encevra Research Round 41: Hard by Construction or Hard by Selection? Disagreement Mined Item Pools, Isomorphic Twin Holdouts, and the Consensus Errors That Disagreement Cannot Find

Round 41 of Encevra Research tests the Round 40 proposal to build harder evaluation pools by keeping items on which a four witness panel disagrees. An exact selection diagnostic (E41-A) shows that disagreement selection keeps 42.2 percent of items, doubles the share of hard and trap items, and more than doubles the measured advantage of forced majority over a single witness (0.055 in sample versus 0.023 on the full pool), about half of which disappears on an isomorphic twin holdout (0.028). 13.4 percent of forced majority errors lie on items that disagreement never selects, and a 3 of 4 quorum releases all of them. A live screen of ten authored contract audit families written as isomorphic twins (E41-B) found no disagreement in any family; the only error (F07, an undocumented delivery of quarterly statements) was unanimous on both twins and was held only by a disagreement routed extract then resolve topology. A track on twenty real SEC EDGAR exhibits fixed in advance (E41-C) produced eighteen unanimous correct decisions and two 2 to 2 splits, both in the execution only stratum (Y11, Y13), each released by forced majority and held by quorum and the deterministic resolver. The round concludes that hard pools should be built by construction and validated on held out twins. Manuscript: https://research.encevra.com/round-41. Experiments: https://research.encevra.com/experiments.