Encevra Research Round 50 (ORT-50): When Both Readers Are Wrong: Open Removal Tests for Review and Rewriting

Round 50 of the Theory of Collaborative Intelligence program performs open removal tests on 24 new authored packets in three contractual and three non-contract policy families, with six newly verified SEC exhibit windows for a limited receipt or delivery question. It separates deterministic finite-state and semantic-mutation checks from live evaluations by Witnesses A through D. An exact quote gate admits 48 of 48 scripted wrong status mutations. In live reads, no same-wrong decisive unsafe agreement appears, so the study does not establish a safety gain from a second read. Shared false holds remain. On the same 21 complete items, paired review scores 16 of 21 with or without procedure memory. Seven of 29 usable rewrites from correct starting ledgers become wrong HOLD statuses with valid quotes, relative to the intended author labels. A blind label review agrees with all 30 labels but flags specific-instrument versus generic-approval wording that makes some HOLD answers defensible. This is a validity limitation, not hidden adjudicated ground truth. From deliberately wrong starting ledgers, the validator makes 22 correct repairs in 30 outputs. A status-preservation gate routes all 22 repairs, admits eight wrong unchanged statuses, and reduces correct final outcomes from 22 to 12 of 30 by losing ten correct decisive repairs. Twelve repairs already end in HOLD and retain that endpoint after routing. The study narrows the interpretation of preservation: it protects the starting status, not semantic truth. Five experimental responses are empty or truncated and excluded explicitly; the provisional synthesis and corrected final audit are distinguished in the reproducibility archive. Web version: https://research.encevra.com/round-50