Encevra Research Round 46: Who Must Look Again? Disagreement Routing, False Hold Load, and What a Reviewer Must See

Round 46 of Encevra Research asks the review routing question of the program's road map: when an evidence resolver and an AI reader disagree and the item goes to review, what does a stricter consistency test in the resolver hold back, how much work disagreement routing sends to review and what still escapes, and what must the reviewer be shown. Five new authored contract families (an underwriting lock up, key employee agreements signed with the acquirer, appraisals by an approved appraiser, unconditional landlord consents, and site acceptance after a burn in period) each state an "each" requirement with an aggregate threshold, in four variants: P, in which two records together meet it; S, a near miss that is HOLD under an open record; C, the near miss plus a completeness record (FAIL); and R, the near miss plus a record that cures it (PASS). Every label carries a two completion certificate, and Witness A reviewed all 20 packets blind before any test call. The same record set ledgers are resolved by four fixed rule sets: the Round 45 rules (R45), R45 with a record set consistency test (F1), R46 (F1 plus a universal condition test, pre registered), and an exploratory variant (R46c). A structural diagnostic (E46-A, no model calls) shows that F1 removes one scripted split condition mistake, that only R46 removes the other, and that R46 holds every correctly read cure and turns each into a wrong FAIL under a closed record; it also certifies that a router releasing only on agreement can let an unsafe outcome through only when both components share the error, and certifies 24 new SEC EDGAR exhibits whose annotated ledgers carry a note quoting real filing text. Live (E46-B, 141 completed calls), the blind review agreed with 40 of 40 labels. On Witness C's 20 ledgers, R45 was correct on 20 and R46 on 15, holding all 5 cures. Witness B split the requirement on 15 of its 20 ledgers; under R45 three of them gave a split condition release (a wrong PASS), and the record set consistency test removed all three, so R46 made no wrong PASS on either witness's ledgers. Both direct readers were correct on 17 of 20, and every unsafe direct or reviewer verdict was a FAIL on a near miss item; no direct reading or reviewer verdict was a wrong PASS. Routing the resolver against the other witness's direct reading sent 11 of 20 items to review with 0 escapes (6 with the Round 45 rules); routing two readers against each other sent 4 and let 2 shared wrong FAILs through. A reviewer shown the packet only was correct on 16 of 20 with 4 unsafe FAILs, shown a brief of the disagreement only 14 of 20 with 6 needless holds, and shown both 20 of 20. On 24 real SEC exhibits (E46-C, 48 validator calls), Witness C rewrote every annotated ledger correctly and followed the note on none. In the first run the planned budget tail stopped Witness B's record set ledgers after 8 of 20; the 12 remaining calls were completed the same day, after the first results had been seen (post hoc), with the same harness and settings, and no completed call was repeated. Witness C both wrote the ledgers and reviewed them, so the reviewer results are self review. No production system changed. Manuscript: https://research.encevra.com/round-46. Experiments: https://research.encevra.com/experiments.