| L-377 | A within-page token shuffle used as the null for boundary-adjacency statistics produced a false cross-line avoidance finding, because line-start and line-end vocabularies differ | F-40's cross-line core deficit (O/E 0.56-0.67 against a page-token shuffle, s87; the same null in roach r5, r7 (c) and r11 (c)) disappears when whole lines are reordered within pages (db-s-core-return-20261019, independent audit PASS: O/E 1.06-1.18, residual suppression 0/3). It had been used as a criterion to reject pool generators. The external antenore toolkit makes the same folio-shuffle choice. Change: any statistic that pairs tokens across a boundary (line, paragraph or page) needs a null that keeps each token's position class, such as whole-unit reordering or comparing against the same position in other units, as well as the plain shuffle. A failure to reject under that null is reported as a bounded scope correction, not as proof of no effect (O-086). |