| D-086 | 2026-10-19 13:41 | **s53 Part 1 pair list is mis-specified; its label lines on F-20 stop counting (reviewer audit).** (a) Astra's source check (yonichmanuscript runs/explore/plant_group_sourcecheck_20261019.*) found that 8 of the 12 ZL_PHARMA_HERBAL_PAIRS in s53_p4a1_p4e1_anchors_and_errors.py do not match their claimed ZL comments. Opus reconciled them against data/external/page-facts/plants (Stolfi Note 091, Zandbergen) and data/external/page-facts/eyecheck/pairs.tsv (L-358): f1v, f18v, f23r and f37v are right; f3v has no published match; f19r (should be f102v1), f32v (f102r2), f47v (f102r2), f48r (f89v2), f90v1 (f99v), f90v2 (f100v) and f96v (f99r/f100r) are wrong. (b) The F-20 s53 impl and data lines (c-377) are marked unverified with the note. F-20 grade moves TRIANGULATED to CORROBORATED, because it now rests on s48/s53 image matching, s54 (correct ZL cell list) and s128 (18 eye-checked pairs, NO_LINK at power 0.958). The finding's direction is unchanged. (c) Not affected: s54 (its PAIRS match the ZL comments), s128, s115 and s53's image run (results_s53.txt). The s53 duplicate-herbal test inherits the wrong targets and is also void. (d) Any new grouped visual/text bridge must take its pairs from eyecheck/pairs.tsv (39 distinct pairs, verdicts and label confidence). Pairs on the same foldout are not independent observations (Astra). | Opus (reviewer, D-080), from Astra's source check | workflow/FINDINGS-GRAPH.yaml F-20; c-377 — status: active |