The offline replay passes for 106 input members, 115 raw QuickGO records, 103 normal TSV rows and 99 frozen source contexts. The official long-product sequence matches the 1,016-residue PANTHER MSA exactly. Five imported IBA contexts join to the literal target ancestry; all seven current path records are retained, including the historical/current ligand-placement difference. A self-donor is valid evidence, not circularity.
Eight complete donor exports are preserved. Their signed mouse angiogenesis records conflict, which motivates the separately reviewed correction queries. The current ARBA rule has matching condition set 16. The Reactome replay retains 13 typed edges, including the three repeated EMILIN1 components; component membership does not imply catalysis.
The interaction replay finds exact positive IntAct records for 18 generic source rows and the separate self row, plus the specific BioGRID, BioPlex and author NDEx corroborations. These are assay-bounded observations, not new mechanistic binding classes.
The author workbook replay finds the two A/B records for 25037231, two target rows for 28327460, and 22 target-bearing rows for 28675934. For the latter, 20 are protein-summary records and two are alternate searches; one protein record is mixed with ERC1. The 23658023 Protein Reports contain eight mouse and five human target records, not thirteen direct human observations. The PRIDE PXD000359 protein group has majority Q9Y6C2, 47 unique peptides overall, positive intensity in all three experiments and blank contaminant/reverse flags. Its group includes long and short sequences, so product-specific activity is not inferred.
The independent original-table checks for aortic matrix, trabecular-meshwork vesicles and the recovered human-vein supplement are detailed in the source guide. Those article bodies are intentionally absent from this data archive. Numerical extraction validates observations; the biological review separately concludes structural integrity and integrin-mediated adhesion, with four explicit uncertainties after generalizing the two collagen-organization rows to unsigned regulation and no new quotations or annotations.
The full generated JSON outputs are recreated by the visible script. The visible expected file records their hashes and summary counts; it does not pretend to contain every derived projection.