# Claim-check dossier: Zenodo 22090940 (25 Aug 2026 snapshot)

Re-verified 22 Oct 2026 by Ash (ash-35@ilands.app). The record has not changed since 2026-08-25.

Why this exists: a count is only worth what a stranger can re-run. Everything below is checkable from the public record with stdlib Python, no dependencies, no trust in me required.

## Source

- Record: Zenodo 22090940, "Problematic images in vendor antibody verification data"
- DOI: 10.5281/zenodo.22090940
- Version: 260825 (published 2026-08-25T08:30:22Z)
- File counted: `260825_catch_all_spreadsheet.csv`, 5,752,227 bytes
- SHA-256: `be36d555d43d5d4ddecc219be6b36e2a718d7a7bf95eb2b744d80476509c3086`
- README: `README.md` from the same record

## Counting rules

Three counts, defined before counting:

1. flagged images = rows in the spreadsheet
2. affected products = distinct (vendor, catalog_number) pairs
3. vendors = distinct vendor strings

Also counted: distinct `original_image_filename` values, to test whether rows and images agree.

Case and whitespace are stripped before comparison. `original_image_filename` is lowercased.

## Result

| vendor | flagged images (rows) | affected products | share of rows |
|---|---|---|---|
| G-Biosciences | 6388 | 6373 | 33.7% |
| Thermo Fisher Scientific | 5581 | 4819 | 29.5% |
| LSBio | 2695 | 2536 | 14.2% |
| Origene | 2236 | 1959 | 11.8% |
| Proteogenix | 362 | 362 | 1.9% |
| Bioss | 305 | 292 | 1.6% |
| R&D Systems | 263 | 171 | 1.4% |
| Antibodies.com | 242 | 242 | 1.3% |
| HUABIO | 209 | 203 | 1.1% |
| Abcam | 185 | 115 | 1.0% |
| Abnova | 154 | 145 | 0.8% |
| Santa Cruz Biotechnology | 125 | 111 | 0.7% |
| Boster Biological Technology | 120 | 96 | 0.6% |
| Millipore Sigma | 39 | 31 | 0.2% |
| GeneTex | 38 | 38 | 0.2% |
| NeoBiotechnologies | 2 | 2 | 0.0% |
| **total** | **18944** | **17495** | |

Top four vendors = 89.2% of all flagged rows.

## Where the record's README and the file disagree, and why

- README says 18,943 images. The file has 18,944 rows and 18,943 distinct image filenames. The gap is one filename appearing twice: `MA5-32928-RAC1-IHC-P-1-20190228112632.jpg` (Thermo Fisher, MA5-32928) sits at image_index 95 and again at image_index 125, filed under two different findings. So README's image total counts distinct filenames, which is defensible; row count is one higher.
- Same cause in the vendor table: README lists Thermo at 5,580 images, the file has 5,581 rows. Every one of the 16 vendor product counts in the README matches my count exactly, and every other vendor image count matches too.

I can't say which row is the duplicate record; both entries describe different problems in one image.

## What this does and does not establish

Does: how many entries are in this version of the record, how they split by vendor, and that the README's own table is internally consistent apart from the one double-entered image.

Does not: that any listed antibody is defective. A flagged entry is a verification image carrying signs of manipulation or reuse. It is not a functional validation result, and the findings are the contributors' work, not mine.

Any rate claim needs a denominator from outside this record. For example, LSBio's response to the database said flagged entries were "less than 0.04% of the total offering"; LSBio's own homepage listed a catalogue of 312,144 antibodies, and 2,536 / 312,144 = 0.81%. That denominator is the vendor's published catalogue figure, not something this file contains. Re-check it before quoting it.

## Reproduce it

```python
import csv, collections
rows = list(csv.DictReader(open('260825_catch_all_spreadsheet.csv', encoding='utf-8-sig', newline='')))
print(len(rows))                                                    # 18944
print(len({r['original_image_filename'].strip().lower() for r in rows}))  # 18943
print(len({(r['vendor'].strip(), r['catalog_number'].strip()) for r in rows}))  # 17495
prods = collections.defaultdict(set)
for r in rows:
    prods[r['vendor'].strip()].add(r['catalog_number'].strip())
print({k: len(v) for k, v in prods.items()})                        # matches the table above
```

If a later snapshot changes any of those four numbers, the change is in the record and should be reported as a diff against this dossier, not as a correction to it.
