The founding-cohort dataset
The frozen founding cohort — 27 products taken apart at label level and the 294 organisms named on those labels — as two CSVs and a JSON file, under CC BY 4.0https://creativecommons.org/licenses/by/4.0/ — Address given, not linked: this site does not link out to the products it audits.. Take it, check it, argue with it, republish it. Attribution is the only condition.
Download
Release 1.9.0, generated from the same records the teardown pages render, last verified .
| File | What it holds | Rows | Size |
|---|---|---|---|
| label-audit-products-v1.csv | What each panel discloses, what of it is verifiable, price, certification, and the dated artefact behind the row. 28 columns. | 27 rows | 25 kB |
| label-audit-organisms-v1.csv | The designator exactly as printed, and what it resolves to when it is looked up. 15 columns. | 294 rows | 122 kB |
| label-audit-v1.json | Both CSVs, plus the findings, the corrections and the published gaps, nested per product. | 27 nested records | 512 kB |
| label-audit-dictionary-v1.csv | Which file each column is in, what it means, and what an empty cell means there. | 43 rows | 5 kB |
This release is the frozen cohort, not every teardown on this site. A further 139 products are deliberately outside these files: adding them would move the denominators every published percentage divides by, under citations already written. The product index lists all 166 and names each one’s cohort.
How to cite it
Under CC BY 4.0https://creativecommons.org/licenses/by/4.0/ — Address given, not linked: this site does not link out to the products it audits. you may copy, remix and build on this data, commercially included, provided you credit the source, link the licence and say if you changed anything. Suggested citation:
US men’s probiotic label audit, 2026: 27 product teardowns, 294 organism-slots. Version 1.9.0, 2026-07-30. bestprobioticformen.com. https://bestprobioticformen.com/data/ — CC BY 4.0.
What the files show
Three figures, each with the denominator it divides by — because a percentage quoted without one is the failure this audit exists to document:
- Products printing a strain designator for every organism
- 51.9 % Fourteen of 27 products. Five of 27 (18.5%) print codes that lead anywhere. Nine of 27 (33.3%) print none at all.
- Products guaranteeing CFU through expiry on the label
- 33.3 % 9 of 27 products. Another 8 of 27 (29.6%) state no potency basis at all. Five state manufacture-only, two state through-expiry off-label only, one contradicts itself between label and website, and two have no CFU number anywhere. The headline is through-expiry 8 plus both-endpoints 1; the seven legs are exhaustive over 27.
- Products with a third-party certification confirmed in the issuer’s own register
- 7.4 % 2 of 27 products. Confirmed means the certifier’s own public database was queried and the specific product was found: Legion (Labdoor, score 100.0, lot 0385A6, tested 14 July 2026) and Sports Research (Non-GMO Project). Checked at the issuer, never at the badge.
All 13 are set out with their arithmetic on what the labels actually say, and all 13
are carried in this page’s structured data and the JSON’s statistics block. If a
measured value in a served release moves, the version is bumped and the correction is logged — a downloaded file travels without the
log that could correct it.
What is wrong with this file
Stated here rather than discovered by whoever reuses it. These limits are also carried in the
JSON release.known_limits block.
- Panels fully transcribed: 26 of 27. Where panel_transcribed is false the brand publishes no Supplement Facts panel in any form, so those organism rows are read from marketing copy and are marked as such rather than counted as a transcription.
- Products with no verifiable US retail price: 3 of 27. They are published unpriced. No price is estimated anywhere in this file.
- Certification status is the result of querying the certifier’s own public register. register-unreachable means the register would not answer, which is not the same as absent-from-register, and the two are never merged.
- The published percentages on the site are computed against the 27-product and 294-slot denominators in this release. Do not recompute them against a filtered subset without saying which subset.
Data dictionary
In the release now, not on this page: one row per column in label-audit-dictionary-v1.csv, and the same
content as the fields object in the JSON. Both are generated from the definitions
that write the CSV headers, so neither can describe a column the files do not ship.
label-audit-products-v1.csv
28 columns, 27 rows. Defined in the dictionary file, under that filename.
label-audit-organisms-v1.csv
15 columns, 294 rows, joining to the product file on product_slug. Defined in the dictionary file, under that filename.
How it was built
On the methodology page, which is the page with that name. The
JSON’s release block carries the same version, licence, coverage and denominator
fields in machine-readable form.