Changelog

What changed, and when

Every “Updated” date on this site points here. A date only means something if you can see what moved on it, so this page lists each substantive change — and for corrections, the old wording beside the new. Pages do not re-stamp themselves on each visit.

By Ned Walsch

Founder and maintainer of Gluten Screen

Updated October 23, 2026 · what changed

Not clinically reviewed — verify with your pharmacist

23 October 2026 Data

Data and research re-run: every study replicated; one source now unreachable

EU leaflets re-parsed from scratch. 1,777 of 1,874 medicines in the EMA register (April: 1,360 of 1,851; the 447 April parse errors are now zero). Still zero declare wheat starch; 271 contain a starch or dextrin and 49.1% name its source (April: 218, 50.0%). The live EU screening now uses this data. Replication section.
Paired US/EU comparison re-captured. 125 molecules now pair (April: 86). US labels name the starch source for 86.4% against 42.4% for EU leaflets; all 55 discordant pairs favour the US (April: 88.4% vs 38.4%, 43 of 43). Four April molecules no longer pair because of the three-label US sample; their pages keep the April data with its date. Replication section.
Plogsted decay re-run against DailyMed's 22 October publication: 31.9% of entries fail a direct lookup, 67.9% name a manufacturer that is no longer a current labeler (August: 31.6%, 67.6%). Amethia and Amethia LO lost their last US label. glutenfreedrugs.com now serves an expired TLS certificate (expired 4 Oct 2026), so the list could not be re-fetched; the unchanged August copy was used. Replication section.
Open Food Facts re-sample attempted for the food-versus-medication paper; the API rate-limited after 200 products. The partial sample is consistent with August and is published, labelled partial.
Labeler directory rebuilt from the openFDA NDC directory: 9,843 labelers and 138,055 products (August: 9,754 and 137,198); 962 with a medical-information phone read off a label.
FDA substance registry (GSRS) re-harvested: 308 substances, 3,028 synonyms. The 15 UNII codes new since August are all unrelated to gluten grains (beta-lactamases, steroids); no dictionary change.
Product counts on "What you can check" refreshed (lip balm 776, toothpaste 218, sunscreen 5,822, hand sanitiser 13,710).
The April figures are not overwritten anywhere: they are what the papers report, and each paper now carries a dated replication section beside them. October datasets are published under /data/ with -2026-10 in the name, CC0.

23 October 2026 Correction

Every figure re-verified against its source; seven corrections; honest dates

Pediatric study re-credited to its authors. Was: "Mangione et al. 2024" / "2025 Pediatric Pharmacy Association study", in one place with a title the paper does not have. Now: Matson KL, Forbes CL, Burton KE. Gluten-Free Options for the Top 100 Pediatric Medications. J Pediatr Pharmacol Ther 2025;30(3):367–371, doi:10.5863/JPPT-23-00084. The figures quoted (689 NDCs; 50.2 / 22.6 / 25.7 / 1.5%) were correct throughout; the attribution was not.
Portuguese excipient study re-credited. Was: "Lizano-Díez 2024". Now: Figueiredo et al., Sci Rep 2025;15:10976 (308 SmPCs). Figures unchanged.
"Liquid and pediatric formulations carry the risk" withdrawn. Was: "The Portugal study found 40–50% of liquid paracetamol products had gluten-derived excipients — mostly wheat starch as a thickener." Now: the study reports 51.2% of solid and 40% of liquid oral analgesics and does not analyse pediatric products separately; Matson et al. name eight tablets or capsules among nine gluten-containing NDCs. Replaced on the methodology page, the drug pages, the top-ten page and llms.txt with "dosage form is not a safe proxy".
Dictionary size. Was: "146-entry" on eight pages. Now: 163 entries, 489 patterns, 155 UNII codes — counted from the dictionary file at render time on PHP pages.
Homepage EU framing. Was: "EMA leaflets are particularly informative because EU regulators mandate a quantitative gluten declaration." Now: the mandate fires on zero of 1,360 central leaflets; US labels name the starch source more often (88.4% vs 38.4%, 86 paired molecules) — which is what our own research page already said.
Refresh claims. Was: "DailyMed is re-fetched on a twelve-hour cache cycle; EMA, Plogsted and the pediatric dataset refresh on longer cycles." Now: DailyMed cache is 7 days with a ~4-hour change-feed sweep; the EU census is frozen; the other two are fixed publications.
Plogsted list. Was: "Sue Plogsted"; last updated "2017" or "2019" depending on the page. Now: Steve Plogsted, PharmD; structured list 23 April 2019, alphabetical list 28 April 2017.
Dates are now honest everywhere. Drug pages, the top-ten page, the research index and the EU census page printed the current date as "Updated" / "Last verified" on every visit. They now carry the later of the reference-data snapshot and the last substantive template change, and every page links here. Research papers take their modified date from the manifest rather than file timestamps.
Unsourced figures removed from the top-ten page ("over 36 million annual prescriptions", "4× more likely to develop autoimmune hypothyroidism", "elevated histamine sensitivity" and similar). The page now says its ten drugs are a selection, not a measured search ranking.
Methodology page restructured: question-form headings, every study with its denominator, inferences labelled as inferences, short inline citation labels with full titles in a sources list, a visible FAQ whose text is identical to the FAQPage schema, a corrections log, and one next step at the end.
Homepage: the search input now sits above the byline so it is on the first screen of a phone; the opening paragraph carries the key numbers.
Accessibility: muted text, eyebrow labels, the tip button and verdict colours re-tuned to meet WCAG 2.2 AA contrast (4.5:1 for body text) on every background they sit on.
robots.txt rebuilt around the three crawler classes providers now document (search index, training, user-triggered), with Claude-SearchBot, Perplexity-User, Meta, Amazon and Mistral tokens added and the Google-Extended note corrected. Content-Signal line added. IndexNow submission script added for the Bing/Copilot ecosystem.
llms.txt and humans.txt rewritten with the corrected figures and all four research papers.
Page titles capped at 60 characters (Bing Webmaster Tools flagged 38 pages as "title too long" on 23 Oct). Research papers now use their short title in the tab; the full title stays in the heading and citation metadata.
Measurement baseline recorded from Bing Webmaster Tools, Search Console and GA4 (MEASUREMENT.md).
This changelog page added. Audit document: METHODOLOGY-2026-10.md in the repository.

18 August 2026 Data

Dataset registry regenerated

datasets.json rebuilt; every "as of" date on the site now reads from it. EMA census and paired US/EU dataset recorded as frozen April 2026 snapshots; Plogsted recorded as upstream, 2017/2019.
Ingredient fingerprint bug fixed in the label corpus (hash was a function of ingredient count, not names). Migration shipped.

5 August 2026 New

Fourth research paper published

"Two Rulebooks, One Protein": 800 US food products from Open Food Facts against 1,360 EU drug leaflets and 86 paired molecules.

2 August 2026 New

Authorship layer, research index, five new surfaces

Person and Organization schema nodes now defined on every page (previously referenced on four templates and defined on one). Visible byline on every page, with an explicit "not clinically reviewed" line.
New: /research/ with three papers and CC0 data; /compare/ (86 paired molecules); /recalls/ (live openFDA); /literature/ (live PubMed); /pharmacy-card/; /data/ provenance index.
README corrected: the claim that EU leaflets carry quantitative gluten declarations in practice was withdrawn (0 of 1,360).
Dictionary extended from 146 to 163 entries after a 13,999-ingredient corpus test and a GSRS synonym harvest; bare grain names, Avena species and 74 UNII codes added.
Safety note added under every flagged result: do not stop a prescribed medicine because of a label flag.

30 April 2026 Data

EU excipient census captured

1,360 EMA centrally-authorised leaflets parsed. Zero declare wheat starch; 218 contain a starch or dextrin; half of those name no source. Capture date is the measurement date and is deliberately not refreshed.

1 April 2026 New

Methodology page published

"Is there gluten in your prescription?" — the plain-English explainer behind the tool.

Next step. If a figure you cited from this site is listed above as corrected, update your copy and the date you read it. To report something that still looks wrong, use the “Tell us what you found” control on any result, or see the corrections policy.

Screen a medication → · How verdicts are decided · Dataset dates