How PharmaTek reads regulatory records
PharmaTek links public regulatory records to substances, organisations and physical sites. This page states what each record establishes, how we count, what we publish, and how current each source is. If a finding on the site does not match this page, it is a defect: report it.
Sources and freshness
Each source carries its own date. One fresh source never stands in for the others.
| Source | What we take | Used for | Cadence | Last refreshed | Licence |
|---|---|---|---|---|---|
| FDA Drug Master File list US FDA | DMF number, type, status (active/inactive), holder, subject, submission date | Current and historical API supplier records | Quarterly at source; checked weekly | Not tracked | US government work, public domain |
| EDQM Certification Online Database EDQM, Council of Europe | CEP number, revision, status, holder, substance, issue date | Valid CEP holders per substance | Daily at source; ingested weekly | 2 Nov 2026 | Commercial reuse permitted in writing (9 Oct 2026) with citation, unaltered data and a no-liability disclaimer |
| FDA Drug Establishments Current Registration Site US FDA | Registered establishment, FEI, address, business operations | Physical manufacturing sites and their registered operations | Weekly | Not tracked | US government work, public domain |
| EudraGMDP European Medicines Agency and EU/EEA authorities | GMP certificates and non-compliance statements: site, authority, inspection date, document number | GMP evidence matched to sites and organisations | Weekly | 2 Nov 2026 | Public EMA register. We reuse the published GMP documents with attribution under the EMA's general reuse terms for its website content; we have no source-specific licence, and will remove or change use on request |
| MHRA GMDP UK MHRA | UK GMP certificates and inspection dates | UK GMP evidence | Weekly | 2 Nov 2026 | No reuse terms are published for the register; we rely on the Open Government Licence v3 that normally covers UK public-sector information, with attribution. Written confirmation has been requested from the MHRA and not yet received. |
| FDA Orange Book US FDA | Approved products, listed patents and exclusivities | Product-level patent and exclusivity dates | Monthly at source | Not tracked | US government work, public domain |
| FDA enforcement: recalls US FDA (openFDA) | Drug recall events and classification | Company enforcement history | Weekly | Not tracked | Public domain |
| FDA Warning Letters US FDA | Warning letters to drug manufacturers | Company enforcement history | Weekly | Not tracked | Public domain |
| FDA Form 483 observations US FDA | Published inspectional observations | Company enforcement history | Weekly | Not tracked | Public domain |
| Company firmographics People Data Labs and official company sources | Name, headquarters, size band, website | Company identity on directory profiles | Weekly | Not tracked | People Data Labs data under CC BY 4.0 |
“Last refreshed” is when PharmaTek last successfully loaded the source, from our refresh log. A source that has not refreshed for more than 14 days is shown in amber.
What each label means
- Current record
- An active US Type II DMF, or a CEP, held by the organisation for that substance.
- Historical record
- An inactive DMF. It shows a filing once existed, not that the organisation supplies today.
- Company-level GMP certificate
- A current EU or UK GMP certificate exists for the organisation. It is not matched to the site making a given substance unless a site page says so.
- Site-matched
- A GMP certificate or registration matched to a specific physical site by identifier (FEI, OMS ID) or by exact name and address. Shown on manufacturing-site pages.
- Supplier-reported
- Information a supplier sent us that we could not check against an official register. Labelled as such wherever it appears and never shown as a regulatory record.
- Not found in covered sources
- No record in the registers listed above. Never a negative finding: suppliers may file under another name, sell outside the US and EU, or use an ASMF.
- Model-inferred
- Never published. Relationships suggested by language models were quarantined in August 2026 and are excluded from every public page.
What the records do and do not establish
- A Drug Master File is a confidential submission to the FDA. The FDA does not approve or disapprove DMFs, and an active DMF does not show current commercial availability or qualification for your product. FDA DMF guidelines.
- A Certificate of Suitability confirms a substance can be controlled by the European Pharmacopoeia monograph. It is not a GMP certificate, and the holder is not necessarily the manufacturer; EDQM states its database is not a directory of manufacturers.
- A GMP certificate records an inspection of a specific site for a stated scope on a date. The three-year review window we show is the EMA guideline default, not an official expiry.
- An FDA establishment registration means a site is registered to manufacture for the US market. It is not an approval, and it says nothing about EU or UK GMP status.
- Orange Book patents are listed per approved product and can cover a formulation, combination or use. A product patent date is not a statement that the ingredient is protected.
- Organisation country is where the organisation is recorded, which is often its headquarters, not where a substance is made. Physical locations appear only on manufacturing-site pages.
- Enforcement events (recalls, Warning Letters, Form 483s) are linked to companies by exact-name and identifier matching. Unlinked events are not shown, so a clean profile can mean “not linked”, not “never cited”.
How we count
- Supplier counts are distinct organisations per substance, not certificates or files. An organisation with several DMFs for one substance counts once.
- One organisation is one legal entity. Records group by linked company profile; an unlinked name joins an organisation when it is the same name once legal forms are removed (“GmbH & Co. KG” and “GmbH and Co KG”), with a compatible legal form and country. Merged spellings are shown on the profile as “also filed as”. Parent and sister companies are never merged.
- A CEP counts only when the latest revision of the certificate is valid; an older valid revision of a certificate that has since been transferred, withdrawn or renewed does not.
- One piece of code does this counting for API profiles, A–Z lists, the homepage and the datasets built from it, and regression tests run on every change. The same calculation drives a page's title, summary, tables and structured data, so they cannot disagree.
- Report headlines are fixed, dated snapshots. Each report states its edition date, base population and denominator, and links to a dataset page with the rows behind its chart.
Matching and its limits
Organisations and sites are matched across registers using identifiers first (FDA FEI, EMA OMS organisation and location IDs, DMF holder names as filed) and exact normalised names second. We do not use fuzzy matching: tests on CEP holders showed trigram matching misattributed about 17% of holders at any usable threshold, so unlinked CEP holders are shown by the name EDQM records. Since 23 October 2026 we also keep the EMA SPOR organisation and location IDs that EDQM publishes for about half of all certificates: two holder entries with the same organisation ID count as one holder. A SPOR location is the holder's location, not necessarily where the substance is made. Exact-name matching still makes mistakes, mostly links to a parent or sister company with a similar name; the measured rates are in the table below.
Measured match quality
On 23 October 2026 we reviewed 360 stored links, 60 per link type, sampled at random within an exact-name and a non-exact stratum and weighted back by stratum size. The review found links attached to a parent or sister company and GMP documents attached to another site of the same company. We tightened the matching rules, repaired every stored link (backing up each change), and reviewed a fresh sample of 360 links. A link is correct when it points to the same legal entity or site. For an FDA site, the right company is the FDA registrant named for that site or the establishment's own legal entity; a link to a parent or sister company of both counts as “same group”, not correct.
| Link | Links | Correct before repair | Exact-name sample | Other-name sample | Weighted estimate | 95% lower bound | Coverage |
|---|---|---|---|---|---|---|---|
| DMF holder to company | 24,127 | 97.5% | 25/25 of 23,543 | 35/35 of 584 | 100% | 86% | 92% of active Type II DMFs |
| Enforcement event to company | 21,140 | 92% | 25/25 of 21,084 | 35/35 of 56 | 100% | 86% | 94% of events |
| FDA site to company | 9,610 | 94% | 25/25 of 8,506 | 33/35 of 1,104 | 99.3% | 85% | 96% of FDA-registered sites |
| GMP document to company | 7,025 | 79% | 25/25 of 3,256 | 35/35 of 3,769 | 100% | 88% | 35% of GMP documents |
| CEP holder to company | 7,421 | 87% | 25/25 of 3,258 | 35/35 of 4,163 | 100% | 88% | 81% of valid certificates |
| GMP document to site | 4,768 | 81% | 25/25 of 3,086 | 34/35 of 1,682 | 99.0% | 85% | 24% of GMP documents |
How to read it. Each sample was drawn at random within two strata, links whose names match exactly and the rest, with very different sizes, so the samples are not pooled. The weighted estimate weights each stratum's share correct by its size. The lower bound is a stratified one: each stratum's one-sided exact (Clopper-Pearson) bound at 97.5%, combined with the same weights, which holds at 95% or better overall. With 25 and 35 links per stratum the bound cannot exceed about 88% even when every sampled link is right, so these samples show that errors are rare in the strata reviewed, not that precision is above 95%. Coverage is the share of records that carry a link at all: an unlinked record is not shown on a profile, so coverage, not precision, limits recall. We have not yet measured how many links are missing. The reviews were internal, not independent; a blind sample of a further 300 links, with our verdicts hidden, has been prepared for a second reviewer and their verdicts will be published here. Every reviewed link, with record IDs, the record detail the verdict was based on and review notes, is in the benchmark dataset. Random samples of links are dominated by the largest holders, so we also reviewed one link for every distinct non-exact name: 10 of 10 enforcement names and 36 of 41 DMF holder names were the right entity. The five wrong DMF names (for example Alpharma ApS and Biocon Sdn Bhd) were linked to a same-name company in another country; we unlinked them and legal forms tied to one country (AG, KG, ApS, AB, Sdn Bhd) now only match a company based there. The other errors in the fresh samples were a site linked through a parent's shared FDA registrant number, a CEP holder linked to a group-level profile (see below), and a GMP document for another building on the same business park. Matching now requires a compatible legal form (an AG never matches a GmbH or an Inc), and a GMP document only counts for a site when its address names the site's street as well as its city or postcode. Drawing a further sample then showed the CEP sample had missed a pattern: EDQM records the holder's name followed by its city, and the linker accepted any leading part of that string, so “Sinopharm Weiqida Pharmaceutical Co., Ltd.” sat on a “Sinopharm” profile. We now cut the holder's name at its legal form and require that exact entity: 158 links were removed, 772 exact links added, and a fresh sample of 60 CEP links was all correct. Every reviewed link
Research release checklist
A report edition is published or revised only when every item holds. The scientific reviewer signs off the edition, not just the headline.
- The population is defined once (what is counted, from which register, on which date) and the title, lede, chart captions, structured data, homepage teaser and report index all describe that same population.
- Every number comes from a numbered dataset revision with a SHA-256 hash; organisation counts come from the same code as the API pages.
- No passage claims more than the measurement: filings are not capacity, output, readiness or market concentration; site counts are not volume; a match rate is quoted with its sample and design.
- Earlier figures that change are explained in the report and logged on the corrections page; earlier dataset revisions stay downloadable.
- Certificate and record types are stated where they differ in meaning (for example Chemical and TSE CEPs).
Publication rules
API profiles
Indexed when the substance has at least one evidenced supplier relationship (current or historical), a CEP on the EDQM register, or approved US products in the Orange Book. Identity-only pages stay reachable but are not indexed.
Company profiles
Published when the company has a description, a country, a commercial entity class and at least one hard signal: a DMF, a regulatory dossier, a partner record, an FDA enforcement event or a GMP certificate.
Manufacturing-site profiles
Published when the site has a resolved address and an attributable official registration or certificate with a date. Unknown ownership is stated, not hidden.
Country pages
Headquarters lists (/directory/location) and physical manufacturing dashboards (/countries) are kept separate. A country with no records is not described as having no manufacturing.
Search and filters
Search results and filter combinations are not indexed and are excluded from the sitemap.
Known gaps
Coverage is limited to the registers above: manufacturers that sell only outside the US and EU, or supply under an EU Active Substance Master File without a CEP, will not appear. Company-to-site links are incomplete. Supplier capability, grade, capacity and price are not in any regulator's data and are never inferred.