OIML-CS Certificates · Story
How the corpus was made, and why
The OIML Certification System has issued certificates of conformity for decades, published as PDFs on oiml.org. The register is public, but PDFs are where machine readability goes to die: an applicant's name, a type designation, a status, locked in layout. Every authority and every manufacturer who needs the register as data scrapes it privately and drifts.
The corpus answers that once: the whole register downloaded with its manifest, and a stratified sample OCR'd into structured, frontmatter-carried markdown — the certificate's number, applicant, issuer, year, status, and full text, with the extraction method and the source PDF named on every document. Per Recommendation family, a schema is synthesized from the parsed content: the field vocabulary that family's certificates actually use.
The two tiers, honestly separated
The raw tier is the scrape itself: the PDFs, one repository, untouched. The digitalized tier is derived from it by OCR with the declared extraction method. The browser shows you which is which at every certificate: full text where the digitalized tier covers it, the raw PDF where it does not. The manifest lists everything, digitalized or not.
What it feeds
The schemas and the corpus vocabulary inform the machine-readable certificate work (the CNML product's schema space). The corpus is the real-world test set for what a certificate of conformity contains, at scale, across decades.