Skip to content

Record a country for each name in the ctgov-officials local corpus, so its hand-typed names can be read per country #1451

Closed

Description

@derek73

ctgov-officials is the one local corpus of names nobody chose for a test or a catalogue: 536,981 officials as their studies typed them, titles and credentials included. It keeps three keys per record (name, role, study), so it cannot say how any one country's names read. The fielded corpora that can (ORCID, PatentsView, PSC) hold names composed from boxes, with no titles, commas or post-nominals.

What the registry has. The same API response carries each study's sites. Measured 2026-11-18 on one page of 1,000 studies (fields=NCTId,OverallOfficialName,OverallOfficialAffiliation,LocationCountry,LocationContactName,LocationContactRole,CentralContactName):

  • 862 overall officials. 716 (83%) belong to a study whose sites are all in one country, 73 to a study in several, 73 to a study listing no site.
  • 778 site contacts with a name, each under a site with its own country. The corpus holds none of these today.
  • 375 central contacts.

Proposed.

  1. Fetch LocationCountry with each study. Give an overall official a country where every site of the study is in one country, and leave it empty otherwise.
  2. Decide whether site contacts join the corpus as records of their own, each with its site's country. They are a second hand-typed population about as large as the first.
  3. Register country as the corpus's group, so config_diff.py diff --local ctgov-officials --group <country> and snapshot.py --local can keep one country.

Questions.

  • Is "every site in one country" the right attribution for an official? It says where the study ran, not where the person is from. The affiliation text is the alternative and would need parsing.
  • Do site contacts belong in the same corpus or a sibling one? Adding them changes every count quoted from ctgov-officials (AGENTS.md, decisions entries, issue bodies), which are dated snapshots.
  • The registry writes country names, not codes (Turkey (Türkiye), South Korea). Keep them as written, as PSC keeps nationalities, or map them to ISO codes as ORCID's are?

Done when: a re-fetch records the country, the corpus's module docstring and meta.json describe the new keys, the existing tallies still run, and tools/differential/README.md's "Local corpora" names the group.

Not in scope: scoring. The corpus has no name boxes, so it measures moves between two trees and how often a shape is typed, per country, and nothing about which reading is right.

Activity

  1. added this to the 3.2 milestone on Nov 18, 2026
  2. self-assigned this
    on Nov 18, 2026
  3. derek73 commented on Nov 18, 2026

    @derek73
    OwnerAuthor

    Decided 2026-11-18, the three questions:

    • Attribution: an overall official takes a country where every site of the study is in one country, and none otherwise. No parsing of the affiliation text.
    • Site contacts: a sibling corpus of their own, so every count already quoted from ctgov-officials stays true.
    • Country names: mapped to ISO 3166-1 alpha-2 codes, so the groups line up with ORCID's and PatentsView's in the per-country reports. A name the map does not hold fails the fetch, naming it.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

Projects

No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions