About ROR Matching — ROR Retriever

What ROR is

The Research Organization Registry (ROR) is an open registry of persistent identifiers for research organizations. A ROR ID (like https://ror.org/05bp8ka05) disambiguates an organization far better than free-text names — “Univ. of Colorado”, “CU Boulder”, and “University of Colorado Boulder” are one organization, one ROR.

RORs are not just for author affiliations. ROR succeeded GRID and absorbed the Open Funder Registry, so in DataCite metadata RORs can identify affiliations (affiliationIdentifier), funders (funderIdentifier), and publishers (publisherIdentifier) — the same connectors the Metadata Connectivity tool measures.

Two ways to search ROR

The ROR API offers two very different searches, and knowing which one you are looking at explains most surprises.

	Affiliation strategy	Keyword query
API	`?affiliation=…`	`?query=…`
Who uses it	The ROR Retriever (first), MGC’s `RORRetriever.py`, DataCite & Crossref matching pipelines	The search box on the ror.org website; the ROR Retriever as a fallback
Designed for	Raw affiliation strings from metadata (“Department of CSE, Dhaka University of Engineering and Technology, Gazipur”) — it tries substrings of the input against registered names	A person typing a name they already know
Confidence	Scores each candidate 0–1 and chooses a match only when it is confident; refuses to choose otherwise	None — results are ranked by token overlap, and even nonsense input usually returns something
Failure mode	Misses organizations whose registered name has extra tokens (see the example below) and no short alias	Confidently wrong top hits on messy strings (see the example below)

How the ROR Retriever combines them

Each affiliation is matched with the affiliation strategy first. Only when it returns nothing usable is the string retried with the keyword query, and the top hit is reported with QUERY in the Match column. Row colors carry the confidence:

Green — the affiliation strategy chose this ROR (score 1.00). Trustworthy.
Amber — a suggestion, not a decision: either a best score match accepted with the “accept best score” option, or a QUERY keyword hit. Verify these by eye before using them.
Red — neither search produced a usable result.

The two checkboxes mirror RORRetriever.py options: exclude acronym matches (--noacronyms) and accept best score if none chosen (--max). The TSV and CSV downloads use the same columns and file naming as the Python tool, with QUERY appearing in match_Affiliation for keyword-search results.

Why results can differ from the ror.org website

The website finds it, the matcher doesn’t. Searching ror.org for Metadata Game Changers returns 05bp8ka05 as the top hit — but the affiliation strategy returns no match. The record’s only registered name is “Metadata Game Changers (United States)”, and with no shorter alias registered the two unmatched tokens push the score too low to choose. The ROR Retriever’s QUERY fallback surfaces the website’s answer in amber. The durable fix is a ROR curation request to add the short name as an alias.

The website answers, but wrongly. The keyword search for Department of Physics, MIT, Cambridge MA returns the Cambridge–MIT Institute — a defunct joint venture, not MIT. The affiliation strategy avoids the trap by refusing to choose. This is why the ROR Retriever runs the careful search first and marks keyword hits for verification, rather than trusting the website’s top hit everywhere.

ROR matching in the browser uses the public ROR API (api.ror.org/v2/organizations); nothing is uploaded or stored. The matching logic is a port of RORRetriever.py. A Metadata Game Changers prototype © Metadata Game Changers, licensed under CC BY-NC 4.0 · DOI: pending.