The services it calls

The app sends data to four services:

Service	What is sent	What is returned
NCBI E-utilities	the query, your email address, and your API key if set. For a PDF you add whose article is not in the archive, its DOI or title.	the PubMed records
Unpaywall	a DOI and your email address	whether a free copy exists, and where
Semantic Scholar	PMIDs, and your API key if set	citation counts and, where known, a link to a free PDF
radiopaedia.work/cite	an identifier you cited	the formatted citation

The app also makes these requests:

• ⤓ Save free PDF downloads the file from the open-access link that Unpaywall or Semantic Scholar returned. The request goes to the publisher or repository in that link, with a browser User-Agent header.
• The PDF reader loads PDF.js (version 4.10.38) from the jsdelivr CDN. The PDF itself is not sent anywhere: pywebview serves it to the reader from a server on your own computer.
• The Radiopaedia window of the study window is radiopaedia.org, opened with your login. With the Radiopaedia search option, the text you search for is sent to radiopaedia.org.

The links on an article card (PubMed, LibKey, DOI, free full text) open in your browser. The LibKey link contains your library ID and the PMID.

The email address is not authentication

NCBI and Unpaywall require a contact address in every request, so that they can reach a user whose requests cause a problem. Neither service sends mail to the address or requires an account. Without an address, Unpaywall returns HTTP 422.

Rate limits

• NCBI allows 3 requests a second, or 10 with an API key. The app waits 0.34 seconds between requests without a key and 0.11 seconds with one.
• Unpaywall states a limit of 100,000 requests a day and no limit per second. The app waits 0.1 seconds between requests, which is about 10 a second.
• Semantic Scholar is queried in batches of 100 PMIDs, 1.3 seconds apart, or in batches of 500, 0.15 seconds apart, with an API key.
• radiopaedia.work/cite is queried one identifier at a time, 0.4 seconds apart.

Phrases PubMed does not find

PubMed lists the phrases it could not match in errorlist.phrasesnotfound. The app shows them as a warning and does not treat them as an error.

The reason is the ISSG filters. They are published search strings several thousand characters long, and they contain terms that match nothing in MEDLINE, for example chemotreatment*. If an unmatched phrase were an error, a search with a guidelines filter and no results would fail with a message about syntax, when the query is valid and there are no results on the subject.

A search field that does not exist (errorlist.fieldsnotfound) is an error, and the app reports it.

Free PDF links: Unpaywall before Semantic Scholar

Both services can return a link to a free PDF. When both have one, the app uses Unpaywall's.