Blogs

What title, keywords, DOI, arXiv id, and PMID mean in Find a paper

Find a paper is one box. After about 300ms it either resolves a DOI, arXiv id, or PMID through OpenAlex, or it searches by title and keywords and shows up to eight hits. Click a hit to fill the form. Find does not spend one of the three generate runs per hour on Free.

Find a paper is one box. Type a title fragment, keywords, a DOI, an arXiv id, or a PMID. conscRAG resolves identifiers first, then fills title, DOI, and abstract.

One field, two kinds of query

Above Title, DOI, and Abstract on the drop form is Find a paper. The placeholder says you can type a title, keywords, a DOI, an arXiv id, or a PMID. Those are not five separate APIs in the browser. After about 300 milliseconds we call GET /api/papers/search. That route looks at the string and decides: identifier lookup or title search.

Click a hit and we copy title, DOI, and abstract into the form. You stay on conscRAG. Find does not use a generate run and does not count against the three Runs per hour on Free. You still press Run yourself when the three fields look right.

Title and keywords: OpenAlex search

If the box does not look like a DOI, arXiv id, or PMID, we send it to OpenAlex works search, sorted by relevance, and show up to eight hits with year, venue, and cite count. This is the right path for a fragment (“scientific inspiration machines”) or a few topical words. It is also how you recover a paper when you remember the nickname, not the identifier.

Keywords here are not MeSH and not conscRAG’s generated profile keywords. They are just query text. Hits can share a phrase and still be the wrong paper. Read the year and venue before you click. OpenAlex relevance is not a guarantee that the canonical 2017 version outranks a later paper with the same title words.

DOI, arXiv id, and PMID: resolve first

A DOI is a 10.… string, optionally wrapped in doi: or a doi.org URL. We ask OpenAlex for that work. If the DOI is a DataCite arXiv DOI (10.48550/arXiv.…) and the exact record is missing, we try the arXiv form once. If OpenAlex still has nothing, we return no hits rather than a keyword search of the DOI digits (that path produced unrelated papers).

arXiv ids

An arXiv id looks like 2305.14259, or arxiv:2305.14259, or an abs/pdf URL on arxiv.org. We map it to the common arXiv DOI and resolve that.

PMIDs

A PMID is accepted as pmid:12345678 or a pubmed.ncbi.nlm.nih.gov URL, not as a bare number (too many false positives). We filter OpenAlex by pmid.

What Find a paper will not do

It will not download the PDF. It will not search your past runs (Your runs is a different page). It will not query Semantic Scholar or PubMed’s own API. And it will not start generation. The button you want after a good hit is still Run, on the same form.

Type at least three characters. Debounce exists so we do not fire OpenAlex on every key. If a hit says “no abstract,” you can still pick it and paste the summary yourself. Generate needs that abstract; Find does not invent one.

Questions

Does a bare number count as a PMID?

No. Use pmid: followed by the id, or a pubmed.ncbi.nlm.nih.gov URL, so we do not treat random digits as PubMed.

Where do arXiv ids come from?

An id such as 2305.14259 is documented by arXiv. conscRAG maps it to the DataCite-style DOI 10.48550/arXiv.2305.14259 and looks that up in OpenAlex.

What is a PMID?

PubMed assigns a PMID to indexed citations. See PubMed. conscRAG asks OpenAlex with filter=pmid:, not PubMed’s own API.