Documentation
Everything about the registry in one place: what a registrant needs to know, first steps for programs, and the schemas every record is held to.
For registrants
- The registrant's handbook — a DOI in five steps, keeping it right, withdrawing and transferring, fees and timings.
- The policies — naming, metadata, persistence and withdrawal, privacy, registrant terms, fees; every version kept.
- The rules — the states a name can be in, the checks before registration, the ledger, resolution.
- The Smart Scholars DOI Metadata Format 1.0 — the record every DOI is registered with; version 1.1 (9 Oct 2026) adds the original-language title, abstracts in other languages, the AI-use disclosure, data availability and archived copies.
- The DOI System Metadata — what every DOI name carries, kind of work by kind of work.
- The quality procedure — every control between a name being taken and a DOI resolving.
- Governance — who runs the registry, how its rules change, and what happens to every DOI if it stops.
- The continuity copy — every DOI, record and change, written out every night.
Help and news
- Questions: mou@smartscholars.in, answered within two working days.
- The community — questions and suggestions, discussed in the open.
- Status — whether the registry and its Handle server are answering, and over the last 90 days.
- The change feed (Atom; RSS) — every registration, new version, new address and withdrawal.
- OAI-PMH 2.0 (
oai.php) — for repositories and aggregators: every registered and withdrawn DOI as Dublin Core (oai_dc) or the Kernel Metadata Declaration (doi_kernel), by prefix or registrant, changes since any date; a withdrawn DOI is a deleted record. 6 Oct 2026: also as DataCite XML in DataCite's own OAI envelope (oai_datacite) and as Crossref UNIXREF (unixref) — a registrant harvests its own deposited metadata withset=registrant:its number, the role Crossref's DepositHarvester plays (the formats). - The XML query (
query.php) — which DOI held here is this citation: Crossref's XML query form in, Crossref's XML result form out (how it matches). - The full API reference — every call, reading and writing.
Licence — what you may do with the metadata
Every record this registry publishes — on its record pages, in its search, its feeds, its bulk files, its continuity copy and its API, in every format listed below — is dedicated to the public domain under CC0 1.0 Universal. Read it, download it, reuse it, mix it into your own service, for any purpose, with no fee and no permission to ask, as the registry's policies promise. Attribution is welcome and not required: "Data from the Smart Scholars Registry (registry.smartscholars.in), CC0". The works the records describe belong to their authors and publishers and keep their own licences — a record's license field says which.
First steps for programs
Reading needs no key and no account: every DOI's record is public. Each answer below comes from the live registry; the DOI is one it holds. Replace it with any other.
- One record, as JSON — its state, where it resolves, its version, who holds it and the full record:
curl "https://registry.smartscholars.in/api.php?action=get&doi=10.46243/jst.2025.v10.i12.pp35-41"
- Search — by title words, an author, a journal, an ISSN or a year; with counts by year and type:
curl "https://registry.smartscholars.in/api.php?action=search&q=food+printing&per_page=3"
- A citation, by asking the resolver for a format — BibTeX here; also
application/vnd.citationstyles.csl+json,application/x-research-info-systems(RIS) andtext/x-bibliography(APA):curl -H "Accept: application/x-bibtex" "https://registry.smartscholars.in/resolve.php?doi=10.46243/jst.2025.v10.i12.pp35-41"
- The DOI System Metadata declaration — the elements the DOI Handbook gives every DOI name, and whether this record carries them all;
-H "Accept: application/vnd.doi.kernel+xml"instead gives the Kernel Metadata Declaration as XML:curl "https://registry.smartscholars.in/resolve.php?doi=10.46243/jst.2025.v10.i12.pp35-41&as=system"
- Everything at once — what can be downloaded, with the counts; then
as=jsonl(one record a line) oras=csv, for the whole registry or&prefix=one prefix:curl "https://registry.smartscholars.in/bulk.php?as=manifest"
6 Oct 2026: the record in the other formats of the DOI content-negotiation service, deep paging and the XML query are described below — every format, paging and date ranges, the XML query.
Every API answer is JSON with "ok": true, or "ok": false and the reason in words, and may be read from any web page. Writing — reserving, registering, updating, withdrawing — needs a registrant's sign-in or the registry key; it is in the full reference.
Keys, limits and the headers
Reading needs no key. Without one, every read door — the API's reads, bulk.php, feed.php, cite.php, oai.php, query.php and the resolver when asked for data — is counted per internet address: 300 a minute and 10,000 a day, under a one-way code made from the address (never the address itself), erased within a day. A free key gives a program an allowance of its own, 600 a minute and 100,000 a day; a Plus key 1,200 a minute and 1,000,000 a day, and the monthly snapshot. A day runs midnight to midnight, India time. The two levels and how to get a key.
Send the key in a header — X-Api-Key: ssr_… or Authorization: Bearer ssr_… — never in the address (a key in the address is refused: addresses are written into logs). A browser sent on to a DOI's landing page is never counted, and neither is a write.
curl -H "X-Api-Key: ssr_…" "https://registry.smartscholars.in/api.php?action=get&doi=10.46243/jst.2025.v10.i12.pp35-41"
Every counted answer carries X-Api-Level (anonymous, free or plus); X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset (seconds until it starts afresh) for whichever window is nearer to running out; and each window on its own: X-RateLimit-Limit-Minute, X-RateLimit-Remaining-Minute, X-RateLimit-Limit-Day, X-RateLimit-Remaining-Day.
Over a limit: HTTP 429 with Retry-After (seconds) and JSON in words — {"ok": false, "error": "Too many requests: …", "window": "minute", "limit": …, "retry_after": …}. Wait that long and ask again. A request refused for the minute does not use up the day. A key not known, revoked or expired: 401, with the reason; a free key at the snapshots' door: 403.
Citations, relations, data citations and funders
The registry reads, from the records it holds, who cites whom, how records relate, which records point at datasets, and who funded what. It counts only the records held in this registry (registered ones): a citation or a link from a work held anywhere else is not seen here, so a "cited by" number here is never a count of every citation a work has.
| Read (no key) | What it gives |
|---|---|
api.php?action=cited_by&doi=… | the registered records here whose references cite the DOI — the DOI may be held here or anywhere |
api.php?action=references&doi=… | a record's references that carry a DOI, each with the record held here under it, if any |
api.php?action=relations&doi=… | both directions: the relations the record states, and those other records here state about the DOI |
api.php?action=data_citations | links from records here to datasets (&doi= either end, &subject=, &object=) |
api.php?action=funders | funders with the registered works naming each (&name=, &ror=, &funder_id=); one funder's works: api.php?action=search&state=registered&funder= a ROR ID or a Funder Registry ID, or funder.php |
Data citations. Crossref switched its Event Data service off on 23 April 2026 and now offers a data-citation endpoint (in beta), which links a work's references and relations to DOIs it knows as datasets. The registry follows the same idea with what it holds: a reference or a relation of a registered record here to a DOI held here as a dataset; a relation that says data by its own name (Crossref's basedOnData); and Crossref's isDataBasisFor on a dataset's record, read from the other end. A dataset registered elsewhere (at DataCite, say) is not known here to be one unless a relation names it so.
Funders and ROR. Crossref's Funder Registry is moving into ROR. A record keeps a funder's Funder Registry ID (10.13039/…) and its ROR ID when the registrant or Crossref gives them, each checked first (a ROR ID's last two digits are its checksum). Where a record names a funder by its Funder Registry ID alone, no ROR ID is made up: the registry's administrator may ask ROR which organisation lists that Funder Registry ID, and a ROR ID found that way is marked "from ror.org" wherever it is shown. Opening a page never asks ROR or anyone else.
Rebuilding the indexes from the bulk file. Nothing here is kept apart from the records: every line of bulk.php?as=jsonl (and of the nightly continuity copy) carries a record whole, and the indexes come from it alone — record.references[].doi (a DOI name, lower case, without https://doi.org/) gives the citations; each of record.relations[], one entry per identifier of the other item, gives the relations (its role, of and crossref_relation; a doi.org address is the DOI it names); record.funding[], with the agents[] whose role is funder and which funding[] does not already name, gives the funders (a DOI identifier that is a Funder Registry ID, a ROR identifier whose checksum is right, the awards). Only the records whose state is registered are counted across records. The one thing not in the file is a ROR ID looked up at ROR; ask ROR again for those.
One DOI, several addresses
Beside its main address a DOI may have up to ten more — a mirror, a copy in another language, a copy for readers in one country — each with a label, a country (ISO 3166-1, e.g. GB), a language (BCP 47, e.g. hi, pt-BR) and a weight from 0 to 1. A registrant sets them on the console beside the address, or with the API's locations call. The resolver then chooses in this order:
- The link —
?locatt=id:1(each address's number; 0 is the main one),?locatt=country:GB,?locatt=language:hi,?locatt=label:…, or?locatt=mode:legacyfor the main address: the form of the Handle System's 10320/loc value and of Crossref's Multiple Resolution. - The reader's country, when the web server tells the registry what it is — never guessed from an address.
- The reader's language, as the browser asks for it (
Accept-Language):ptfindspt-BR,en-GBfindsen. - When none of these picks one: a page that lists every address, the main one first (HTTP 300, Multiple Choices).
A program asking for metadata — an Accept header or &as= — still gets it, whatever the addresses; a withdrawn DOI still shows its tombstone; a DOI with no other address resolves exactly as before. In the Handle System the main address stays the DOI's URL (index 1) and every address travels in one 10320/loc value (index 1000), the main one first with weight 1: a Handle proxy reading it chooses by the link, then the reader's country, then at random by weight.
curl -X POST -H "Content-Type: application/json" -H "X-Registry-Token: …" "https://registry.smartscholars.in/api.php?action=locations" \
-d '{"doi": "10.46243/jst.2025.v10.i12.pp35-41", "locations": [{"url": "https://…", "label": "Hindi edition", "language": "hi"}]}'
The other addresses come back as locations in every record — api.php?action=get and resolve, the resolver's JSON, bulk.php (JSON Lines; other_urls in the CSV), the nightly continuity copy — and on the record page. Every change is a row in the DOI's history.
Format 1.1 — languages, AI use, data availability, archived copies (9 Oct 2026)
Version 1.1 of the Smart Scholars DOI Metadata Format keeps every 1.0 record exactly as it is — the same fields, byte for byte, the same checksum — and adds six things, all optional. A record that carries one of them says "format": "smartscholars-doi-metadata/1.1"; every other record still says 1.0. The rules are in the 1.1 JSON Schema (every 1.0 record is valid against it) and in the registry's own check, which refuses a broken field in words.
| Field | What it holds |
|---|---|
titles[].original | true marks the title in the original language of a translated work (Crossref's original_language_title). Every title may carry its language (lang, a BCP 47 tag such as hi or pt-BR); a translation of the title has the type TranslatedTitle. |
abstracts | The abstract in other languages, beside the main abstract: [{"value", "lang"}], one a language. |
ai_disclosure | Whether and how AI tools were used — Smart Scholars' own disclosure, ss-ai-disclosure version 1.0 (the categories are ours, not an industry standard): used, uses ([{"category", "tool", "note"}], the tool at most 80 characters, the note 200; an "other" use is described in its note), and, when it was made on a Smart Scholars form, accountability, declared_by (author or editor) and declared_on. Shown in its plain words: "No AI tool was used in preparing this work.", or "AI use: Language editing or proofreading (the tool); … The authors checked all AI-assisted content and take full responsibility for it." |
data_availability | The data-availability statement (at most 2,000 characters) and the datasets: [{"id", "type"}], a DOI name itself ("type": "doi") or an address ("url"). |
archive_copies | Copies of the work an archive keeps: [{"archive", "url", "sha256", "timestamped_at", "verified_at"}] — the archive's name, the copy's https address, the SHA-256 of the file as archived (so anyone can check it is the same file), when the archive took it, and when it was last found there unchanged. |
crossmark | The Crossmark policy the record is under: {"policy"}, the policy page's DOI, with the domains Crossmark is shown on. |
The AI-use categories — the code a record holds, and the words shown for it (Smart Scholars' own, version 1.0):
language— Language editing or proofreadingtranslation— Translation of the authors' own textliterature— Searching or summarising the literaturedrafting— Drafting or rewriting textanalysis— Data analysis or statisticscode— Writing or checking software codefigures— Creating or editing figures or imagesreview— Help with peer review or editorial decisionsother— Other (described in the note)
Read from Crossref. When the registry reads a DOI's Crossref record, it fills: the language; the original-language title, marked (Crossref's API gives its text, not its language); the abstract's own language when Crossref's abstract names one (xml:lang), and several abstracts when it holds several, each in its own language; the Crossmark policy (update-policy) and its domains; the AI-use disclosure from the Crossmark assertions of the group ai_disclosure (one a use, ai_use_ and the category; ai_use_none when no AI tool was used) — read exactly or not at all, never in part; the data-availability statement from the assertion named data_availability; the datasets from the record's isSupplementedBy relations. What Crossref's record cannot carry or does not carry — the archived copies, the abstracts in other languages, the translated titles, a declaration Crossref holds none of — stays as it is held here: a re-read never takes it away, and the record's history says what was kept.
Where each one goes. The record as JSON, the JSON Lines file and the continuity copy carry everything as held. The CSV has six columns more, after other_urls: language, other_titles, ai_use, data_availability, datasets, archived_copies. DataCite XML: every title with its xml:lang and titleType, every abstract a description of type Abstract with its xml:lang, the AI-use and data-availability statements as descriptions of type Other, the datasets as IsSupplementedBy. Crossref UNIXREF and UNIXSD: the original_language_title, every abstract a jats:abstract with its xml:lang, and — only for a record under a Crossmark policy — the crossmark block with the declarations as its assertions and the funding and licence inside its custom_metadata; the archives Crossref's list names, in archive_locations. RDF: every abstract with its language. OAI-PMH (oai_dc): every abstract as a description with its language, the datasets as relations. The feeds: the title's language. The Kernel Metadata Declaration keeps to the Kernel's elements: the titles with their languages, as before.
When the publisher's page stops answering. The registry reads every address on a schedule. Once an address has failed two checks in a row (404, 410, or nothing answering at all — a refusal or a server in trouble never counts) and the record holds an archived copy, the resolver no longer sends readers to the dead page: it shows a small page — "The publisher's page is not answering; here is the archived copy from …" — with both links and the copy's SHA-256, and api.php?action=resolve says the same as fallback. A DOI whose page answers is never affected, and when the page answers again the DOI goes straight there again. The copies are set with the API's archive_copies call (the reference) or in a record update.
The checksum. record_sha256 changes only for a record that gained one of these fields: a record that gained nothing keeps its 1.0 form and the checksum it always had.
"ai_disclosure": {"schema": "ss-ai-disclosure", "version": "1.0", "used": true,
"uses": [{"category": "language", "tool": "…"}], "declared_by": "author", "declared_on": "2026-10-09"},
"data_availability": {"statement": "…", "datasets": [{"id": "10.5281/zenodo.…", "type": "doi"}]},
"archive_copies": [{"archive": "Internet Archive", "url": "https://web.archive.org/web/…",
"sha256": "…", "timestamped_at": "2026-10-02T10:15:00Z", "verified_at": "2026-10-08"}]
Format 1.2 — multilingual by design (17 Oct 2026)
Most of the world's journals print a title in one script and are cited in another. Version 1.2 keeps every 1.1 record exactly as it is and adds two optional things so a record can carry a work the way its own readers write it: a title's role, and a name in other scripts. A record that carries either says smartscholars-doi-metadata/1.2 in format; one that carries neither keeps its version and its checksum. JSON Schema: ssmd-1.2.schema.json · the plain-words page with records in six scripts: Multilingual metadata · every rule on one page: Metadata Profile 1.0 · the round-trip report.
| Field | What it holds |
|---|---|
titles[].role | transliteration — the same words in another script (a Devanagari title in Latin letters); its lang names the script: hi-Latn, ur-Latn, ta-Latn, or the other way, en-Deva. translation — the title in another language, with its lang. Never on the PrincipalTitle, and never together with original: true (1.1's mark for the title in the work's original language, which stays what it is). |
agents[].name.alt | The same person or organisation written in other languages or scripts: [{"display" | "family"+"given" | "org", "lang"}], one form a language or script, at most five — so an author is found under the name on the printed page and the name on the record alike. |
An example — a Telugu article with its title transliterated and translated, and its author in both scripts:
{"format": "smartscholars-doi-metadata/1.2", …,
"titles": [{"value": "తెలంగాణ గ్రామాల్లో భూగర్భ జలాల నాణ్యత", "type": "PrincipalTitle", "lang": "te"},
{"value": "Telangana gramallo bhugarbha jalala nanyata", "type": "Title", "lang": "te-Latn", "role": "transliteration"},
{"value": "Groundwater quality in the villages of Telangana", "type": "TranslatedTitle", "lang": "en", "role": "translation"}],
"agents": [{"role": "author", "name": {"family": "Reddy", "given": "Lakshmi", "alt": [{"family": "రెడ్డి", "given": "లక్ష్మి", "lang": "te"}]}}]}
Where each one goes. The record as JSON, the JSON Lines file and the continuity copy carry both as held. DataCite 4.7 writes a translation as a TranslatedTitle and a transliteration as an AlternativeTitle, each with its xml:lang; Crossref's format takes one original_language_title, which the 1.1 mark already fills; Dublin Core carries the other titles as dcterms:alternative with their language. The record page shows each title with its language and says which is the same title in another script and which the translation.
Every format of a record
The DOI content-negotiation service (citation.doi.org/docs.html; Crossref's content negotiation) hands a DOI's metadata out in the format a program asks for with its Accept header. The resolver here does the same for every DOI it holds — resolve.php?doi=… with the header, or &as= the key below; api.php?action=get&doi=…&as= gives the same; OAI-PMH harvests two of them. Each format is made from the registry's record and from nothing else.
| Format | Accept · as= | What it carries, from the record |
|---|---|---|
| the record (JSON) | application/json · json | the whole record in the Smart Scholars DOI Metadata Format, with its state, address, version and history links |
| DOI Kernel Metadata Declaration | application/vnd.doi.kernel+xml, application/xml, text/xml · xml | the DOI Foundation's Kernel (every element the format maps to it); as=system the System Metadata declaration as JSON |
| CSL-JSON · BibTeX · RIS · APA | application/vnd.citationstyles.csl+json · application/x-bibtex · application/x-research-info-systems · text/x-bibliography; style=apa · csl, bibtex, ris, apa | the citation forms (cite.php) |
| schema.org in JSON-LD | application/ld+json, application/vnd.schemaorg.ld+json · jsonld | a ScholarlyArticle (or Periodical, PublicationIssue, Book, Chapter, Dataset, Report, Thesis, CreativeWork by the record's type): identifier, name, abstract, author (with ORCID iD and affiliations), editor, publisher, datePublished, inLanguage, isPartOf (issue → volume → periodical, with ISSN), pages, url, license, keywords, citation, funder and funding, encoding (the files). The same block sits in every record page's head. |
| RDF — Turtle and RDF/XML | text/turtle · turtle; application/rdf+xml · rdfxml | the same triples in two syntaxes (W3C Turtle and RDF/XML): Dublin Core Terms, BIBO, FOAF, PRISM and OWL — the work as https://doi.org/DOI with its type, title, abstract, creators and contributors (ORCID and ROR as the node), publisher, dates, language, the journal (urn:issn:) with volume, issue and pages, identifiers, license, references, relations, owl:sameAs info:doi/DOI |
| DataCite XML | application/vnd.datacite.datacite+xml · datacite | DataCite Metadata Schema 4.7 (documentation): identifier, creators, titles, publisher, publicationYear, resourceType (mandatory); subjects, contributors, dates, language, alternateIdentifiers, relatedIdentifiers (the journal's ISSN as IsPublishedIn, references, relations), rightsList, descriptions (the abstract), fundingReferences, relatedItems (the journal, book or proceedings with volume, issue, pages). Not made for a record without a creator, a title, a publisher or a publication year — the schema requires them and nothing is invented. A funder known by its identifier alone is left out (funderName is required). |
| Crossref UNIXREF | application/vnd.crossref.unixref+xml · unixref | doi_records/doi_record/crossref against unixref1.1.xsd: journal articles, journals, issues and volumes, books, datasets, reports, standards, components and posted content with titles, contributors (ORCID, affiliations with ROR), abstract (JATS), publication dates (online, print, or other when the record does not say which), pages, article number, funding (fundref), licence (AccessIndicators), relations, doi_data with the address and the text-mining and similarity-checking links, the reference list. Not made where the schema requires what the record does not hold: a chapter (the book's own publication date), a book series (a volume's title and date), a conference paper or proceedings (the conference name and the proceedings' date), a thesis whose record does not hold the date it was approved (dates.approved) or its institution (its publisher), an article or dataset whose record names no journal or database. Posted content is written type="other", since its kind is not held. |
| Crossref UNIXSD | application/vnd.crossref.unixsd+xml · unixsd | the UNIXREF record inside crossref_result/query_result/body/query/doi_record (crossref_query_output3.0.xsd) with the crm-items the registry holds: publisher-name (the registrant the DOI is held for), owner-prefix, deposit-timestamp, created and last-update (the registration and the last change here, UTC), citedby-count — the registered records held here whose references cite it, never a count of every citation. Made for the same records as UNIXREF. |
A format the registry does not make is answered 406 (RFC 9110, "Not Acceptable") with the list of the formats it makes: text/html, application/json, application/vnd.citationstyles.csl+json, application/x-bibtex, application/x-research-info-systems, text/x-bibliography; style=apa, application/vnd.doi.kernel+xml, application/xml, text/xml, text/turtle, application/rdf+xml, application/ld+json, application/vnd.schemaorg.ld+json, application/vnd.crossref.unixref+xml, application/vnd.crossref.unixsd+xml, application/vnd.datacite.datacite+xml. A format the registry makes but cannot make for one record — its schema requires what that record does not hold — is also 406, with the reason in words and the formats that can be made for that DOI; in OAI-PMH such a record is simply not in that format. ONIX for DOI (application/vnd.medra.onixdoi+xml, mEDRA's format) and DataCite's own JSON, Codemeta and JATS answers are not made here.
curl -H "Accept: application/vnd.datacite.datacite+xml" "https://registry.smartscholars.in/resolve.php?doi=10.46243/jst.2025.v10.i12.pp35-41" curl "https://registry.smartscholars.in/api.php?action=get&doi=10.46243/jst.2025.v10.i12.pp35-41&as=turtle"
The mapping is written once, in includes/regformats.php, and every door reads it; tests/test_registry_formats.php validates every XML format against its schema (the copies in tests/fixtures/schemas) and parses the Turtle, RDF/XML and JSON-LD.
Deep paging and date ranges
Cursors. As Crossref's REST API does it (its tips, rest-api-doc): on api.php?action=search or action=list add cursor=* (and rows=, up to 1000) to the first request; every answer carries next-cursor; pass it as cursor= for the next page, URL-encoded; when fewer rows come back than asked, or no next-cursor at all, the end is reached. offset and page are not used with a cursor. A cursor is opaque, tied to the query it was given for (the same words, filters, sort and rows — any other query is refused in words), and lasts 24 hours; after that, start again with cursor=*. Paging is by the sort keys, so a list stays consistent while records are added: nothing is seen twice or skipped. The total is counted once, on the first page, and carried along.
Date ranges, with Crossref's filter names, on search and list: from-pub-date / until-pub-date — the publication date as the record holds it (dates.published; a record dated 2025-12 counts as 1 December, as Crossref counts it); from-update-date / until-update-date — the record's last change in this registry; from-created-date / until-created-date — its registration here (for a DOI first registered at Crossref and brought in, the day it was brought in). Values are YYYY, YYYY-MM or YYYY-MM-DD, inclusive at both ends, UTC (the update and created filters also take a time: 2026-10-01T12, T12:30, T12:30:00). Each is a parameter of its own (&from-pub-date=2025-01) or, as Crossref writes it, inside one filter= (&filter=from-pub-date:2025-01,until-pub-date:2025-12, where prefix:, type:, state: and funder: may stand too). An answer with a date range repeats the filters it applied as filters, and every search item carries published, the date as held.
curl "https://registry.smartscholars.in/api.php?action=search&rows=100&cursor=*&from-update-date=2026-10-01"
The XML query — which DOI is this citation?
query.php takes Crossref's query_batch — POSTed as application/xml, passed as qdata=, or one query given as plain parameters (issn, journal_title, author, volume, issue, first_page, year, article_title, isbn, doi, unstructured_citation …) — and answers with a crossref_result (crossref_query_output3.0.xsd): one query element per hit with the DOI and the fields it was matched on, each marked match="exact", fuzzy or optional; status is resolved, multiresolved, unresolved or malformed. With format=unixref or format=unixsd the matched records come whole instead.
How a match is made. The search's own word matching narrows the candidates (the ISSN, the year, the words of a title, the author's surname); then each candidate's record is compared field by field: a DOI, an ISSN, an ISBN, a volume, an issue, a first page and a year match exactly or not at all (a year one off counts as fuzzy); a title matches exactly when its words are the same, fuzzily when one holds the other or they share most of their words; an author matches the first author's surname (every author with search-all-authors="true"), fuzzily with a typo or two; an unstructured citation matches by the DOI inside it or by most of the title's words. A field marked optional never excludes a record; exact refuses a fuzzy match; null demands that the record has no such field; any other field that does not match excludes the record. A record counts as a hit when the query names it outright (DOI, title, citation, identifier, ORCID iD) or places it in a known journal, book or series (ISSN, ISBN or title, with a first page, a volume and issue, or an author and year). Hits are ordered by score; with enable-multiple-hits="false" (the default) more than one hit is multiresolved with no DOI, as Crossref answers an ambiguous query; "true" lists them; secondary-query="author-title" tries the author and title alone when the full query finds nothing.
Only the records held in this registry are searched. The query is counted like every read (door query, keys and limits). Crossref's own query service also takes an e-mail address (pid=); none is needed here, and one given is echoed in the answer's head.
The schema library
| Schema | What it is | Version | Read it |
|---|---|---|---|
| Smart Scholars DOI Metadata Format | The record every DOI is registered with: every element of the DOI Kernel, and what journals, indexes and readers need. 1.1 adds the original-language title, abstracts in other languages, the AI-use disclosure, data availability, archived copies and the Crossmark policy; every 1.0 record is a 1.1 record. | 1.0 · 1.1 (9 Oct 2026) | specification · JSON Schema 1.0 · Markdown · what 1.1 adds · JSON Schema 1.1 |
| DOI System Metadata | The elements every DOI name carries (DOI Handbook 10.1; ISO 26324:2025, Annex B), and the basic metadata each kind of work needs. | 1.0 | specification · JSON · Markdown |
| DOI Schema (Kernel) | The DOI Foundation's schema for a Kernel Metadata Declaration; every declaration the registry publishes is checked against it. | doi:10.1000/276 | doi.org/10.1000/276 · XSD |
| DOI Schema Attribute Value Sets | The DOI Foundation's lists of allowed values (types, roles, identifiers). | 2.3 (2023), doi:10.1000/282 | doi.org/10.1000/282 · XSD |
| DataCite Metadata Schema | What a DataCite XML record is checked against; the registry writes it for application/vnd.datacite.datacite+xml and oai_datacite. | 4.7 (3 Mar 2026) | documentation · XSD · oai_datacite |
| Crossref UNIXREF | Crossref's unified output schema, "member deposited metadata"; the registry writes it for application/vnd.crossref.unixref+xml and unixref in OAI-PMH. | 1.1 | documentation · XSD |
| Crossref query output (UNIXSD) | The result of a query and the UNIXSD envelope: crossref_result, crm-item, doi_record. | 3.0 | documentation · XSD |
| Crossref query input | The query_batch the XML query takes. | 2.0 | documentation · XSD |
| schema.org · Turtle · RDF/XML | The vocabulary of the JSON-LD; the two RDF syntaxes. | — | ScholarlyArticle · Turtle · RDF/XML |
What the registry hands out, and in which forms: a record as JSON (api.php?action=get, or the resolver with Accept: application/json); the Kernel Metadata Declaration as XML; the DOI System Metadata declaration as JSON (&as=system); a citation as CSL JSON, BibTeX, RIS or APA (cite.php shows all four); the whole registry as JSON Lines or CSV (bulk.php); the changes as Atom or RSS (feed.php); since 6 Oct 2026, schema.org JSON-LD, RDF (Turtle, RDF/XML), DataCite XML and Crossref UNIXREF / UNIXSD (every format).
Browsing, participation reports, the submission log, the public data file
Browsing. Every registrant (registrant.php?id=N) and every prefix (prefix.php?prefix=10.…) has a public page: what it holds by state, year and type, the titles its records name, its newest records, and the ways in — the feed, the bulk files, its OAI-PMH set. The title list names every journal, book series and proceedings the records name, with the ISSN or ISBN the records give it, who holds it, how many DOIs and which years. Every figure on these pages is the search's own: it links to search.php narrowed to the same rows (registrant=, prefix=, container=, year=, type=, state=), so a number and the list behind it are one query.
Participation reports. report.php?registrant=N or ?prefix=10.… shows, element by element, the share of the records that carry the eleven elements Crossref's Participation Reports measure — references, abstracts, ORCID iDs, affiliations, ROR IDs, Funder Registry IDs, award numbers, Crossmark, text-mining URLs, licence URLs, Similarity Check URLs — for current records (this year and the two before) and back records, as Crossref defines them, over the registered records of the work types Crossref's report covers; each share links to the DOIs missing the element, and two CSV files take the summary and every DOI's yes/no away. Crossmark cannot be read from the registry's format and the page says so instead of showing 0. The figures are read from the records held here at the moment the page opens — for a DOI brought in from Crossref, what Crossref held on the day it was read.
The submission log. submissions.php?registrant=N is the registry's answer to Crossref's deposit harvester and submission logs: every ledger entry of the registrant's DOIs, newest first — reservations, registrations (a DOI brought in from Crossref is one whose note says so), new versions, new addresses, withdrawals, restorations, transfers, other addresses — with who did it, which fields changed and the record's checksum after; filters by kind, date and DOI; the whole log as CSV; the same as JSON at api.php?action=submissions®istrant=N. It is open to the registrant's own members and the registry's administrators; each DOI's own history stays public on its record page. The ledger is the log: nothing is kept beside it.
The public data file. datafile.php — like Crossref's annual public data file: the whole registry, every registered and withdrawn record, as one JSON Lines file (gzip) with its SHA-256 and a manifest, made once a year (or once a month, as the administrator sets it) by cron.php datafile and kept; anyone may download it, no key, counted like every other read. The same records are open at any moment, made fresh on request, at bulk.php; a Plus key's monthly snapshot is a separate set of files.
Joining
A publisher, institution, society, funder or scholar becomes a registrant by applying on the Join page, signed in with the account that will act for it: the organisation, what it will register, whether it already registers DOIs elsewhere, and its acceptance of the policies in force. Smart Scholars reads every application and answers within two working days — approved (the registrant is made, the account acts for it, a prefix is assigned), a question (answered on the same page or on My DOIs), or declined with the reason. Every step is written to the applicant and kept with the application; every decision is a row of the registry's ledger.
A sponsor is a registrant that acts for others, as Crossref's sponsoring members act for organisations that are not members themselves: a member of the sponsor sees each sponsored registrant on My DOIs and acts for it exactly as its own member would, under its prefixes; the ledger names both. A sponsorship is set by Smart Scholars — on approving an application through a sponsor, or later — and, once ended, gives no rights from that moment. A suspended or closed registrant stays refused whoever acts for it.
Continuity
Every night the registry writes its escrow set: every registered and withdrawn DOI with the address it resolves to and its full record, the whole change history, and the prefixes with the registrants that hold them — open files with their SHA-256 checksums, which need no software of ours to read. The set is published on the escrow page and kept the same night in Smart Scholars' private escrow repository.
No outside partner holds a copy yet. Smart Scholars will lodge the set with one — a university library or a research institute that keeps its own copy, taken from the registry every night — under a continuity agreement, so that if Smart Scholars ever stops running the registry, the DOI Foundation and a successor can keep every DOI resolving from it. The partner will be named here once it agrees.
What happens to every DOI if Smart Scholars ceases is set out in the policies, section 3 — persistence and withdrawal.



