Skip to main content

Methodology and coverage

Every verdict needs a source. Every missing source needs a refusal.

This page documents the live reference-resolution pipeline, the evidence required for each result, the cases the checker refuses to decide, and what is retained after a run.

Pipeline

Parse, resolve, guard, enrich, classify.

  1. 01

    Parse the input

    Detect DOI lists, BibTeX, RIS, or plain text. Structured title, author, year, journal, and DOI fields are retained when present. Parsing failure remains visible.

  2. 02

    Resolve DOI metadata

    Query the Crossref REST API with an identified User-Agent. A found work, an explicit 404, and a timeout or service failure are separate outcomes.

  3. 03

    Guard the existence verdict

    If Crossref returns 404, query the DOI Foundation registration-agency service. A DOI registered with DataCite or another agency is never labeled absent merely because Crossref lacks its metadata.

  4. 04

    Enrich the record

    Check the Retraction Watch production mirror, Crossref post-publication relationships, and any exact clone-domain match from the Hijacked Journal Checker source. Record whether the mirror answered for this run.

  5. 05

    Classify with refusal states

    Return one highest-priority evidence state and the supporting source. Missing metadata, unavailable services, and inconclusive searches remain unchecked rather than guessed.

Decision rules

What each result means.

The label summarizes the strongest supported state. It does not infer intent, misconduct, study validity, or journal quality.

Retraction
Retraction Watch or Crossref records a retraction, unless a Retraction Watch record says it was reinstated.
Expression of concern
A formal expression-of-concern relationship is present. Retraction Watch does not claim comprehensive EoC coverage.
Correction
A correction, erratum, or addendum is linked. This is not treated as a retraction.
DOI-title mismatch
A structured BibTeX or RIS title has very low overlap with the title registered for the DOI. Plain text is excluded from this test.
No record found
Crossref 404, DOI Foundation no-registration response, guarded title search no-match, and a structurally complete DOI are all required. Manual confirmation is still required.
Not checked
Any missing or unavailable evidence leaves the result unresolved. No accusation is made.
No flags found
The work resolved and no tracked flag was found in the sources queried at that time. This is not a quality assessment.

Primary sources

The checker integrates independent infrastructure.

Crossref REST API

DOI metadata, bibliographic search, and post-publication relationships

Queries use an identified User-Agent, concurrency 3, response-code separation, and a 24-hour metadata cache.

Open the official source

Retraction Watch Database

Curated retractions and selected related notices

The upstream dataset is updated each working day. Manusights uses a DOI-keyed production mirror, records whether that mirror answered for each run, and labels the result incomplete when only the bundled safety sample was available.

Open the official source

DOI Foundation resolution service

Registration-agency check independent of Crossref coverage

The doiRA response prevents a DataCite or other non-Crossref DOI from being mislabeled as unregistered.

Open the official source

Retraction Watch Hijacked Journal Checker

Known clone-domain evidence

Only an exact listed clone domain in the pasted reference can trigger this flag. Journal title and ISSN matches are deliberately excluded.

Open the official source

Data handling

Exact behavior, not a blanket privacy slogan.

The full pasted text is processed on the server. A computed result that includes parsed reference text may be cached for up to 24 hours. The cache avoids repeating the same external lookups.

Saved-run records keep DOI and input-format fields plus sanitized findings. Before database persistence, the exact parsed.raw strings are replaced by a DOI label, parsed title, or neutral placeholder.

Operational email contains only counts, result categories, and detected format. It does not include submitted bibliography text. Submitted text is not sent to a language model or used for model training.

Durable share pages are addressed by unguessable tokens and marked noindex. They contain computed bibliographic fields, not the exact pasted strings. Anyone who receives the link can view the snapshot.

FAQ

Method questions, answered

How does the checker resolve a DOI?
It queries the Crossref REST API with an identified User-Agent and concurrency three. A found record, an explicit 404, and a timeout or service failure remain separate outcomes. Metadata responses may be cached for 24 hours.
Why is the DOI Foundation queried after a Crossref 404?
Crossref is one DOI Registration Agency, not the DOI registry itself. The DOI Foundation registration-agency service prevents a valid DataCite, mEDRA, EIDR, or other non-Crossref DOI from being mislabeled as absent.
How does the checker use Retraction Watch?
A DOI-keyed production mirror supplies curated retraction records and related notice details. Every new result records whether that mirror answered. If it is unavailable, the run is labeled incomplete instead of treating a small bundled safety sample as full coverage. Retraction Watch is treated as comprehensive for retractions but not for expressions of concern or corrections, consistent with its user guide.
How are hijacked journals detected?
The checker extracts domains from the exact pasted reference and compares them with known clone domains from the Retraction Watch Hijacked Journal Checker source. It does not flag by journal title or ISSN because clone sites copy legitimate identifiers.
What about references without a DOI?
The checker attempts a guarded Crossref title, author, and year search when enough metadata was parsed. No confident match becomes an unchecked result, not a claim that the reference is fabricated.
When can the checker return no record found?
Only when Crossref returns 404, the DOI Foundation reports no registration, a guarded title search finds no confident match, and the DOI appears structurally complete. The result still calls for manual verification and does not attribute a cause.

Use the method

Run a bibliography through the documented pipeline.

The result preserves source attribution and unchecked states so you can decide what requires manual follow-up.

Open the Reference Checker

Provenance

Credit the underlying data providers.

The checker is an integration layer. Review provider credits and the correction path before citing or teaching with it.

Open provenance and credits