Limitations

What this check
can miss.

Every result depends on three things: what the databases held on the date of the check, which of them answered, and whether the reference was matched to the correct article. Each can fail.

Before you read any report

These apply to every reference returned, including those reported as carrying no notice.

Reading the result

No notice on record is not a clean bill

The result states that no notice was found in the sources that answered, on that date. It does not state that the article is sound.

Three situations produce the same output: the article is unproblematic; a concern exists but no notice has been published yet, a delay measured in years; or the reference was matched to the wrong article, so the search never covered the paper you cited.

Illustrative result

Publication record– No notice found

What that meansSearched, not verified

The state is therefore grey, not green. Green is reserved for a retraction reversed on the article itself — the only positive finding available, and narrower than it sounds: where the notice says retract and replace, the version cited stayed withdrawn and the report says so in the colour of a correction.

Identification comes first

A reference without a DOI is matched on weaker evidence

With a DOI, identification is exact. Without one, the match rests on title, journal, year and first author, any of which can agree with the wrong record. Two failure modes, both observed in real reference lists:

A short generic title fits inside a longer one

A two-word book chapter, Endotracheal Suctioning, scored a perfect title match against a reference intended for a different paper. Candidates covering less than half of the submitted title are now penalised, but the failure mode is not eliminated.

A translated title shares almost no words with the record

Three correct records were retrieved and then discarded on token overlap alone. Multilingual reference lists are routine in nursing research, and are where title matching under-reports without signalling it.

How often this happens

In a 200-reference list we use for regression testing, 17 references — 8.5% — could not be matched to a record with confidence and were reported as unresolved rather than as carrying no notice. That figure describes one list, drawn from nursing and health-informatics literature, and will differ with the reference style and the subject.

Verify that each entry carries authors, journal, title and year, or a DOI. Entries missing several of these may be reported as unresolved, or matched to the wrong record.

What can leave a result incomplete

These apply to some checks and not others, or define the boundary of the question answered.

Source availability

A source can fail without saying so

For several weeks Europe PMC contributed nothing to any report, and no error appeared. Reports stated that three databases had been searched when two had. The cause was ours: a single incorrect word in an endpoint URL.

Reports now name the sources that responded and those that did not, in place of a fixed count. Where a source is briefly unavailable the call is retried before it is reported as absent. An incomplete check and a negative result are reported differently.

Choosing between records

Where two records fit, one is chosen silently

When several candidates match a reference and no field separates them — same title, same year, no journal given — the tool reports the closest rather than declining to answer. It does not yet tell you that a choice was made.

Read the Matched to panel on any entry that matters. It names the record and the fields that agreed, which is what a reader needs to catch this.

Timing

A notice published today may not be here yet

Retraction Watch publishes on working days and our snapshot is refreshed on the same schedule. A notice issued since the last refresh will not appear.

Repeat the check close to submission. A result holds for the date printed on it.

Scope

Publication status is the only question answered

Study quality, sample size, methods and statistical validity are outside scope, as is whether a reference supports the claim it is cited for.

Reinstatement means the journal restored the article to the record, or published a corrected version in its place. Neither is a judgement on the science. Critical appraisal remains the reader’s.

About our own evidence

What the benchmark does not establish

Twenty selected references, one input condition, a single assessor, one day. The result describes those twenty cases.

It does not estimate accuracy across the literature and does not establish durable superiority over any other tool, since all of them change. That is the scope wherever the benchmark is cited.

Benchmark scope

20references

1condition

1assessor

1day

A tool that names its limits is not a weaker tool.It is one you can tell when to trust.