Skip to content
Tinnitus Clarified

How we check

What this site does to catch its own errors, what those checks cannot see, and the numbers they produce. Every figure below is counted when the site is built or read from a dated run of the check that produced it — none is typed by hand.

Last changed

The site by the numbers

Counted when this page was built, on , unless a figure says otherwise.

articles
219
Counted at build
citations, to 1,266 unique sources — every one listed
1,757
Counted at build from each article's source list
sources per article, the median. 25 articles cite fewer than three, shown here rather than left for you to find.
5
Counted at build
corrections on 96 pages, between September 3, 2026 and September 26, 2026, each with the old wording quoted
117
Counted at build from the corrections log
articles in the Treatments section carry an evidence grade. The rest are ungraded, not graded “unknown”.
30 of 53
Counted at build from the treatment registry
articles reviewed by a clinician. Why, and what happens instead
0
Counted at build from the reviewer register
cited papers checked for retractions and expressions of concern: none found. What that covers
919
Run (the day this page was built)
current Cochrane reviews about tinnitus cited somewhere on the site; 12 of 12 on the article about their subject. Which, and where
12 of 12
Run (the day this page was built)
PubMed-linked citations whose first author and year were compared with NCBI’s record: none disagreed. What that covers
1,210
Run (the day this page was built)

What the sources are

The 1,266 unique sources, by type. Types are assigned by fixed rules — from the citation’s own wording and a maintained list of publishers — rather than decided one citation at a time; a citation that does not state its design is counted by where it was published. Which publisher belongs on which list is itself an editorial call, made once per publisher.

  • Clinical guideline28 (2%)A clinical practice guideline: a professional or government body's recommendations on what clinicians should and should not do, usually built on a systematic review of the evidence.
  • Systematic review196 (15%)The citation describes a systematic review, meta-analysis or Cochrane review — evidence pooled across multiple studies.
  • Clinical trial77 (6%)The citation describes a randomised or controlled trial: an intervention tested against a comparison group.
  • Observational study108 (9%)The citation describes a cohort, cross-sectional, prevalence or case study — observed rather than assigned.
  • Narrative review31 (2%)The citation describes a review that summarises a field without a systematic search protocol.
  • Journal article647 (51%)Peer-reviewed literature where the citation does not state the study design.
  • Health authority93 (7%)Guidance or reference material from a government health body, regulator, or major medical centre.
  • Patient organisation30 (2%)Material from a professional body or patient charity such as ASHA, the ATA or Tinnitus UK.
  • Preprint or repository3 (0%)Posted to a preprint server or repository and not peer-reviewed at the point we link to it. Worth weighing accordingly.
  • Commercial or vendor7 (1%)A page belonging to a clinic, device maker, retailer, insurer or law firm — an organisation with something to sell or a case to argue. Sometimes exactly the right source, for a device's own specification or a stated legal criterion. Never strong evidence that a treatment works.
  • Unclassified46 (4%)The citation and host together are not enough to classify it. Shown rather than hidden or guessed at.

Filter the whole list by any of these types in the research database.

The checks

Each of these is a script that reads the site’s own content and, where it needs to, asks an outside authority — PubMed, through NCBI’s public interface — rather than trusting the site’s own records. The result of each run is saved with its date, and that saved result is what this page shows. None of them reads a paper and confirms it supports the sentence that cites it: that is done by reading, and it is the part of the work no script can vouch for.

Retractions and expressions of concern

What it checks. Every paper the site links to on PubMed or PubMed Central — in article sources and bodies, and in the site's own pages and components. PubMed Central links are resolved to PubMed records through NCBI's ID converter. Each record is checked for a Retracted Publication type and for retraction, expression-of-concern, erratum and correction notices.

What it cannot check. Only papers with a PubMed record are checked. A source linked through its publisher, a DOI or another site is not examined by this check, and neither is a health-authority or organisation page. It relies on PubMed having recorded the notice, so a retraction PubMed has not yet indexed would be missed. A clean result ages: a paper can be retracted after the date of the run.

Last result. Run (the day this page was built): 919 papers checked; no retraction and no expression of concern found. 14 papers carry a routine erratum or correction notice, which is usually a fixed typo or figure label and is listed rather than counted. Of the 1,266 unique sources the articles list, 1,034 link to PubMed or PubMed Central and 910 of those were covered by this run; the other 232 link elsewhere and are not checked this way.

Cochrane coverage, site-wide and by topic

What it checks. Every review in the Cochrane Database of Systematic Reviews that PubMed indexes with tinnitus in its title or abstract, collapsed to the newest version of each review, with protocols (plans for a review, containing no results) set aside. Each current review with tinnitus in its title is checked twice: is it cited anywhere on the site, and is it cited on the article about its subject.

What it cannot check. Covers Cochrane reviews only, and only those PubMed indexes with tinnitus in the title or abstract. Being cited is not the same as being reported accurately — the check confirms the link, not what the article says about the review. Which article a review belongs on is an editorial assignment kept in the script, and a new review is reported for a person to assign rather than checked automatically.

Last result. Run (the day this page was built): 12 of 12 current reviews about tinnitus cited somewhere on the site, and 12 of the 12 assigned to an article cited on that article.

Citation labels against NCBI

What it checks. Every source in an article's source list that links to PubMed or PubMed Central. The first author's surname and the year in the citation label are compared with NCBI's own record of the paper.

What it cannot check. Checks the first author and the year, not whether the paper supports the sentence it is cited for — that takes reading it. Journal names are compared and reported but never counted as errors, because NCBI and publishers write the same journal differently. Citations whose label has no author and year cannot be checked this way, and sources linked through a publisher, a DOI or another site have no NCBI record to compare against.

Last result. Run (the day this page was built): 1,210 citations compared; every first author and year agreed with NCBI.

The corrections log against the live pages

What it checks. Every entry in the public corrections log, against the page it names: the page exists and carries the title the entry gives it; the old wording is gone from the page and the new wording is on it; the old wording does not survive on its own in the title, description or FAQ that search results and feeds quote; and no entry prints a stray backslash.

What it cannot check. Whether a correction was the right call, and whether its stated reason is honest, are judgements this check does not make. Quoted wording is compared on overlapping runs of words rather than exactly, so a correction that hedges a sentence rather than replacing it can pass with much of the old wording still in place — deliberately, because hedging is a legitimate fix.

Last result. Run (the day this page was built): 117 entries checked; each points at a real page, names it correctly, and was applied.

What counts as a revision

Each article shows the date of its last substantive revision. A revision is a change a reader receives: the body text changes by at least two lines, a source is added or removed, the title or description changes, or a question or answer in the FAQ changes. Rewording a citation label, retagging an article or editing a date is not a revision, and does not move it.

The dates are checked against the site’s commit history by a script, which reports a date that moved without a change behind it, or stayed behind when one happened. What a revision date does not mean is that every claim on the page was re-checked that day. Where a page has been checked, the check is named and dated on its own.

Also checked, with no result published here yet

  • Grades against their articles. Checks that every graded treatment has an article, that every figure the comparison table quotes appears in that article, and that the guideline positions are filled in consistently. It cannot say whether a grade is still right, which is a judgement.
  • Revision dates against the commit history, as described above.

Read more

  • How we rate evidence — the grading scale, which sources carry more weight, and how a grade is assigned
  • Medical review policy — why no clinician has reviewed an article, and what a review would involve
  • Corrections — every error found and fixed, with the old wording quoted