Editorial standards and corrections policy

A publication is only as good as the rules it will not break under deadline. Ours are written down, versioned, and enforced by a standards editor who does not report to the writers whose work he sends back. This page is the current version. Where an earlier version differed materially, the change is recorded at the foot.

1. Sourcing

1. Primary documents over summaries. We read the paper, not the abstract; the docket, not the news release; the notice, not the trade-press account of the notice. Where the two differ, we report the primary document and note the discrepancy.
2. The statistical analysis plan before the paper. For any trial result, we establish the pre-specified endpoints and hierarchy before reporting what was found.
3. The estimand, every time. Where a trial published more than one analysis, we state which one we are quoting. This is not optional and has no exceptions.
4. Methods with measurements. An analytical result is reported with the laboratory, the technique, the detection wavelength and, where disclosed, the integration threshold. A figure without them is reported as a claim, and labelled as one.
5. Named sources where possible, characterised sources otherwise. We do not use unnamed sources for factual assertions about named companies. Where a source cannot be named, we say why and we describe their position precisely enough for a reader to weigh it.
6. Denominators. A percentage without a denominator is not a finding. Sub-editors are instructed to reject one.

2. What we will not assert

• That a company has behaved dishonestly, unlawfully or fraudulently, unless there is a document we have read and can describe.
• That a product is safe, effective, or suitable for any person. We report measurements and published evidence; we do not certify anything.
• That an individual will respond to a treatment in a particular way. The published response distributions do not support it.
• That a supplier’s material is "verified" on the strength of a report describing a different lot.
• That an absence of evidence is evidence of absence, or the reverse.

3. The vendor dossier programme

Twenty companies are covered. Material is bought at retail, as an ordinary customer, with no disclosure to the vendor beyond a statement — made before the order — that any result may be published. Samples are posted unopened to independent laboratories and the Journal pays. Where we can obtain more than one vial from a lot, we split the sample between laboratories and publish both results without averaging them.

Every company receives specific written questions each quarter. The correspondence log records the question, the date and the outcome, without editorialising, including "no response after two follow-ups". Vendors do not see a dossier before publication, cannot review or approve it, and cannot pay for any part of it. No vendor advertises in the Journal.

How the Journal score is constructed

The score on a dossier is a letter grade, A to C, composed of four components, weighted as follows: the analytical record on material we purchased (40%), documentation quality (25%), responsiveness to written questions (20%) and fulfilment reliability (15%). It is revised only at a quarterly review and only against the record. It is not a safety assessment, not an endorsement, and not comparable with any other publication’s figure.

We publish a single grade reluctantly, because readers construct one whether or not we provide it, and a badly constructed one is worse than a stated methodology. The dossiers say so on their face.

4. Corrections

1. We do not silently edit. A published article’s text is not altered to remove an error. The erroneous sentence is corrected and a correction notice is printed with the piece, in the same typeface, stating what the earlier version said.
2. Every correction is logged. The running log at /departments/corrections/ carries every correction and editor’s note, dated, linked to the piece.
3. A correction is not a clarification. A correction means we published something untrue. A clarification means we published something true that was capable of misleading. We label them differently because the difference matters.
4. Prominence matches the error. A wrong figure in a headline is corrected in the position the headline occupied, not in a footnote.
5. Reader-originated corrections are credited. Roughly one in five of our corrections arrives from a reader.
6. Nothing is unpublished. We do not remove an article because it became inconvenient. If a piece was wrong enough to withdraw, we say so on the page and leave the page.

5. Complaints

Write to standards@compoundjournal.com. A substantiated factual challenge receives a reply and, where it is upheld, a published correction. A complaint about conduct — a writer’s behaviour, a conflict of interest, a failure to seek comment — is investigated by the standards editor and the outcome is published if the complaint is upheld. Complaints about the editor-in-chief are handled by the standards editor and reported without her involvement.

6. Advertising and independence

Advertising is labelled Advertisement or Sponsored on every appearance. Sponsors are named on the funding page, see no editorial copy before publication, and have no influence over what is covered or how. No advertiser has ever been given advance notice of coverage concerning them. Two current advertisers have been criticised in these pages while advertising in them.

7. Conflicts of interest

Nobody who writes for the Journal accepts speaking fees, consultancy, honoraria, travel or equity from a pharmaceutical manufacturer, a compounding pharmacy, a testing laboratory or a research-chemical vendor. Where a contributor has a prior relationship with an organisation we cover, it is stated on their contributor page and in the piece.

8. Anonymity and reader safety

We do not publish a reader’s full name without explicit consent. Letters are printed with an initial, a surname and a city, all verified. We do not publish information that would identify a person’s medical circumstances beyond what they have chosen to disclose, and we decline material that names a private individual as a purchaser of anything.

9. Annual accuracy audit

Each year the standards editor samples published pieces at random, re-checks every checkable assertion in them against primary sources, and publishes the error rate by category. The audit is published whether or not the result is flattering. It is the only number in this publication that we would rather not have to print.

Revision history

• Version 4, 1 July 2026 — added the scoring weights for the vendor programme, which were previously described but not quantified.
• Version 3, 14 January 2026 — added the requirement to state the integration threshold with any purity figure.
• Version 2, 2 June 2025 — added the distinction between corrections and clarifications, and the prominence rule.
• Version 1, 8 January 2024 — first publication.