Data, video and the corpus

The recording is the article's primary data. What may be done with it beyond reading the article is governed by a separate consent, and that separation is deliberate.

The article and its video

The article, including the video, is published under CC BY 4.0. Anyone may view, download, redistribute and build on it with attribution. That is the licence on the published work.

The corpus is a separate thing

Annotated endoscopic video is scarce and valuable for training and for machine learning. This journal therefore maintains a corpus of de-identified recordings with their structured clinical annotations. Inclusion requires two conditions, both hard gates in the platform:

• the patient's consent explicitly extends to commercial sublicensing of the de-identified video, and
• the de-identification review has been counter-signed by the editorial office.

Either the patient's consent scope excludes commercial sublicensing, or the de-identification plan has not been counter-signed by the editorial office. Both are hard gates: the article publishes normally, the recording never enters a licensed corpus.

Authors and patients may decline corpus inclusion. Declining has no effect whatsoever on whether the article is accepted, how it is reviewed, or how it is presented — the article publishes normally and the recording simply never enters the corpus. This is stated plainly because a journal that quietly disadvantages authors who decline is running a consent process in name only.

What the annotations contain

Each eligible recording carries the article's coded clinical fields as record-level labels — procedure, indication, intervention, operator grade, optical modes, preparation score — and the author's marked procedure steps as temporal annotations with start and end times. The corpus record is published in the MLCommons Croissant format, so a dataset consumer reads the same structure they would read from any other Croissant dataset rather than a bespoke schema.

Chapter tracks

Every published video carries a WebVTT chapter track generated from the author's marked steps. It is served alongside the video and available as a file at /articles/<number>.vtt. The chapter list on the article page seeks the player directly, and each chapter is also addressable as a media fragment for citation.

Metadata

Article metadata is open. It is deposited with Crossref, exposed over OAI-PMH at /oai in both Dublin Core and full JATS, offered as an Atom feed, and available per-article as JATS XML, JSON, BibTeX, RIS and EndNote. No registration and no API key.