How this is made

The document

The reference was assembled from five separately researched chapters — history and nature, mechanism, epidemiology, types and diagnosis, treatment, citizen science, and the research frontier. Assembly was editorial only: material was rearranged, deduplicated and smoothed, but nothing factual was added, altered or resolved in the process.

Three conventions carried through from the research and still govern the text:

Every claim is tagged with its provenance. The tags are described in the important notice. Community reports are never discarded for being anecdotal — they are labelled honestly and kept, because patterns in anecdote are worth recording.

Conflicts are shown, not settled. Where two sources disagree, both are kept and the disagreement is stated. “Unknown” and “contested” are treated as valid final answers where that is what the evidence supports. The cross-chapter contradictions are collected in Unresolved Conflicts rather than adjudicated.

Errors spotted during assembly were left in place and logged. Anything that looked like a research or transcription error was not silently fixed; it is listed in Editor’s Queries so the correction is visible.

The update cycle

A weekly research pass monitors the literature, trial registries, regulator decisions and patient-community sources. It feeds a monthly revision in which material is added, corrected or removed. The Watchlist is the live monitoring table — the specific things being tracked and roughly when they are expected — and the Changelog records what actually changed.

Questions the document cannot answer are collected rather than guessed at. When the assistant on this site meets a question that is not covered by the text, it offers to add it to the queue for the next research pass.

Sourcing beyond English

The research explicitly sought non-Anglophone sources — German, Japanese, Danish, Italian, Chinese, Korean and Spanish-language literature and guidelines — because several conclusions change when you look outside English. Oxygen access policy is the clearest example: the strongest positions belong to Germany and Japan, and a review confined to English-language sources understates that. Where language bias in the underlying literature is measurable, the document says so.

The site

Everything on this site is generated from a single markdown file, which is the canonical version of the document. The build splits it into pages, generates the navigation and the full contents, builds the search index and the knowledge graph, prepares the retrieval index that the assistant answers from, and produces the PDF, Word and Markdown downloads.

That means there is exactly one place to edit and no possibility of the site and the document drifting apart. A monthly revision is a single edit followed by a rebuild.

Two things are rewritten during the build, both navigational rather than editorial. Footnote markers become live links into the relevant reference list. Section pointers of the form §x.y become links where the target resolves unambiguously — and are deliberately left as plain text where it does not, because the document discloses that some of its own pointers reflect an earlier numbering, and a confidently wrong link is worse than no link.

Reading it

The document is long. Some suggested ways in:

• The executive summary covers the whole argument in about five minutes.
• Search works across the full text, including tables.
• The knowledge graph is the best way to see how treatments, mechanisms and trials connect to each other.
• The assistant on the ask page answers from the text and links to the passage it used.