Skip to content
GMedACJournalGMedACJournal
SearchGMedAC

Legal and regulatory content policy

How this publication handles jurisdiction-specific legal and regulatory claims, and what the build refuses to publish

The failure this policy exists to prevent has a name: describing preliminary research as legal approval. A jurisdiction note that was read once, summarised honestly and never put to counsel looks, on a rendered page, exactly like a settled position. Prose cannot carry that difference reliably — a hedge is the first thing lost in a tidying pass, and nobody notices it go. So the difference is carried in structured fields: a registered claim whose declared basis does not support it does not become a page. That control reaches the claims an author registered and nothing else, which is a limit stated in full at the end of this document.

What counts as a jurisdictional claim

Any assertion that an obligation, restriction, registration requirement or eligibility condition holds in a named place, for a named kind of professional, for a named kind of service. Softening the geography does not exempt the sentence. "Most European regulators take this view" is a jurisdictional claim in vaguer clothing and is handled as one, because generalising across a border is the commonest way an unsupported claim enters a draft.

What every jurisdictional claim declares

Each entry in the article's claims list carries, and the check requires:

  • The jurisdiction the rule is asserted to hold in. For the United States this is never the country. It is a state, and a claim written at country level is refused rather than quietly narrowed by a reviewer's guess.
  • The professional category it applies to. A rule for physicians is not a rule for psychologists, and inheriting one from the other is an inference, not a reading.
  • The service type. Remote consultation, in-person consultation and asynchronous advice are separate questions, and the answer to one is not the answer to another.
  • At least one entry in sourceIds resolving to a real source record.
  • The effective date of the instrument where it is known, and an explicit statement that it is not known where it is not. An undated rule is a rule of unknown vintage.
  • dateRetrieved and retrievalRoute on every source behind it.
  • validityAsOf on the article, with nextReviewAt.
  • basis — whether this is primary-source fact, reasoned inference, or something waiting on counsel.

Which bases reach a page

PRIMARY_SOURCE_FACT, REGULATOR_GUIDANCE and PROFESSIONAL_BODY_POSITION may support a published jurisdictional claim. REASONED_INFERENCE, COUNSEL_CONFIRMATION_REQUIRED and UNKNOWN may not. An inference can be careful and correct, and it is still an inference — printing it in the same typeface as a statute converts this publication's own reasoning into apparent authority. The check reads the basis on every claim declaring a jurisdiction and fails the build on the last three, rather than waiting for a reviewer to notice at the end of a long document. A policy that depends on remembering is not a control.

Sources, and what counts as having read one

LEGISLATION, REGULATOR_PUBLICATION and OFFICIAL_REGISTER sit above every other source type for this material, and a HIGH or CRITICAL article may rest only on sources at or above PEER_REVIEWED_STUDY — which puts INSTITUTIONAL_SUMMARY, PROFESSIONAL_BODY_POSITION, JOURNALISM and COMPANY_STATEMENT out of reach for exactly the articles most likely to be quoted back. A professional body's position can still stand behind a claim as a basis; source type and claim basis are separate taxonomies, and satisfying one does not satisfy the other. Every source record requires retrievalRoute in prose: how the document was actually opened and read, not that a request returned. A successful HTTP status is not proof the source was read.

Nothing published here is legal advice, and nothing here is a substitute for a professional's own regulator or their own adviser. This publication describes what was read, when it was read and from where. It does not tell a professional what they may do, and a sentence that starts to read that way is a defect, not a style preference.

The same restraint applies to verification. A verification badge on the marketplace is not permission to practise and not permission to book. It is a statement that somebody checked a document, published as a summary of what was checked, never as an endorsement. Whether verifying and publishing makes GMedAC an endorser rather than an intermediary is an open question carrying COUNSEL_CONFIRMATION_REQUIRED, and this publication does not resolve it by writing confidently in either direction.

The corridor position, stated plainly

No GMedAC launch corridor is approved. Every gate in the Czech launch matrix is open: three RESEARCHED_PENDING_COUNSEL, seven COUNSEL_QUESTION_OPEN, two EVIDENCE_MISSING, none approved. EVIDENCE_MISSING is not a clearance. Excluded means unresearched. It never means permitted. The United States is never one corridor; it resolves state by state, and what the marketplace holds on it is corridor discovery — a shortlist of candidate states with their sources recorded — which is expressly not a permission. Its own note on the clearest candidate reads that a registration route existing is a pathway's existence, not a permission, and every US position remains counsel-required per state. The marketplace itself is at Phase 0 — no checkout, no live payment provider, no deployment — so an article implying that a place is served, or a launch near, contradicts the record before any legal question arises.

Setting publicationStatus: LEGAL_REVIEW does not create a review, and editing it to READY does not complete one. What the build can test is structural: whether reviewerIds names a reviewer record at all. Whether that person holds standing in the declared jurisdiction is asserted by an editor and weighed by a human; nothing here checks a credential against a register. So an article carrying a jurisdictional claim with no named reviewer fails whatever its status says, while one reviewed by the wrong person passes the build and rests entirely on the editor who named them.

When a published legal claim turns out to be wrong, the route is a correction file recording previousClaim verbatim beside correctedClaim, with reason, editorId, reviewerId, material and dateModifiedChanged. A material: true correction with dateModifiedChanged: false is refused, and so is a material correction naming no reviewer. Those rules constrain the correction file; they cannot notice an edit that was never recorded as a correction at all. Not rewriting a published sentence in silence is a commitment kept by editors, not a thing the build enforces.

Where this control is weak

The check sees claims that were registered as claims. A legal assertion written as ordinary prose and never entered in claims is invisible to it and depends entirely on a reviewer reading carefully. Freshness is uneven on purpose: a stale HIGH or CRITICAL article fails the build, and so does any stale article carrying emergencyInformation: true, but LOW and MODERATE staleness is a warning and an editorial-queue item, because failing a build over an explainer a week past its date teaches people to push the date rather than reread the article. And the privacy questions around this material are open in the same way the corridor questions are: whether a booking pairing is Article 9 data under GDPR is UNKNOWN and awaits counsel, and the absence of clinical fields does not answer it. This is honest, not clean.