Standards
The Institute

Editorial & evidence standards

An institute that writes about AI governance should publish its own. These are the rules every page, book and platform feature is held to — not aspirations; the working system, with its enforcement named.

1. Sourced, or downgraded in the sentence

Every quantitative claim carries a named, dated source. Where a claim cannot be sourced, the sentence itself says so — "unverified", "inference", "the record stops here". A reader should never have to guess which kind of sentence they are reading.

2. What is unpublished stays unpublished

Where a standard, fee schedule or survey process has not been made public — as with much of the Responsible Use of AI in Healthcare (RUAIH) certification at the time of writing — our pages say exactly that, and label everything built on top as inference. We treat honesty about the gaps as the product, not a disclaimer on it.

3. Every page carries its provenance

Reference pages carry a last-reviewed date, the reviewing author, and where useful a "how this was built" note describing sources and method. Pages past their review interval are re-verified against their sources before the date moves — the date is a promise, not a decoration.

4. Composites are labelled composites

Case material drawn from multiple engagements is labelled composite wherever it appears. No invented clients, no invented testimonials, no implied adoption numbers. Reader reviews shown on this site are excerpts from public retail listings, attributed, reproduced without editing beyond truncation.

5. Independence, in writing

No vendor sponsorship, no paid placements, no affiliate commissions, no vendor equity, no referral fees. Vendor assessments score against a published method; where the Institute has a product of its own in a category, the conflict is disclosed on the page.

6. Automated gates, human sign-off

An automated pre-flight system checks every book manuscript for placeholders, structural defects, figure quality and print specification; the site's build blocks publication on schema violations, broken commercial links, or missing provenance fields. A failing gate stops a release. Passing gates does not publish anything — a human decision does.

7. Corrections

Errors are corrected in the next printing or the next deploy, and material corrections are noted on the page. Write to [email protected]. Corrections from working operators are read first and answered personally.

Why publish this? Because the Institute's whole argument — in the books, the reference, and the platform — is that trustworthy systems are ones whose rules are written down, dated, and enforced by something sturdier than good intentions. This page is that argument, applied to ourselves.