Editorial standards

What you can expect from a page here

Six standards every page is held to, and how to check each one for yourself on the page in front of you. If a page fails any of them, it is a fault worth reporting rather than a limitation to work around.

594 citations·62 state a version scope·105 carry a review date
01

Every factual claim names its source

A page is written from primary material: the project documentation, the specification, the regulation, the paper. Not from a summary of it, and not from recollection. Defaults, version numbers, dates and thresholds are quoted from the source and can be checked against it.

How to check it: Scroll to the foot of any page. The sources are listed there with publisher and link. Numbers in the text carry a citation or a link on the same line.

Where two sources disagree, both are reported rather than one being chosen silently. The Transformer paper explainer cites conflicting BLEU figures from arXiv and the NIPS proceedings for exactly this reason.

02

Every claim is tied to a version

A statement about software is only meaningful against a version. A default that is true in PostgreSQL 18 may not be true in 14, and advice written against an unnamed release is advice you cannot act on safely.

How to check it: The bar under the title states which product versions the page applies to. Technology pages also carry a release table showing which versions still receive fixes and which are past end of life.

Release data is generated from published support information rather than typed by hand, so the table on a page and the site-wide timeline cannot disagree.

03

Every page states when it was last checked

Technical detail ages. Rather than leaving you to guess whether a page is current, each one records the date it was last verified against its sources and how often it is due to be checked again. Fast-moving subjects such as pricing are on a ninety-day cycle; stable mechanics are on an annual one.

How to check it: The status line under the title shows when the page was last reviewed. Pages that are overdue against their cadence are listed openly on the what-changed page.

The review date is deliberately separate from the last-edited date, because fixing a typo re-checks nothing. The status line combines them into one date only while both fall in the same month, and shows them apart as soon as they diverge.

04

Fact, analysis and projection are labelled differently

Mixing a verifiable statement with an opinion about what it means is the most common way technical writing misleads without stating anything false. Each page declares which of the three it is, and pages that are not straight fact carry a visible badge.

How to check it: Look for the badge above the content. Its absence means the page is factual reference material.

05

What is not known is said, not filled in

Where something could not be verified it is removed rather than hedged with cautious phrasing. Where a subject is not covered, the page says so instead of producing filler. Where a source is paywalled or unreachable, the page states that and describes only what can be established.

How to check it: Technologies with no guide still appear in dependency diagrams, shown without a link, so a gap in coverage is visible rather than hidden.

The finance pages are shorter than they were because eighteen figures with no traceable source were removed rather than restated with qualifiers.

06

Derived material carries its original licence

Where a page is built on openly licensed material, that licence is named on the page and the derived text is published under the same terms. Material under a confidential or proprietary licence is not used at all, however useful it would be.

How to check it: Pages derived from openly licensed sources display the licence beside their source list.

Reading the labels

Three kinds of statement appear here, and they carry different weight.

Fact

Statements that can be checked against a cited source: defaults, version numbers, dates, thresholds, what a specification requires.

A source is listed. If a figure has no source behind it, it should not be on the page at all.

Analysis

Interpretation. What a fact implies, which option suits which situation, what a change is likely to cost you.

The underlying facts are sourced, but the conclusion is a judgement. Someone with different constraints could reasonably reach another.

Projection

Expectation about what has not happened yet.

Treat as opinion. It is labelled separately precisely so it is not mistaken for the first two.

What this does not promise

None of this makes a page correct. It makes a page checkable: sourced, scoped to a version, dated, and consistent with the rest of the site. Those are different things, and the difference matters.

A source can be wrong. Documentation goes out of date between a page's review date and the day you read it. A summary can lose the detail that mattered for your case. Verify anything load-bearing against the current official documentation before you rely on it, and read thedisclaimer, which is not boilerplate.

If a page here is wrong, out of date, or missing a source, that is worth reporting. Write tometaintelligencesystems@gmail.comand it will be corrected. Corrections that change what a page concludes are recorded on the page itself and collected on what changed, so a reader returning after a correction can see what moved.