Editorial standards

How articles are produced

Articles on UltraInstinct are drafted with the help of a large language model and then put through an automated editorial gate before anything is published. We say this plainly because you deserve to know how what you are reading came to exist, and because a technical audience can usually tell anyway.

The gate rejects a draft outright if it fails any of the following:

• Length below the editorial floor, or structure missing an introduction, section headings, a conclusion, or a references list.
• More than 40% title overlap or 80% body overlap with anything already published here, which is how we stop the same subject being covered twice under different words.
• Fewer than two contextual links to related material on this site.
• Any non-English text, or a missing or invalid lead image.

When the language model fails to produce something that passes, nothing is published that day. The pipeline explicitly refuses to fall back to generic filler.

What we measure ourselves

An explainer restates knowledge that already exists. To be worth your time, this site also has to produce something that does not exist anywhere else, so we run benchmarks on the server that hosts this site and publish the numbers along with the complete source code, the exact method, and the machine specification.

You can disagree with a method, and you can run any of them yourself: see the benchmarks.

Corrections

Technical writing gets things wrong. If you find an error — a misstated mechanism, a wrong number, a claim that does not survive contact with the source — tell us through the contact page and we will correct the article and say what changed.

We publish no sponsored articles and no paid product placements. Where advertising appears, it is clearly marked as advertising and has no influence on what we write.