Disclosure

AI assistance

Where AI is used to produce this publication, what it is not permitted to do, and what stops it publishing something wrong.

Most articles here are drafted by an AI model and published without a human reading them first. That sentence is the point of this page: you should know it before you decide how much to trust what you read, rather than finding it in a footer somewhere.

What follows is what that actually means in practice.

What the AI does

Triage. Decides whether an item from a monitored feed describes a real change a developer would need to act on. Most do not, and are rejected here.
Duplicate detection. Decides whether two headlines describe the same development, so several outlets covering one announcement produce one article rather than four.
Fact extraction. Reads the primary source page and pulls out claims, each with the exact sentence from the source that supports it.
Drafting. Writes the article from that extracted fact set.
Review. A separate pass checks the draft against the facts for anything unsupported or contradictory, and a further pass judges whether the article is actually useful.

What the AI is not permitted to do

These are enforced in the prompt and then checked again mechanically, because an instruction to a model is guidance, not a control.

It never sees the source article's prose while drafting. The drafting step receives only the extracted fact set. It cannot paraphrase text it was not shown, which is both why the output is a synthesis and why the fact check means anything.
It cannot invent a quotation. Every quoted string in a draft is matched against the verbatim text we fetched from the source. A quotation that is not found there blocks publication outright.
It cannot invent a link. Any external URL in a draft whose domain is not among the researched sources blocks publication.
It cannot invent a version, date, price or capability. Facts are only accepted when the supporting sentence is found in the fetched page. Facts that fail that check do not count toward the article's confidence.
It does not decide whether to publish. That is arithmetic (below), not a model's opinion.

What decides whether an article goes live

Each draft gets a confidence score between 0 and 1, weighted across seven components: whether a primary source exists, how many independent sources corroborate it, what share of claims are backed by verified evidence, how specific the article is, the fact-consistency verdict, the quality verdict, and how original it is against what we have already published.

An article publishes automatically only when it scores at least 0.78 and trips none of the hard blocks. Any one of these holds it back regardless of score:

no primary source
a claim that contradicts the extracted evidence
a quotation not found verbatim in a source
a link to a domain that was not researched
too similar to something already published
too short, or with no usable structure

Anything that falls short is held for a person to look at, or rejected. It does not quietly publish at a lower standard.

What a human does

A person sets the thresholds, chooses and rates the sources, writes the editorial rules, reviews everything held back, and can edit, unpublish or lock any article at any time. Articles in Field notes and Guides are written from firsthand experience of shipping the Ali3N Studios apps — the AI helps structure them, but the experience and every factual claim in them is a person's.

When we get it wrong

Automation of this kind will eventually publish something inaccurate. When it does, the correction is made on the article itself with a note saying what changed — not silently edited away. If you spot an error, tell us and it gets fixed.

What we will not do

Republish or rewrite someone else's article. Sources are read for facts and cited prominently; their prose stays theirs.
Invent an author. Everything here is published by Ali3N Studios LLC. There are no fake bylines and no fake headshots.
Publish for volume. The system has a daily publishing cap and a minimum gap between posts precisely so that "more" never becomes the goal.