How this is made
This site is produced by an automated pipeline. It is worth saying so plainly, up front, rather than leaving you to work it out.
What is automated
Articles are drafted by a language model working from a source — a news story, a film’s production record, an album’s actual track list. Every draft passes through a chain of checks before it can be published: a structure check, and a fidelity check that compares every name, number and quotation against the source material and rejects anything that is not there.
A second model acts as editor. It reads each draft against the same source, sends back corrections, and rejects what cannot be fixed. A fair share of drafts do not survive it.
What is not
The judgments are mine. The scores, the rankings, which films and records and games get covered at all, the house favourites scored above their critical consensus — those are editorial decisions made by a person, and the pipeline carries them out.
When a review scores something well above the critical average, the piece says so and names the wider figure. It is our opinion, offered as one.
The limits, honestly
The model’s knowledge ends before some of what is covered here. Reviews of films currently in cinemas, or records released this year, are written from fetched source material and are explicit that they survey the reception rather than claim a viewing. Where the source does not explain something, the article says so instead of filling the gap.
It still gets things wrong. When it does, the correction is made to the article, and usually to the pipeline as well, so the same mistake is harder to make twice.
Who is behind it
Mitch. I stream at The Mitchery, where a fair amount of this gets built on camera. The premise of the show is the same as the premise here: describe the thing, let the machine build it, find out whether it works.