How it’s written
The prose on this site is written by a language model, not by a person at a keyboard. We say so plainly, because a site whose whole promise is a source beside every claim owes you the truth about its own method.
Grounded, not generated
The model does not write from what it happens to remember. For every myth, the pipeline first searches the corpus, a library of mythology scholarship held and read directly, and hands the model only the passages it found. The model retells what those passages say, and nothing else. A fact with no passage behind it does not get written, however plausible it sounds. This is the same rule the whole project runs on: memory is not a source.
What the machine checks
Before a page is allowed to ship, automated gates read it. Every myth must resolve to a real corpus book and a primary text, or it is held back, not published. The house voice is enforced by rule. A narrative that comes out too thin is sent back to be filled from its own passages, never padded. Citations must point at a tradition the figure actually belongs to, so a Greek myth cannot end up quoting the wrong shelf of the library.
What people do, and do not do
People set the rules, choose which figures to build, decide what counts as a source, review the work in samples, and fix every error a reader reports. What people do not do is hand-write or line-edit every page. At the scale this reference is built, that would be a promise we could not keep, so we do not make it. The guardrail is not a human reading each line. It is that no line can cite a source that is not really there.
So you can check it yourself
Because the writing is machine work, the sourcing has to carry the trust, and it is all on the page: the primary text by chapter or stanza, the corpus book that holds it, the scholar behind a disputed reading. If a claim here cannot be checked, that is a fault worth reporting. The method and corrections pages carry the rest.