A mythology encyclopedia used to be judged by how much it contained. That test has stopped meaning very much. The web is filling up with mythology pages produced at industrial speed, thousands at a time, most of them confident, cleanly formatted, and wrong in ways a casual reader has no way to see. Volume has become the easiest thing in the world to fake.
So we would rather be judged by something harder to fake. This page is an account of how an entry earns its place on Gods & Monsters, what we refuse to publish, and, at the bottom, the things we already know are still wrong. A site that is honest about the limits of the record is worth more than one that pretends the record is complete, and that honesty is only worth anything if it is specific. We have tried to make it specific.
How an entry earns its place
Every listing here is built the same way: research first, writing second. The research has to clear a short set of tests before a single sentence of the article gets written. Each of those tests exists because we once got something wrong and had to go back and undo it. They are worth spelling out, because the mistakes they guard against are the same ones filling the rest of the web.
We are never our own source
The most common way a modern reference site fools itself is by quoting itself. A page goes up. Other sites, many of them automated, copy it or reword it. A search then returns several “independent” sources that all trace back to the same origin, which is the original page. The claim looks corroborated. It is only an echo.
We learned this the hard way. A routine check once turned up entries on this site, presented as genuine deities of a specific tradition, that on inspection existed nowhere at all except our own pages and the content mills that had rewritten them. They were not obscure or thinly attested. They were invented, and we were the ones who had invented them, and they had been live for months before anyone asked the right question. We audited every candidate page on the site, removed the fabrications, and traced them back out of the larger overview pages they had seeped into. The rule now is absolute: our own site, and anything that merely repeats it, counts for nothing when we are deciding whether a figure is real.
We count sources by their roots, not their number
Two sources are only two sources when they are genuinely independent, and it is surprisingly easy to be fooled here as well. A reference book states something; a popular wiki cites that same book; a search summary paraphrases the wiki. That reads like three sources. It is one source wearing three coats. We once accepted a figure on exactly that basis and later found that the single writer underneath the whole stack had been describing something the tradition’s actual recorders never mentioned at all.
So before a source is counted, we follow it back to where it begins. A general encyclopedia or a popular wiki is a place to start looking, never proof on its own. If everything traces to one root, we have one source, and one source is not enough to write from.
A close cousin of this problem is the modern invention dressed as ancient myth. A striking number of “old” figures were actually coined by one identifiable person in the modern era: a national poet inventing a war god in the nineteenth century, a folklorist reading a pagan fertility cult into medieval church carvings in the twentieth, a beloved American tall-tale animal that started life as a lumber company’s advertising. None of that makes them less interesting. It makes them modern, and when a figure’s trail ends at one named author and date, we say so and name them rather than pretend the trail runs back to antiquity.
We check the culture, not just the name
Filing a figure under the wrong people is not a tagging slip. It puts words in a living tradition’s mouth. This is easy to get wrong at scale and hard to catch, especially when the sorting is done by a machine matching text instead of a person who knows the material. One classification pass we ran keyed on the letters in a name rather than its meaning, and quietly filed a figure from a Chinese classic under an unrelated African tradition, and an ancient Egyptian god under Chinese, purely because each name happened to contain the right handful of letters. Mistakes like that look perfectly reasonable in a spreadsheet. They are visible only to someone who already knows the answer, which is why a person who knows the material reads the classification instead of trusting it.
We verify the being, not the translation
Some of the most convincing errors come from translating a name into a plausible god. A great many mythological names are really just descriptive phrases: owner of the forest, lord of the sun, mother of the waters. Translate the words and you get something that sounds exactly like a deity, and it is tempting to hang a real myth on it. But the phrase is very often only a by-name for a figure who already exists under a different name.
We carried one such listing, well written and full of specific detail, that presented a “lord of the sun” as a major sky god with a complete myth cycle. The phrase was real. It was a regional by-name for a minor forest trickster, and the grand solar myth attached to it belonged to an entirely different figure. Someone had translated the words, produced a convincing god, and bolted a borrowed story onto it, and every individual step had looked like careful research. The detail we kept from that case, because it is the whole lesson: our own site had already recorded the correct version months earlier, buried in an answer to a reader’s question that nobody had gone back to read. The encyclopedia had caught its own mistake and been ignored.
What we refuse to publish
These tests mean we regularly decide not to write something at all. When we research a figure and find the record simply is not there, that the entire modern portrait rests on one unread book or a single compiler’s paragraph, we stop and say so rather than pad it into a persuasive article. Those non-entries never appear on the site, which is exactly what makes them hard to point to, so it is worth stating plainly: a standard that never costs anything is not a standard. We turn work away at this stage often, and we would rather keep a smaller encyclopedia that is right than a larger one that is quietly guessing.
Where the machines fit, and where they do not
It is an editorial decision of this site to be open about how it is made, so here it is plainly: we use artificial intelligence throughout this project. Software helps us research candidate figures, drafts entries in our house structure, produces the artwork, and runs the routine checks that keep a catalogue of more than two thousand listings consistent. We think a page devoted to honesty would sit badly on a site that was coy about that.
What the machines do not supply is the part this whole page is about. Every test described above was designed by a person, after a failure a person caught. The classifications are read by someone who knows the material instead of being trusted. Entries are reviewed before they are published, and every removal, correction, and refusal on this page was a human decision, made one at a time. Thousands of human hours have gone into building this encyclopedia and into the slower, less visible work of standing behind it — tracing sources back to their roots, tearing out fabrications, turning away figures the record cannot support, and answering the readers who write in. Drafting turned out to be the easy part. The hours went into the judgment, and the judgment is ours.
You may notice that several of the failures admitted on this page were made by automation and caught by a person. That is the honest shape of the partnership. A tool that works at industrial speed can be wrong at industrial speed, which is exactly why the tests above exist and why a person enforces them. Our newsletter carries this same disclosure, for the same reason: we would rather be trusted for what we actually do than admired for something we are not doing.
When the machinery fails
Not every failure is about research. Some are about the tools. We once identified a batch of entries that had leaned on an unreliable automated source, rewrote every one of them properly, and moved on satisfied, before noticing that a quirk in our own publishing process meant not one of the corrected versions had actually reached the live site. The repair had reported complete success and changed nothing. That is worse than doing nothing, because it hides the problem behind a green light. We fixed the process, and we mention it here because this is precisely the kind of failure a site is least likely to admit to and most likely to repeat.
What is still wrong
This section exists because almost no one publishes one, and that is exactly why it is worth publishing. These are open problems we know about right now:
- Some of our internal planning data is still incomplete, which means parts of the process cannot yet be fully reviewed by eye before anything is written.
- A few traditions in our queue rest on a single contested source from more than a century ago. We have flagged them, and if our own process ever reports them back as solidly attested, we will treat that as a sign the process failed rather than as good news.
- For at least one past error, we do not know exactly how long it was live before we caught it. The honest answer is “months,” and we would rather say that than invent a precise number.
We will update this list as we close items and, inevitably, as we find new ones. Quietly dropping something from it without fixing it would defeat the entire purpose of keeping it.
A standing invitation
None of this makes us immune to being wrong. It makes us a site that expects to be wrong sometimes, checks accordingly, and tells you when it happens. If you find an error here, and especially if you have reason to think a figure was never real, we want to hear about it. Getting a living tradition right matters more than getting it fast, and the record is always better for a reader who knows the field and takes the time to write in.