About · method
Method
The archive makes one claim: every line is sourced. This page is the account of what
that means in practice — the standard a parallel clears before it appears, the things
this archive will not do, where the collection is thin, and how to tell it that it is
wrong.
34 parallels · 68 checks · 47 corrections published
One person edits this archive. The parallels were researched with AI assistance and
each was read back against its sources by two independent fact-checks before it was
admitted; the checkers' notes are published unedited beside the claims they correct,
including where they narrow them. Where that process failed, the fault is editorial.
01 · how a parallel gets in
Both sides quoted, every source openable, every correction kept.
The 34 parallels were assembled along seven research angles — flood,
serpent, fire, the gap, names, transmission, and one modern case. They are not summaries
of a resemblance. Each names a specific edition on both sides, quotes the passage it turns
on, and carries the checks that were run against it.
01 Both sides quoted, not paraphrased. Every parallel names a work and an
edition on each side — Kovacs's Gilgamesh, Eggeling's Śatapatha Brāhmaṇa, the
King James Genesis — and prints the passage it turns on. A reader who disagrees
can see the words the claim was built from before reading the claim.
02 Every source openable. The 34 parallels carry 86 source references
between them, two to four each, across 28 hosts: Project Gutenberg,
the Internet Archive, sacred-texts, Perseus, Wikisource, ctext, Avesta.org.
85 are links that leave this site. The remaining one is a print citation
with no scan to point at, and is printed as
a citation rather than dressed as
a link.
03 Two adversarial checks, or it does not appear. Each parallel was read back
against its sources twice, independently, by checks briefed to break it — 68
in all. Only double-confirmed parallels were admitted, which is why the file
that holds them records a verdict for every one.
04 Corrections recorded against the claim. Those checks returned 47
corrections across 33 of the 34 parallels. Each is printed on
that parallel's own page, verbatim, above the sources and below the evidence —
including where a checker narrows the claim the page is making. The one parallel
that came back with nothing to correct
says so in the same place.
What this standard establishes is narrow, and worth saying plainly: that both passages
say what the page says they say. It does not establish that the two texts are related.
Borrowing, shared inheritance, coincidence, or a common shape in how minds tell
stories — that is the argument the six doctrines are having, and the
archive does not settle it.
20 of the 34 parallels cite a text you can open here in full; the
rest cite chapters, single hymns, and editions the library does not hold complete, and
link out to a scan instead.
02 · what the archive will not do
Four refusals, and what each one costs.
01 It publishes no number it cannot count. Every figure on this site is
derived from the catalogue or the canon when the page is built: 324 works,
20,789,616 words, 34 parallels, 28 findings, 18
records, 9 open questions, 6 doctrines. There are
no reader counts, no view counts, and no totals counted before the thing exists.
02 It does not withhold evidence to make you earn it. Every work is readable
in full, immediately, with no account and no wait. Every parallel shows both
quoted passages, the evidence, the corrections, and the sources on one page.
Nothing here is held back behind a sign-in or a level.
03 The composed records say so on their own plates. Eighteen records
are written for this archive: their objects and provenance are invented. Each one
says that on its own page, in the line above its text rather than in a metadata
row a reader can skim past. The traditions those records read and the parallels
they turn on are real, and both are one link away.
04 It endorses no doctrine. Six doctrines disagree about why the
parallels exist. The archive gives each the same room and takes no side. The
classification interview tells you where you fall among them; it does not tell you
which one is right, because the archive does not know.
03 · where this collection is thin
An honest count of the shelves.
The catalogue holds 324 complete works under 25 tradition labels. It
is not a balanced sample of world myth. It is a sample of what English-language
public-domain digitization has made available — every work here is a Project Gutenberg
text — and that shapes the collection far more than any editorial intent. The full
distribution, largest shelf first:
9 Indigenous North American
3% 1 Australian Aboriginal
<1% Read down that column and the gaps are the interesting part. Six of them are worth
naming, because a reader filtering the library by tradition will meet each one.
01 Two labels name no tradition at all. "Folklore" and "World mythology" hold
59 works between them, 18% of the catalogue, and neither is a
tradition. The King James Bible and The Legends of the Jews are in the second of
them, because there is no Hebrew shelf and no Jewish shelf — though Hebrew and
Jewish sources account for 9 of the 68 sides of the
34 parallels.
02 Nine works were shelved under the wrong continent, and have
been moved. They were filed as "Vedic · Hindu" — the Iroquois, the Sioux, the
Algonquin, Old Indian Legends, The Song of Hiawatha among them — because the
label sorted on the word "Indian". They now stand under "Indigenous North
American", a shelf of nine, and "Vedic · Hindu" is down to
ten works that belong there. The correction does not close the gap
underneath it: all nine are collections made by outsiders, and
the sources the parallels actually rest on — Mooney's Cherokee myths, cited by
three, and Swanton's Tlingit texts, cited by
two — are still not held here.
03 One tradition is cited and entirely absent. A parallel turns on the
Zoroastrian Vendidad in Darmesteter's translation. The catalogue holds no
Zoroastrian text. The seven works under "Persian" are literary — the
Shah Nameh, the Rubaiyat, Persian fairy tales — and none of them is Avestan.
04 Australia has one book. Parker's Australian Legendary
Tales was filed under "Folklore" and now has its own shelf, which is a more
honest label for a shelf of one. Two parallels cite
Aboriginal Australian sources, Euahlayi and Yolngu; the Yolngu source one of
them rests on is a 1937 monograph that is not public domain and so cannot be
held here at all.
05 Fourteen works stand for the whole African continent — fewer than
Chinese, Celtic, or Roman. All fourteen are collections of folk tales
and customs, or literary retellings of them; the catalogue holds no African epic
and no African cosmology. Three
of the fourteen are African-American collections from the United States
South: Uncle Remus, Bypaths in Dixie, and The Conjure Woman.
06 The Kalevala was shelved twice, once under "Celtic". Volume 01 of the poem
sat on the Celtic shelf while the complete Kalevala stood alone under "Finnish".
Both are Finnish now, a shelf of two — but they are still two
copies of one poem, and the catalogue counts them as two works.
None of this is a plan. It is what the collection is today, stated so that a reader
does not have to discover it by being misled. Where a label is wrong here it is wrong
on the shelf too, and the shelf is the thing to fix.
04 · corrections
How to report an error.
The archive expects to be wrong somewhere. A library whose whole claim is that every
line is sourced has to be easier to correct than it is to doubt.
01 Where it goes. Requests is the channel — choose "bug" and say what is
wrong. It reaches the editor. Signed in, you can watch the status of your own
reports on the same page: new, seen, planned, done, declined. Signed out, you
can leave an email that is used only for a reply and is shown nowhere.
02 What to include. The page, the line, and what the source actually says. A
correction that names an edition and a locator can be checked in minutes; one
that says a page is wrong cannot. Every record body block and every connection
carries a permanent locator for exactly this — an address, once issued, never
renumbers.
03 What happens to it. A correction that lands is recorded in that page's own
corrections block, where the fact-checkers' notes already sit, and stays there.
The claim is corrected in the open rather than quietly rewritten into agreement.
04 There is no published email address. The archive does not print one, and
it runs no public issue tracker. Requests is the whole channel, and it is read.