Ask a straightforward question — who were Aphrodite's parents, how did Baldr die, which god made the world — and the honest answer is frequently "depends which text you open, and the texts are not reconcilable." Readers usually take this as a failure of scholarship. It is the opposite. The disagreements are the best-preserved evidence we have of what these traditions were actually like before anyone tried to codify them.
A mythology with no internal contradictions is a mythology that has been through an editor. The ones that reach us messy reached us less filtered.
Four mechanisms that manufacture variants
Contradiction in myth is not random. It is produced by identifiable processes, and once you can name the process you can usually work out which version is doing what.
Performance variation. A story told aloud is re-made every time it is told. Length is adjusted to the audience's patience, details are localised to flatter the people in the room, and a singer who forgets a passage improvises a bridge that may survive into the next telling. This produces variation with no ill intent and no original: there was never a first correct version that later ones corrupted.
Regional cult. Gods were local before they were literary. The goddess worshipped at one sanctuary and the goddess of the same name a hundred miles away could have different parents, different consorts, different sacred animals and genuinely different characters, because they were, historically, different deities who acquired a shared name. Panhellenic poetry then had to weld these into one biography, and the welding seams are visible as contradictions.
Political editing. When a city rises, its god rises. Someone rewrites the succession so that the local deity is the one who did the important thing, and the older account is demoted to a variant or dropped. This is the most common single cause of large-scale change in a pantheon, and it usually leaves fossils behind: an obsolete epithet, a festival whose name no longer matches its patron, a god with prestige nobody can account for from the current stories.
Scribal choice. By the time material reaches a manuscript, a copyist is choosing between exemplars, sometimes silently. Editors of ancient texts work with a rough principle that the harder reading is likelier to be original, on the reasoning that scribes smooth difficulties rather than create them. Applied to myth, the same instinct is a useful cynicism: a version that makes perfect narrative sense may have been made to.
First example: two Medusas, and the one that came later
The oldest surviving account makes Medusa one of three Gorgon sisters, born monstrous to primordial sea parents, and mentions almost in passing that Poseidon lay with her in a spring meadow. There is no transformation in it, because there is nothing to transform: she is what she was born as, and her mortality among immortal sisters is the odd detail the early poets seem interested in.
The version everybody knows — beautiful priestess, assaulted in a temple, punished by the offended goddess for the offence committed against her — is Roman, from Ovid, some seven centuries later, and it is told inside a frame that gives the reader every reason to find the goddess's justice grotesque.
The disagreement changes what the story is about. On the older account, Perseus kills a monster and the interest is in the mechanics of surviving a gaze. On Ovid's, the monster was manufactured by a god's cruelty, and every subsequent reader who has treated Medusa as a figure of violated innocence — feminist criticism, contemporary novels, a great deal of tattoo art — is reading a Latin poem, not a Greek myth. Neither version is fake. But claiming "the Greeks saw Medusa as a victim" is a straightforward error of century and language.
Second example: Baldr's death as tragedy or as love triangle
In the Icelandic prose account, the death of Baldr is a cosmic hinge: dreams of doom, an oath extracted from every substance except one overlooked plant, a blind brother's hand guided by Loki, and a failed universal weeping that leaves the loss permanent. It is where the Norse material turns irreversibly toward its ending, and it is enormously affecting.
Saxo Grammaticus, writing in Latin in Denmark at roughly the same period, tells something that overlaps only in names. His Balderus and Hotherus are rival warriors — a demigod and a mortal prince — competing over a woman, fighting several pitched battles, with a magical sword and prophetic woodland maidens involved and Hotherus as a sympathetic protagonist rather than a duped instrument.
You cannot merge these. What you can do is ask what each is for. The Icelandic version belongs to a systematised cosmology heading for a scheduled apocalypse. The Danish version belongs to a national history that needed heroic ancestors and had no use for gods who were not men. The existence of both is evidence that the story was flexible enough, in the twelfth and thirteenth centuries, to be pulled into either shape — which tells us something real about the state of the tradition that either version alone would conceal.
Third example: Egypt's rival creation accounts are not confusion
The Egyptian material is often described as bewildering because it contains several incompatible creations. It contains several because several cities each had one, and each was arguing.
At Heliopolis the model is generative and genealogical: a creator who arises by himself, brings a first couple into being out of his own substance, and stands at the head of a descending line of nine. At Memphis, the priesthood of Ptah made a case that the world was thought by the heart and spoken by the tongue, with the Heliopolitan gods reduced to the products of that utterance: a takeover bid conducted in theology. At Hermopolis, eight primordial beings representing the qualities of the pre-creation waters do the work instead.
Read as competing claims by institutions with budgets, these stop being contradictions and become the clearest evidence in the ancient world for how theological politics operated. The Memphite text does not deny the Heliopolitan account. It swallows it and puts something else on top — which is precisely what a rival capital would do, and precisely what a later empire would do again with Marduk in Babylon.
What to actually do when the sources fight
There is a discipline for this, and it is not difficult.
- Date every claim before weighing it. "Earlier" is not automatically "truer", but a difference of seven hundred years and two languages is the first thing to establish, and most popular retellings do not tell you.
- Ask what each source needed. A handbook wants completeness. A hymn wants the local version. A chronicle wants a legitimate ancestor. A poem wants the ending to land. Their distortions are predictable in direction.
- Never average. The worst treatment is a smooth composite assembled from four incompatible texts, because it produces a version that no one ever believed and that cannot be checked against anything.
- Prefer the inconvenient detail. Material that a source keeps despite its being awkward for that source's own argument is much likelier to be inherited than invented.
- Let the disagreement be the finding. Where the record genuinely does not resolve — and it often does not — saying so is more useful than a confident sentence with no evidence behind it.
That last point is the editorial line of this site. Pages here show competing versions beside each other, name the work each comes from, and mark which one a given claim depends on. The family trees draw a disputed parent as disputed rather than choosing. The comparison pages test equations instead of assuming them. It makes for slightly more demanding reading than a tidy summary, and it is the only version of this subject that can be checked.
Follow-up questions worth asking
Do modern scholars ever declare a winner between variants?
Sometimes, on evidence rather than taste. A version can be dated by loanwords or metre, tied to a datable political event, or shown to depend on a text we can date, and that can settle priority. What specialists generally will not do is declare one variant the authentic myth and the rest corruptions, because the assumption of a single lost original is the thing a century of comparative work has most thoroughly undermined.
Does a story appearing in more sources make it more reliable?
Not by itself, and treating frequency as evidence is a common trap. If nine of ten sources are copying the same handbook, that is one witness with nine echoes, not nine witnesses. Textual scholarship weights independent lines of transmission, not raw counts — and a single early inscription can outweigh a dozen late compilations that all descend from one another.
Are there myths where every source agrees?
Very few, and the agreement usually indicates a short transmission chain rather than a secure fact — a story recorded once, early, and copied from that record thereafter. Uniformity in this field is more often a sign that only one witness survives than that everyone independently confirmed the same thing.
Which tradition here has the worst-behaved evidence?
The Celtic, on any reasonable measure: the gap between the practising cultures and the surviving texts is the widest, the recorders were the most ideologically distant from the material, and the modern secondary literature is the most contaminated by nineteenth-century invention. The Celtic pages consequently spend more space than any others distinguishing what a medieval manuscript says from what a Victorian antiquarian wished it said.