# Civilisers, version 2 — source comparison

Status: exploratory source audit. Begun 2026-09-05 after inspecting the original lineup
results and locating several relevant accounts. This is **not** a preregistration of an
unseen result. The source selection is purposive and the researcher knows the identities.

## Question

Which proposed correspondences between Viracocha, Quetzalcoatl, Bochica and Oannes are
actually stated in identifiable accounts? Do the same kinds of detail also occur in other
teaching or founding figures?

This first release is an evidence audit, not a global rarity test and not a test capable of
establishing an Ice Age civilisation. Its useful result is a checkable account of what the
comparisons do and do not share.

## Units and scope

The unit of evidence is a located passage within a named source, not an AI-written life of
a deity. Keep several passages from one source in one source family. Compare early
accessible accounts with a clearly labelled later comparative retelling where available.
"Earliest inspected" never means "earliest extant" without an exhaustive source search.

Start with the four named figures. Add teaching or founding figures as role comparators,
including local analogues; do not describe all comparators as unrelated negative controls.
Record source coverage for each. This small, uneven, deliberately selected set cannot
estimate a worldwide base rate.

## Claim definitions

- **Sea arrival:** the figure enters human society from the sea. A journey from the east,
  emergence from a lake, an unspecified appearance and departure by sea do not qualify.
- **Sea departure:** the figure leaves human society over or into the sea. Keep a recurring
  nightly return separate from a final departure.
- **Specific instruction:** the account explicitly attributes a named skill or institution
  to the figure. Record the skill; general abundance or wise rule is not teaching a craft.
- **Catastrophe sequence:** record who causes or survives the event and whether instruction
  occurs before or after it. A flood elsewhere in the same work is not automatically the
  civiliser's arrival episode.
- **Return promise:** the figure explicitly promises to return. A narrator's prediction,
  later expectation or a return that already occurs is a different claim.
- **Distinctive combination:** retain companions, objects, prohibitions and event order.
  Do not pool attributes across source versions and call the resulting bundle one account.

An observation is `present`, `contradicted`, `not_mentioned` or `uncertain`, with an
explanation and passage references. Negative statements apply only to the inspected
passages. Missing evidence cannot establish that a tradition lacked a detail.

## Workflow

1. Locate and read passages in accessible editions; use page/chapter/line locators that a
   reader can follow without this site's software. Visually check relevant PDF pages.
2. Save brief source excerpts and explicitly labelled paraphrases. Record translation
   status, source mediation and unresolved questions. No invented quotations or dates.
3. Compare the same narrowly defined claims across the named figures and role comparators.
   Expose source coverage and mismatches alongside matches.
4. Publish an evidence ledger, a source-linked comparison and a small set of findings.
   No p-value or ranking will be computed from this deliberately selected pilot.
5. A later rarity test would need a defined sampling frame, comparable source effort,
   independent source families, a fixed claim scheme, and additional accounts held aside
   from discovery. That test is not complete merely because this audit is published.

## Limits

The researcher and AI are unblinded. Online editions are uneven; translation and colonial
mediation matter. Sources with different purposes need not mention the same things.
Recording after contact shows an opportunity for influence, not proof of borrowing.
Resemblance can reflect transmission, shared inheritance, convergent storytelling or later
editing. This audit does not select one explanation in advance.

## Version history

The original experiment remains at `docs/heroes.html`, with unchanged raw and derived
results in `data/s4/`. Version 2 does not replace, combine with, or count the original model
judgments as additional historical witnesses.