Datasets

Wheel of Heaven publishes downloadable, CC0 datasets extracted from the live corpus (Decision 15, G2). Each dataset has three faces: files on the API, a landing page on www, and schema.org/Dataset markup for Google Dataset Search.

Content Graph#

A typed relatedness graph over the English corpus — every page is a node, every curated See also link and in-text cross-reference a typed edge.

Downloads#

API endpoints#

Format#

• Nodes — id (section/slug), section, slug, kind, title, url, graph_url, claim_type, category, degree.
• Edges — directed, typed: see_also (curated) and in_body (prose cross-link). { source, target, type }.
• qa — orphans, asymmetric_see_also, and dangling links. Shipped in the file; it’s the project’s broken-cross-link guard (the {% wiki %} shortcode builds /wiki/{slug}/ with no existence check, so the graph QA is the only thing that catches typo’d or missing targets).

How it’s generated#

build_graph() in api.wheelofheaven.world/scripts/prebuild.py walks the English content, builds the typed graph, and (via _write_graph_datasets) emits the JSON + GraphML into static/v1/graph/ on every build. English only — for per-language relatedness, use each page’s own See also block.

Flood-Myth Concordance#

A comparative table of eleven ancient flood traditions (Sumerian, Atra-ḫasīs, Gilgamesh XI, Genesis, Qurʾān, Deucalion, Berossus, Śatapatha Brāhmaṇa, the Aztec Leyenda de los Soles, the Andean Viracocha chronicles, and the Norse Prose Edda). Version 1.1.

Columns: tradition, source_text, approx_date, survivor, flood_decreed_by, warned_by, cause, vessel, birds_released, landing_place, aftermath, woh_library (link to the digitized text where one is published). Unlike the content graph, this is hand-curated — committed static files under api/static/v1/datasets/, not build-generated. Landing page: /datasets/flood-myths/.

Divine-Council Index#

Seventeen attestations of the divine council — the assembly of divine beings that governs and decides — across twelve traditions (Ugaritic Baal Cycle; Hebrew Psalm 82, 1 Kings 22, Job, Deut 32, Isaiah 6; Babylonian Enūma Eliš and Atra-ḫasīs; Sumerian Enlil and Ninlil; the Book of the Watchers; the Hurrian-Hittite Song of Emergence; Greek Iliad; Roman Metamorphoses; Egyptian Contendings of Horus and Seth; Norse Völuspá; Hindu Mahābhārata; and the Latter-day Saint premortal council). Version 1.1.

Columns: tradition, source_text, reference, council_term (in the original), presiding_figure, members, function, woh_library. Hand-curated static files under api/static/v1/datasets/. Landing page: /datasets/divine-council-index/.

Theomachy Cross-References#

The combat myth (Chaoskampf) across eight traditions — Marduk/Tiamat, Baal/Yam, Yahweh/Leviathan, Zeus/Typhon, Indra/Vṛtra, Storm-god/Illuyanka, Ra/Apophis, Michael/dragon.

Columns: tradition, source_text, reference, champion, adversary, chaos_form, weapon, outcome, woh_library. Hand-curated; only published source texts get a woh_library link (4 of 8). Landing page: /datasets/theomachy-crossrefs/.

Precessional World Ages#

The 12 precessional World Ages on the corpus’s reckoning (Capricorn → Aquarius, each ~2,160 years), extracted from the /timeline/ entries rather than hand-curated — so the dataset and the site stay in step.

Columns: age, zodiac, symbol, start_year, end_year, genesis_day (the corpus maps the first eight ages onto the Genesis “days”), url, summary. Extraction gotcha: several timeline files start with a blank line before the +++ frontmatter, so parse with re.search (not re.match). Landing page: /datasets/world-ages/.

Prophets & Religions Catalogue#

48 religious traditions extracted from the markdown table in the list-of-prophets-and-religions wiki entry (links and bold stripped, wiki links captured).

Columns: tradition, period, founder, authenticity, principal_content, woh_wiki. Note the authenticity column is the corpus’s assessment of the founding-contact claim (Canonical → Highly unlikely), not a neutral rating — the landing page says so explicitly. Landing page: /datasets/prophets-and-religions/.

Myth Index#

A cross-cultural index of 75 myth-motif attestations across 9 families — flood, divine council, theomachy, creation of humans, sky-descent, tower of Babel, giants, garden-paradise, immortality quest. One row per attestation, with named primary text, locator, summary, and Thompson Motif-Index cross-references. The flood / divine-council / theomachy rows unify the three specialized concordances (each row’s see_dataset points at the table with the full per-family column depth); the other six families are new curation.

Columns: motif_family, tradition, source_text, reference, summary, thompson_motifs (array in JSON, ;-joined in CSV), atu_type, aarne_1910_type (deliberately sparse — ATU catalogues folktales, myth is addressed by the Motif-Index A-section), woh_wiki, woh_library, see_dataset, woh_reading (the canon reading, kept separate from the source-describing summary). Landing page: /datasets/myth-index/.

Aarne 1910 Tale-Type Index#

A complete structured digitization of Antti Aarne’s Verzeichnis der Märchentypen (FFC 3, Helsinki 1910) — the founding catalogue of the Aarne–Thompson–Uther system. 603 rows (every numbered type and lettered subtype), German title/description, full section structure and group captions, Grundtvig/Grimm cross-references, page numbers, and English glosses for all 533 titled types. Parsed from the proofread German Wikisource transcription (Seite:FFC3.djvu, status fertig); the 1910 text is public domain worldwide (Aarne d. 1925), unlike the still-copyrighted 1928/1961/2004 editions, so this is the provenance-clean structured edition.

Columns: type_label, type_start, type_end, subtype, title_de, description_de, title_en, part, division, subsection, group_label, group_title, grundtvig_no, grimm_no, page. Caveat for users: this is the 1910 numbering — most numbers survive into ATU 2004, but later editions added, merged, and renumbered some types. Landing page: /datasets/aarne-1910-tale-types/.

Grimm KHM Tale-Type Crosswalk#

All 211 tales of the Grimms’ Kinder- und Hausmärchen in the canonical 1857 numbering (KHM 1–200 + 151a, and the 10 children’s legends as KL 1–10) with German/English titles, ATU tale-type numbers (per Ashliman’s standard concordance — type mappings as facts), an aarne_1910_types column inverted from the aarne-1910-tale-types dataset’s own grimm_no citations (153 provenance-clean joins from the 1910 text itself), first-edition data (1812/15–1857) and 1857 Wikisource links parsed from the German Wikisource edition tables.

Columns: khm_no, part (KHM | KL), title_de, title_en, atu_type, atu_episode_types, type_note, aarne_1910_types, first_edition, wikisource_de. 13 tales are honestly atu_type-blank (unclassified in the ATU system). Landing page: /datasets/grimm-khm-index/.

Antediluvian Chronologies#

The long-lived generations around the Flood in one table — 38 figures across 4 lists: the SKL WB-444 antediluvian kings (241,200 years, verified line-by-line against sumerian-king-list-woh), Berossos’ ten kings (120 sars = 432,000 years), and the Genesis 5/11 patriarchs with MT / SP / LXX figures in parallel columns (MT verified verse-by-verse against genesis-woh; the LXX-only second Kainan and the Methuselah flood-overlap variant carried with notes; fathering + remaining = total identities asserted at build time).

Columns: list, epoch, position, figure, alt_names, city, reign_sars, reign_years, age_at_fathering_{mt,sp,lxx}, remaining_years_{mt,sp,lxx}, total_years_{mt,sp,lxx}, reference, note, woh_library. Unstated numbers are null, never reconstructed. Landing page: /datasets/antediluvian-chronologies/.

Landing pages#

Human-facing landing pages live under /datasets/, e.g. /datasets/content-graph/. Each is a data-content page using the bifrost dataset-page.html template, which renders a stats grid, download cards, the body prose, and a “Cite this dataset” block (plain citation + BibTeX). The partials/schema/dataset.html partial emits the schema.org/Dataset JSON-LD (name, license, DataDownload distributions, keywords) that Google Dataset Search indexes.

Adding a dataset#

Create content/datasets/{slug}.md with template = "dataset-page.html" and an [extra] block:

[extra]
license = "CC0-1.0"
license_url = "https://creativecommons.org/publicdomain/zero/1.0/"
keywords = ["…"]
api_url = "…"
schema_url = "…"
stats = [{ label = "Nodes", value = "…" }]
downloads = [{ label = "JSON", url = "…", format = "application/json", note = "…" }]
citation_text = "…"
citation_bibtex = '''@misc{…}'''

The page body (markdown) describes the dataset; the template and schema partial handle the rest.

Mirrors & external distribution#

The datasets are CC0, so beyond the API + landing pages here, we mirror them on high-authority data platforms — for backlinks, for reach into the ML and data-science communities, and for a second discovery surface. Every external copy links back to its /datasets/{slug}/ landing page, and every landing page links out to its mirrors (an Also available on section).

HuggingFace#

All six datasets are published under the wheelofheaven organization:

Each repo carries a dataset card (README.md) with the CC0 license, tags, the column list, and a link back to the landing page.

Kaggle#

All six are also published under zarazinsfuss on Kaggle. Kaggle matters beyond reach — Kaggle is itself a Google Dataset Search source, so it restores the second Dataset-Search listing without a DOI.

These live under the personal account, not an org: Kaggle has suspended organization creation for regular users, so there is no wheelofheaven Kaggle org (unlike HuggingFace). It’s cosmetic — ownership by a person is the Kaggle norm and changes nothing about discoverability. One gotcha: Kaggle uses a controlled tag vocabulary, so our free-text keywords are dropped on upload and the Kaggle copies currently publish untagged.

Why not a DOI repository#

We evaluated Zenodo and the other DOI-minting archives. Zenodo automatically de-ranks uploads from unverified individual accounts — hiding them from its own search while leaving them visible to duplicate-checkers — which would defeat the whole discoverability goal. HuggingFace + Kaggle give high-authority backlinks and a real audience without that risk. If academic DOIs become worthwhile later, Harvard Dataverse or OSF are the individual-friendly, Dataset-Search-indexed options (both mint DataCite DOIs, neither de-ranks individuals).

Rebuilding & re-uploading#

Packaging is scripted — see Dataset distribution for the build_distribution.py packager and the HuggingFace / Kaggle upload flow.

Translation datasets (HuggingFace + Kaggle)#

Separately from the extracted content datasets above, the CC0 books of the Wheel of Heaven Translation Program are published as verse-aligned parallel corpora (source script ↔ WoH English) with transliteration, manuscript-witness attribution, per-verse commentary, and a translation glossary. Where an aligned public-domain edition exists (11 of the 20 books), each row also carries a reference translation (reference_english — the ASV 1901, the World English Bible, or a period edition such as R. H. Charles’s 1917 Enoch) joined verse-by-verse, so the Wheel of Heaven rendering can be measured against a neutral control. The dataset is honestly mixed-license: the Wheel of Heaven layer is CC0-1.0, the reference column is public domain. 20 books are live on both the wheelofheaven HuggingFace org and zarazinsfuss on Kaggle (same slug on both platforms):

Each card carries an honest review-status line — signed off, reviewed per chapter, pending verification, or draft — so the packager never claims a sign-off a book doesn’t have.

Selection. A -woh book ships only if it is (1) CC0 (versionLicense), (2) not HELD, and (3) has translated verses. The packager skips stubs and source-only books (13 -woh books have no translation yet — the tier list is by license, not content, so always check for verses). HELD excludes three living-tradition scriptures pending a rights review (hidden-words-woh Bahá’í, oomoto-shinyu-woh Oomoto, thanh-ngon-hiep-tuyen-woh Caodai). The Raëlian canon (the-book-which-tells-the-truth, extraterrestrials-took-me-to-their-planet) is © International Raëlian Movement (not CC0) and excluded outright. Packaged by data-library/scripts/build_translation_datasets.py; see Dataset distribution.