llms.txt and llms-full.txt

The Wheel of Heaven project publishes both flavours of the llms.txt convention: a short manifest and a long full-corpus reference. They sit at stable URLs and are designed to be ingested by AI tools directly, with no transformation.

The two files#

Both files are served by www.wheelofheaven.world and are authoritative for the English corpus. The API site also serves its own llms.txt — a separate manifest for the API surface itself, not for the reading content. See api.wheelofheaven.world/llms.txt.

llms.txt — the navigational manifest#

The short manifest follows the llms.txt spec: a brief project description, a > blockquote summarising what the project is, then sections of links with one-line annotations.

It contains:

• A one-paragraph statement of the project’s working hypothesis.
• Three paragraphs of context (what the project is and is not).
• A “Content sections” list — the seven top-level sections of the reading site, each with a one-line role description and live URL.
• A “Twelve precessional ages” list — each age with its dates and a one-line summary.
• A “Key concepts” list — Elohim, precession, Great Year, World Age, Raëlism, The Book Which Tells the Truth, Genesis, Hamlet’s Mill, Jean Sendy.
• A “How to route queries” block — for which kind of question, hit which section.
• A “Languages” block — per-language entry points.
• An “API access” block — discovery URLs for the JSON twin.
• A “Citing this source” block — CC0 + suggested citation format.
• Project information and optional sections.

The file is handwritten and editorially curated, not auto-generated. It is the project’s most carefully maintained AI-facing surface. When the project’s framing shifts (new editorial pass, new section, new key concept), this file is updated as part of the same pass.

Direct use#

For a quick chat assistant:

You are answering questions about the Wheel of Heaven project. The
manifest below is the project's own short description of itself.

<paste of https://www.wheelofheaven.world/llms.txt>

For an agent’s tool description (e.g. a tool called query_wheel_of_heaven):

This tool exposes the Wheel of Heaven project. The project's manifest:

<paste of https://www.wheelofheaven.world/llms.txt>

Use this tool when the user asks about ancient cosmology, the Elohim
hypothesis, the precessional ages, Raëlism, or related comparative
mythology. The tool returns either a URL to follow up on or a short
prose summary.

llms-full.txt — the full-corpus reference#

The full file follows the same llms.txt convention but is intended as a single ingestion point for the full project, not just a manifest of links. It contains:

• The full short manifest content (so you can paste either file, not both).
• The Working Hypothesis stated in five interconnected components.
• The Project’s Epistemic Stance — claim labels, methodological commitments, self-positioning.
• The Precessional Framework — the astronomical phenomenon and the corpus’s working dates.
• The Twelve Ages — each age summarised at ~2–3 substantive paragraphs covering its content, sources, and place in the larger arc.
• The Project’s Positioning relative to adjacent traditions (Raëlism, ancient astronaut theory, biblical scholarship, comparative mythology, contemporary science).
• Methodological Notes on source families and authority tiers.
• Key Concepts Glossary in prose form.

The file is approximately 22,000 tokens. It fits comfortably in any modern model’s context window (GPT-4o, Claude Sonnet/Opus, Gemini, Llama 70B+, Mistral Large all handle this trivially). For a 100K+ token model, this is the simplest “give the model everything” pattern available.

Direct use#

You answer questions about the Wheel of Heaven project. Below is the
project's own full-corpus reference document. Treat it as ground truth.

When you answer a substantive claim, surface the corpus's epistemic
labelling: "direct" (a source says this and the mainstream agrees),
"framework" (the project's canon states this, mainstream scholarship
does not endorse it), "inferred" (scholarship concludes this), or
"speculative" (the project proposes this).

When the document doesn't cover a specific detail, say so and point to
the relevant section URL.

<paste of https://www.wheelofheaven.world/llms-full.txt>

How to keep your ingest fresh#

Both files are stable URLs. The project does not version them — when the corpus is updated, the files are updated in place. If you cache them downstream:

• The reading site’s edge cache returns these with the default Cloudflare TTL (~1 hour).
• The site rebuilds on every push to main, and Cloudflare purges the zone on deploy.
• For most ingestion patterns, refetching weekly is sufficient. For applications where accuracy is critical, refetch daily.

If you need to detect changes without refetching, the API publishes a sitemap with <lastmod> per endpoint at api.wheelofheaven.world/sitemap.xml. The llms.txt and llms-full.txt files themselves don’t carry change metadata; the API sitemap is the closest signal.

Multilingual#

/llms.txt and /llms-full.txt are English only. The corpus is published in nine languages, but the manifests are not per-language — the English manifests are authoritative.

The reasoning: an LLM consuming these files is overwhelmingly likely to be using English as its internal working language, and the project’s editorial discipline (claim types, terminology, source program) is most precisely expressed in English. If you need per-language access, hit the JSON API under /v1/{lang}/... — see Multilingual.

How the files are generated#

llms.txt is handwritten in www.wheelofheaven.world/static/llms.txt. It is treated as editorial content — owned by the editorial pass, not the build pipeline. When the project’s framing shifts or a section is added/retired, the file is updated by hand in the same PR.

llms-full.txt is similarly handwritten in www.wheelofheaven.world/static/llms-full.txt. It is more time-consuming to keep in sync with the live corpus — each of the twelve age summaries is its own paragraph of editorial work — but the manual maintenance is what lets the file stay sharp, opinionated, and stable in style. The trade-off is intentional.

Both files are served from /static/ by Zola, which means they appear at the root of www.wheelofheaven.world.

What llms.txt deliberately does not contain#

• Per-page bodies. The manifest is navigational. For full bodies, fetch the JSON API.
• Live numbers. The manifest does not list “we have 111 wiki entries” because the number changes; it points at the listing endpoint instead.
• The project’s full bibliography. That lives at /v1/sources/. The manifest only mentions key reference works.
• Schema or enum values. Those are machine surfaces — the manifest is for a human-language-trained model. Schemas live at /v1/schema/ and enums at /v1/enums/.