Quickstart

The fastest path to an LLM that answers accurately about the Wheel of Heaven project, in order of effort.

Path A — single paste, small context#

For a chat assistant, a quick prototype, or a model with a tight context window (under 16K tokens).

Paste the contents of https://www.wheelofheaven.world/llms.txt into the system prompt. It is ~7K characters, ~1,800 tokens. It contains:

• The working hypothesis in one paragraph.
• A list of what the project is and is not.
• The seven content sections with one-line descriptions and live URLs.
• The twelve precessional ages with their dates and one-line summaries.
• Key concepts with links to wiki entries.
• Routing rules (“for definitional questions, use the Wiki”).
• Language entry points.
• API discovery URLs.

Prepend a short instruction:

You are answering questions about the Wheel of Heaven project using
the manifest below as ground truth. When the manifest does not cover
a question, say so and recommend the relevant URL. Never invent dates,
quotes, or page contents. Cite URLs verbatim when you reference them.

---

<paste llms.txt here>

This gets you grounded answers about the project’s structure, chronology, and key terms. It will not let the model answer detail-level questions about specific wiki entries or library texts — for that, go to path B or C.

Path B — full-corpus paste, large context#

For a model with a 100K+ token context window where you want answers without retrieval.

Paste the contents of https://www.wheelofheaven.world/llms-full.txt. It is ~85K characters, ~22K tokens. It contains everything llms.txt contains plus:

• The hypothesis stated in full (five interconnected components).
• The project’s epistemic stance and editorial method.
• The precessional framework explained.
• Each of the twelve ages summarised at ~2–3 paragraphs.
• The project’s positioning relative to adjacent traditions.

This is the corpus’s most ingestion-ready single artefact. It is designed as a one-file context block.

You are answering questions about the Wheel of Heaven project using
the full corpus reference below as ground truth. Where you cite a
specific claim, label it "direct" / "framework" / "inferred" /
"speculative" per the corpus's own epistemic discipline. Never invent
quotes.

---

<paste llms-full.txt here>

Path C — programmatic, retrieval-backed#

For an agent, RAG pipeline, or anything that needs to scale beyond a single paste.

Hit the curated context endpoints once at boot:

curl https://api.wheelofheaven.world/v1/context/
curl https://api.wheelofheaven.world/v1/context/hypothesis/
curl https://api.wheelofheaven.world/v1/context/terminology/
curl https://api.wheelofheaven.world/v1/context/timeline/
curl https://api.wheelofheaven.world/v1/context/sources/
curl https://api.wheelofheaven.world/v1/context/method/

Concatenate the data.body fields of those six responses into your system prompt. Then, for per-question retrieval, hit /v1/wiki/{slug}/, /v1/timeline/{slug}/, /v1/articles/{slug}/, or /v1/library/books/{slug}/chapters/{n} as needed.

Walk-through with Python:

import httpx

BASE = "https://api.wheelofheaven.world"

def curated_context() -> str:
    """Concatenate the six curated context documents into a block
    suitable for a system prompt."""
    keys = ["hypothesis", "terminology", "timeline", "sources", "method"]
    blocks = []
    for k in keys:
        r = httpx.get(f"{BASE}/v1/context/{k}/").json()
        blocks.append(f"## {r['data']['title']}\n\n{r['data']['body']}")
    return "\n\n---\n\n".join(blocks)


def wiki_entry(slug: str) -> dict:
    """Fetch a single wiki entry."""
    return httpx.get(f"{BASE}/v1/wiki/{slug}/").json()["data"]


def chapter(book_slug: str, n: int) -> dict:
    """Fetch a single library chapter."""
    return httpx.get(
        f"{BASE}/v1/library/books/{book_slug}/chapters/{n}"
    ).json()["data"]


system_prompt = f"""
You answer questions about the Wheel of Heaven project. Use the context
below as ground truth. Cite URLs verbatim. Surface the corpus's claim
type (direct / framework / inferred / speculative) when answering a
substantive claim.

{curated_context()}
""".strip()

This pattern keeps the system prompt small (~5K tokens) and pulls per-query detail on demand. Pair with retrieval — see Embeddings and RAG — for any non-trivial production deployment.

Choosing between A, B, and C#

If you don’t know which to pick: start with A, upgrade to B when the model starts inventing details, upgrade to C when B doesn’t fit or when you need per-query freshness.

What to do next#

• Read System-prompt patterns for the wording that has worked well in practice.
• Read Curated context endpoints for what each /v1/context/* document contains and when to include each one.
• Read Attribution before shipping a product that surfaces the corpus.