The Edgar Cayce Readings Wiki

A searchable, linked edition of the complete Edgar Cayce readings corpus. The project turns 14,306 readings from the A.R.E. Complete Readings CD-ROM into an Obsidian-compatible Markdown vault and publishes it as a Quartz v5 knowledge site.

Live site: https://cayce-wiki.vercel.app

What is included

The repository contains 24,301 Markdown notes:

Collection	Notes	Purpose
content/readings/	14,306	One note per Cayce reading, including transcript text, metadata, reports, background, and linked mentions
content/entities/	9,975	People, places, concepts, remedies, organizations, and other linked entities
content/series/	19	Reading-series indexes organized by subject and reading-number range
content/index.md	1	Site home page and high-level corpus index

The published site provides:

• full-text search across reading titles, tags, and compact excerpts;
• Obsidian-style internal links and backlinks;
• local graphs for individual notes;
• a curated global graph of prominent entities and reading series;
• reading metadata such as reading number, series, year, sex, entities, and tags;
• RSS and sitemap output;
• responsive light, dark, and reader modes.

Corpus and editorial notes

The source corpus covers readings dated 1901–1944 and was extracted from the A.R.E. Complete Readings CD-ROM.

Entity notes and links are partly machine-generated. Surface-form linking and LLM-assisted extraction are useful navigation aids, but they are not scholarly authority. Verify quotations, entity identifications, dates, and interpretive claims against the reading text before citing them.

The repository currently reports some slug collisions among similarly named entity notes. Quartz uses the last processed note for a collided URL and prints the complete collision list during builds. These warnings do not stop deployment, but contributors should avoid introducing new collisions and should consolidate existing duplicates carefully.

Repository layout

.
├── content/                         # Published Markdown vault
│   ├── readings/                    # 14,306 reading notes
│   ├── entities/                    # Linked entity notes
│   ├── series/                      # 19 series indexes
│   └── index.md                     # Site home page
├── quartz/                          # Quartz framework source
├── quartz.config.yaml               # Site, theme, plugin, and layout configuration
├── scripts/
│   ├── patch-quartz-performance.mjs # Exact, guarded patches for large-vault clients
│   ├── strip-content-index.mjs      # Splits and compacts search/graph data
│   ├── vercel-build.mjs             # Canonical optimized production build
│   └── prepare-vercel-output.mjs    # Creates a Vercel Build Output API artifact
├── .github/workflows/deploy.yml     # GitHub Actions build and prebuilt Vercel deploy
├── vercel.json                      # Clean-URL and Vercel build configuration
└── public/                          # Generated site output; ignored by Git

Requirements

• Node.js 22 or newer
• npm 10.9.2 or newer
• Git

A full build processes more than 24,000 Markdown files. Expect high CPU and memory use and a build time of roughly 35–50 minutes on a two-core CI runner.

Local development

Install the pinned dependencies:

npm ci
npm run install-plugins

Run the complete test suite:

npm test

Run TypeScript and formatting checks:

npm run check

Start Quartz's local development server:

npx quartz build --serve

The development server is useful for content and layout work, but it does not apply the production-only large-vault patches described below.

Production build

Use the canonical production command:

VERCEL_PROJECT_PRODUCTION_URL=cayce-wiki.vercel.app npm run vercel-build

This command performs the following steps in order:

1. rewrites Quartz's build-time baseUrl to the deployment host;
2. applies guarded performance patches to the installed Quartz search and graph clients;
3. builds all Markdown content into public/;
4. compacts public/static/contentIndex.json for lazy search;
5. creates public/static/graphIndex.json for lazy local and global graphs.

The build-time baseUrl rewrite updates the working copy of quartz.config.yaml. In CI this happens in an ephemeral checkout. After a local production build, restore the file before committing:

git restore quartz.config.yaml

Why the custom build exists

Quartz's standard content index is too expensive to parse and traverse on every page when the vault contains more than 24,000 notes and hundreds of thousands of graph links. This project therefore:

• initializes search only after the first query;
• coalesces concurrent search initialization and discards stale results;
• keeps excerpts short in the search index;
• downloads graph data only when a graph is requested;
• precomputes bidirectional local adjacency at build time;
• limits the global graph to prominent entities and series;
• requires an explicit click before rendering a local graph.

scripts/patch-quartz-performance.mjs uses exact, count-checked replacements. If an upstream Quartz package changes, the build fails instead of silently publishing an unoptimized site.

Deployment architecture

Production deployment has one authoritative path:

push to main
    ↓
GitHub Actions
    ↓
clean install → TypeScript check → full test suite
    ↓
npm ci → optimized Quartz build → compact indexes
    ↓
Vercel Build Output API artifact
    ↓
prebuilt, compressed deployment to Vercel
    ↓
https://cayce-wiki.vercel.app

The Vercel project's direct Git integration is intentionally disconnected. A direct Vercel Git build repeats the full Quartz build inside Vercel and can exceed Vercel's 45-minute build limit. GitHub Actions builds the site first and uploads the completed static artifact with vercel deploy --prebuilt --archive=tgz.

Required verification gates

.github/workflows/ci.yml runs on pull requests targeting main and is reused by the production workflow for the same commit. It performs a clean install, generates the ignored .quartz/plugins index with npm run install-plugins, then runs npx tsc --noEmit and npm test without production secrets. The deployment job requires this verification job to succeed before it starts the expensive build, prepares the artifact, or accesses the Vercel token. Manual production dispatches are also gated and publish only from main.

These gates deliberately do not run whole-corpus formatting: npm run check includes a separate Prettier audit with existing formatting debt. The CI-equivalent local checks are:

npm ci
npm run install-plugins
npx tsc --noEmit && npm test

The workflow contract is covered by scripts/vercel-build.test.mjs. These checks do not replace production HTTP/browser verification, and they do not configure GitHub branch protection. The optimized build and Vercel prebuilt upload remain unchanged; the disconnected direct Git integration is not re-enabled.

Required repository secret

The GitHub repository must define:

Secret	Purpose
VERCEL_TOKEN	Authorizes the prebuilt production deployment

The Vercel organization and project IDs are non-secret identifiers defined in .github/workflows/deploy.yml.

Manual deployment

To redeploy the current main revision without another commit:

gh workflow run deploy.yml --repo Nostem/cayce-wiki --ref main

Then watch the run:

gh run watch --repo Nostem/cayce-wiki

Tests and verification

Run all tests before pushing:

npm test
npm run check
git diff --check

Focused deployment and index tests can be run directly:

node --test scripts/vercel-build.test.mjs
node --test scripts/strip-content-index.test.mjs
node --test scripts/patch-quartz-performance.test.mjs

After deployment, verify at minimum:

• / loads successfully;
• a representative reading such as /readings/3744-1 loads;
• an entity page such as /entities/atlantis loads;
• search returns results after normal multi-keystroke input;
• result navigation remains client-side and responsive;
• /static/contentIndex.json and /static/graphIndex.json return 200;
• generated HTML contains no retired /cayce-wiki/ deployment prefix.

Contributing

1. Create a focused branch.
2. Make content or code changes without modifying generated public/ output.
3. Preserve frontmatter, reading identifiers, and existing wikilink conventions.
4. Run the relevant focused tests and the complete suite.
5. Check the build output for new slug collisions.
6. Open a pull request describing the source and verification for content changes.

For corpus corrections, include the reading number and quote enough surrounding text to make the correction auditable. Avoid turning generated entity summaries into uncited factual claims.

Licensing and rights

The Quartz software in this repository is distributed under the MIT License.

The Edgar Cayce readings, A.R.E. source media, transcriptions, names, and related archival material may have separate copyright, trademark, or usage restrictions. The MIT license for the software does not grant rights to third-party corpus material. Users are responsible for determining whether their use of the content is permitted.

Acknowledgments

• Quartz by Jacky Zhao and its contributors
• The Association for Research and Enlightenment (A.R.E.) as the source publisher of the Complete Readings collection
• The archivists, transcribers, indexers, and researchers who preserved and organized the Cayce material

---

Created by Nostem.