# AI / LLM Files — Purpose & Maintenance

This directory contains machine-readable and plain-text artifacts intended to provide authoritative, canonical information about RHEDYN for AI crawlers, retrieval-augmented systems, and other automated indexing agents. The files are short, factual, and formatted to be easy for LLMs and search crawlers to consume.

## Intended files
- `rhedyn-llm.jsonld` — JSON-LD (schema.org) LocalBusiness and Service structured data. Use for machine-readable discovery and rich results.
- `rhedyn-knowledge.md` — Concise canonical knowledge base suitable for retrieval and embedding (RAG) systems. Designed for direct ingestion by LLMs and AI indexing crawlers.
- `README.md` — (this file) explains purpose and maintenance practices.

## Goals
- Provide a small, authoritative set of facts about RHEDYN that AI systems can rely on.
- Keep content factual, non-promotional, and permission-safe (no private data).
- Encourage consistency between JSON-LD, site metadata, and public profiles so crawlers can verify signals.

## Publishing guidelines
1. Populate authoritative contact details only when they are public and verified:
   - `telephone`
   - `address` (structured PostalAddress)
   - `sameAs` (social profiles)
2. Ensure the `url` and canonical `@id` in JSON-LD use the canonical HTTPS domain (e.g., `https://rhedyn.cymru`).
3. Include the JSON-LD file in the site (or make it available at a stable path), and reference it from the sitemap if possible so crawlers discover it.
4. Keep `rhedyn-knowledge.md` concise (short sections and bullet lists). Large essays or marketing copy reduce usefulness for embedding.
5. Do not include personal, sensitive, or contractual information (private emails, client secrets, or NDA content).

## How to improve AI discoverability (recommended)
- Add `sameAs` links to verified social profiles (LinkedIn, Twitter/X, GitHub, Facebook).
- Publish short, factual team bios (name + role + public link) without private contact details.
- Publish curated case studies with explicit client permission — include outcomes, dates, and non-sensitive metrics.
- Maintain a clean sitemap and make the `/ai/` folder reachable from the site or robots.
- Use stable canonical URLs and avoid redirect chains for pages referenced in JSON-LD.

## Maintenance checklist before publishing
- [ ] Confirm domain is canonical and uses HTTPS.
- [ ] Add `sameAs` array with verified social profile URLs.
- [ ] Add `telephone` and `address` if publicly available.
- [ ] Verify logos and images referenced use absolute URLs (prefer HTTPS).
- [ ] Ensure `rhedyn-knowledge.md` reflects any recent business changes (services, pricing, area served).
- [ ] Link the `/ai/` resources from the main site or include them in the sitemap.xml.

## Privacy & permission notes
- Obtain explicit written permission before publishing client names, direct quotes, or detailed case study metrics.
- Avoid storing or publishing personal data about individual employees beyond what is already public and non-sensitive.
- If a piece of content could identify a private individual or expose internal strategy, do not include it.

## Format and style tips
- JSON-LD: Keep the top-level type as `LocalBusiness` and list offerings as `Service` entries. Keep descriptions short (1–2 sentences).
- Knowledge markdown: Use short headings, bullet lists, and clearly labelled factual sections (Overview, Services, Pricing, How we work).
- Keep each file under a few kilobytes where possible — small, canonical files are easier for indexing pipelines to ingest.

## Where to link from the site
- Add a link in the footer or an `/about`/`legal` page to the `/ai/` folder (or the specific files).
- Ensure the files are included in `sitemap.xml` or otherwise discoverable by crawlers.

## Example publishing workflow
1. Update `rhedyn-llm.jsonld` with verified `telephone`, `address`, and `sameAs`.
2. Validate JSON-LD with an online structured data tester.
3. Update `rhedyn-knowledge.md` to reflect any business changes.
4. Add links to these files from a discoverable page and update the sitemap.
5. Re-run validation and monitor indexing over the next few weeks.

## Contact & ownership
Maintain responsibility: the site owner or an assigned maintainer should update and sign off on changes to the AI files. Keep an internal changelog or note the `Last reviewed` date in the markdown file.

---
Keep these files factual, concise, and up-to-date. They are intended to be a trustworthy source of truth for machines — not marketing copy.