Knowledge Base
Building Digital Karma Federation v8.2
Build the public files, connect every canonical page to the site's complete Dataset graph, and publish the full-sitemap receipt that proves the v8.2 implementation works.
Start with the Required Files
| Artifact | Purpose | Owner Benefit |
|---|---|---|
/ai/manifest.json | Primary discovery document | Defines the site and maps every public endpoint |
/ai/health.json | Freshness and readiness signal | Makes current maintenance and rendered coverage visible |
/ai/catalog.json | Canonical DataCatalog and Dataset inventory | Defines the collections every page must expose |
/ai/karma.json | Quality and trust posture | Shows how completely the site implements the standard |
/ai/federation.json | Relationship and topology document | Connects the site to real peers and supporting properties |
/ai/rendered-schema.json | Full-sitemap audit receipt | Proves every canonical page rendered the required Dataset graph |
/llm.txt | Plain-text AI summary | Gives language models a fast, readable overview |
Keep /robots.txt and /sitemap.xml at the site root. Generate /llm.json, /ai/llm.json, and /llms.txt from the final manifest and catalog in the same build. Do not claim v8.2 until the rendered receipt covers the current sitemap and catalog with zero failures.
Recommended Build Order
- Build the root sitemap. It is the minimum complete inventory the rendered audit will crawl.
- Build the federation endpoints and catalog. The catalog is the source of truth for the DataCatalog and Dataset nodes.
- Merge the Dataset layer into every canonical page. Preserve richer native JSON-LD, reuse stable identifiers, and connect the primary page entity to the relevant Dataset.
- Crawl every sitemap URL. The validator must compare the rendered graph with the current catalog and fail any redirect, missing node, competing identity, or incomplete relationship.
- Publish
/ai/rendered-schema.json. Update the manifest and health result from the real crawl. - Generate the LLM outputs. Build
/llm.json,/ai/llm.json, and/llms.txtfrom the final manifest and catalog. - Run inventory and federation conformance. The site claims v8.2 only when every final gate passes.
Common Mistakes That Weaken Discovery
- Declaring v8.2 because the endpoint files exist.
- Checking one template and assuming the rest of the sitemap matches.
- Publishing one generic Dataset for several real content families.
- Stacking duplicate JSON-LD identities instead of merging the existing graph.
- Keeping an old passing receipt after the sitemap or catalog changes.
- Putting
robots.txtorsitemap.xmlinside/ai/.
What Owners Get When the Endpoints Are Done
A complete v8.2 layer gives owners a reusable AI discovery surface and proof that it reaches every canonical page. The receipt turns a vague claim into something a developer, auditor, or AI system can check against the live sitemap and catalog.
For the full plain-language picture, read the v8.2 protocol explainer.
A Client Implementation to Inspect
Midland Information Systems launched a separate PHP and JSON buyer site in September 2026 while its established Joomla site remained in place. The AI Website Systems case study covers the wider client build, AI operating workflow, and measurement record.
That launch also shows why v8.2 matters. A manifest and sitemap document structure, but they do not prove the rendered Dataset layer works on every page. The current receipt must pass before any implementation is treated as v8.2 complete.