This practical builds the part of Lesson 5.6’s system you can run today: an entity registry, JSON-LD drafted by Claude from that registry for a few pages, and the checks that make it trustworthy. It deliberately stops short of templates and deployment.
What this lesson covers
- The entity registry and its rules
- Claude-drafted JSON-LD with evidence
- Graph integrity in code
- Schema and Google validation
- Shipping without creating a second graph
- Re-checking at thirty and ninety days
The registry is a short file with one entry per real-world thing — the organisation, its people, its products, the website — each with an identifier, one type, its canonical name and only the sameAs links you would defend. The lesson explains the rules learned from building our own: an identifier lives on the domain that owns the entity, one registry rather than copies, one type per node, and relationships expressed by reference. Claude then drafts JSON-LD for three pages, referencing registry entities by identifier rather than re-declaring them, and lists every claim it made with the sentence on the page that supports it.
Code checks what no public validator does: that every reference resolves, that no node has two types, that no entity is re-declared, that no new sameAs link appears and that every supporting sentence is really on the page — Google’s rule that structured data must be a true representation of the page content, turned into a test. The Schema Markup Validator and the Rich Results Test then cover vocabulary and Google’s requirements. Before anything goes live, the lesson has you count the JSON-LD blocks already on the page, because a second emitter is the most common way a fix becomes a new contradiction.
Every script runs as written, with comments, and the lesson explains how to read its two check columns: one for graph integrity, one for claims whose supporting sentence could not be found on the page. Fixing the registry or the prompt rather than the output is the habit it builds, so the next run is right as well.
This practical follows Lesson 5.6 and completes Module 5 of The AI-powered Entity Audit. It suits learners with an Anthropic API key who want structured data that agrees with their content; implementation across whole sites is the subject of the Structured Data Automation course.