Knowledge graph lab¶
Practice question. Which link, reference or reading path should the CDI-BoK maintainer fix first? Prioritize changes that help readers reconstruct evidence or reach an application from its guidance.
Open the CDI-BoK in the explorer
Load the book and follow a relationship¶
- The link loads the CDI-BoK directly. To load it manually or pin a different version, choose Load another vault → GitHub. Enter
jaforero/conversational-decision-intelligence-bokand set the subfolder todocsunder Options. Usemainfor the current corpus or a tag to reproduce a historical version. The repository is public and loads without a token, subject to GitHub API limits. - Open
learn/02-evidence-context.mdand follow the link to the experience design module. At the destination, inspect the backlink and its context. Repeat withresearch/research-agenda.mdand find the candidate GraphRAG protocol. - Classify one finding. A broken link points to a target that fails to resolve. A page without inbound content links can still be reachable through navigation. An implicit relationship requires checking both documents before adding an editorial link.
- Verify in the portal and on GitHub. MkDocs conventions and
.en.mdtranslations may behave differently in a generic Markdown explorer. Record the source page, expected target, edition, evidence, owner and proposed correction. Verify both editions after the change.
The explorer's Health panel counts Markdown files in both languages and flags notes with neither recognized incoming nor outgoing links. The editorial inventory below measures a different property: rendered canonical pages with no inbound content links. Compare each count using the same corpus and criterion. English Markdown references point to the matching .en.md file so the links work in the portal and in the explorer.
Deliverable. Prepare one reproducible finding and a small correction. If a relationship already exists, use it as a control and record what you checked.
Inspect the CDI-BoK editorial graph¶
On each build, the site generates a JSON inventory from registered canonical pages, their rendered links, and the source and claim registries. It distinguishes four relationships: page links to page, page lists source, page lists claim, and claim lists source. Each label describes what the repository records.
Download the editorial graph as JSON
The pages_without_inbound_content_links field lists review candidates, excluding the home page. Navigation links are excluded and each page is counted once, in the canonical Spanish edition. A declared reference helps trace provenance. Its relevance and strength require reviewing the evidence.
Acceptance criterion. Trace a path from a page to a source or claim, check it against the original file, and document what editorial action would change. The GraphRAG protocol describes the additional test for relationship-based retrieval.