Knowledge Base graph
Open Knowledge Base > Peta Pengetahuan. Topics use cached AI classification of published entry content. New or changed entries appear as awaiting categorization, rather than falling back to FAQ/agenda/announcement types. Published/manual articles and PDF sections appear as sources; saved retrieval questions appear as test nodes. Choose a topic, search its items, select a node or use the adjacent accessible list to inspect connections, open a source or test a question.
Coverage is calculated from current source eligibility and saved retrieval tests, without AI calls. Verified means every displayed saved check in the content topic passed. Partial means published evidence exists without complete passing test coverage. Gap means a saved question lacks its expected retrievable evidence. Sources excluded by approval, expiry, visibility or authority rules appear as needing review. Replacement/conflict/coexistence relationships come from source governance.
Limits: up to 1,000 source nodes, 1,000 source relationships and the latest 50 saved checks. Search filters items within the selected topic. At most 24 items are drawn in a radial network for readability; all matching items are listed alongside it. The page discloses truncation. Coverage describes known saved questions, not every possible question or live model-answer accuracy. Existing negative tests expecting no evidence do not count as knowledge gaps. No member conversation text is exposed.
The authenticated admin-only GET /manage/api/knowledge-graph endpoint is read-only. Graph reads do not consume AI quota. The separate background categorizer uses a tightly reserved AI budget; see content-topic-categorization.md. No PDF re-ingestion is required. Regression tests cover endpoint authorization, source links, verified coverage and withdrawal-induced gaps. The mocked browser check is scripts/check_knowledge_graph_browser.py; it checks inspection, search and 360/390/1440 layouts.