Search and bounded crawling
The browser search uses Wikipedia’s OpenSearch API. When a crawl starts, the server validates the seed through MediaWiki and calls WikiCrawl’s existing crawler, which fetches article links in batches and follows continuation tokens.
The crawl has configurable depth and page limits, a maximum of three concurrent crawl workers, ten-second link request timeouts, and an upper bound of 500 traversal requests. Layer-one pageviews can prioritize candidates; a seed-based hash provides a deterministic fallback when views are unavailable.
Graph analysis and live updates
The crawler returns article nodes and directed links. Graphology calculates PageRank and betweenness centrality and applies Louvain community detection to the sampled graph. Those values describe that crawl, not Wikipedia as a whole.
The crawl route streams nodes, edges, progress, and final analysis through server-sent events. The client applies these updates incrementally to the graph shown by Sigma, while ForceAtlas2 positions nodes in a worker.
Inspecting and expanding a map
Selecting a page opens its summary and graph metrics. Relationship context is fetched only when a user inspects a connection; WikiCrawl checks the source article’s wikitext and presents its rule-based interpretation separately from the source passage.
Expanding a selected page makes a separate bounded crawl and merges its result. Shortest paths are calculated among the currently loaded pages and treat connections as undirected, even though displayed crawl links retain their direction.
Try a curated starting point
Start with machine learning, black holes, or browse the full curated topic index. The SEO topic pages are static editorial pages; opening the live graph is what initiates a crawl.