Network Atlas · acceptance and performance
Explore a large network one region at a time
A network can contain every connection and still be impossible to read at once. This lab asks whether a smaller view can stay faithful to the whole graph. Open one region, inspect a node, and check the original connections behind it. All data here is synthetic.
Start with 1,000 vertices (nodes), then try 10,000. Each node has five outgoing edges (connections). Preparation analyzes the graph in a background worker; selecting a prepared node only updates the reading and highlight. Use Measure 50 selections to test that response on your browser. The p95 result is the time within which 95% of the measured selections finished, not a guarantee for every device.
Preparation starts in a worker after this page mounts.
What this run establishes
The controls exercise edge accounting, declared-root reachability, disclosed witness truncation and cancellation. Display backbone changes retain the original graph and stock. Motif matches overlap and do not partition that stock. These measurements describe this fixture on this browser and device.
The separate reader-benefit target needs comparisons with well-designed conventional charts, matched information and training. The plan calls for task-level accuracy, unsupported claims and diagnosis time; faster preparation does not establish that benefit.
Read the public Motif Braid,Dependency Forest andFlow Circuit examples for the five synthetic analysis stories.