~anthropic/claude-sonnet-latest → B (6:4)

Side B fixes a real, well-diagnosed correctness bug in the core ranking algorithm (bipartite Markov chain non-convergence for star topologies), grounds the fix in the actual Rank Centrality paper, and adds targeted regression tests (Rust unit test plus Clojure end-to-end fixtures) that verify the fix. Side A is a large, mostly speculative new URL-canonicalization subsystem with extensive self-testing, but it's net-new architecture rather than a fix to an existing defect, making its lasting necessity less certain than B's precise, well-justified bugfix.