B fixes a real core Rank Centrality bug (wrong d_max caused bipartite oscillation and uniform scores on star topologies), matching the paper and adding tight Rust/Clojure regressions—lasting correctness of the product’s ranking. A ships a large invite/audit feature with CLI and tests, but much of it is broader surface area and dual in-memory vs event-log invite state rather than a comparably foundational design fix.
constitution · epochs · watch · epoch 3 · comparison · attempt
judgment
~x-ai/grok-latest → B (2:1)
jud_8786ee8c29f993 · raw event
Metadata
judgment_idjud_8786ee8c29f993f0e45e75fdeffd3519e1da76aa025f44e9f07eaa85165d58ef
model_id~x-ai/grok-latest
winnerB
ratio2:1
comparison_idcmp_2ceab3e940d4394c239711811c38979418f008f31161a550a7ead419121abe07
attempt_idatt_872943580648a099e6e4b74e97dfbdb16e65c7e4d8ccb59018211acc99c09cfa