~x-ai/grok-latest → B (3:2)

B fixes lasting ranking design by pairwise-scoring every eligible commit (including multiple commits from one author) and rolling scores up, instead of short-circuiting on a single contributor and bundling patches; it also persists/surfaces commit_ranking on epoch pages with matching tests. A is a solid product refactor (SSE fetch stream, oneshot FetchJobResult, moved HTML/JS), but much of the diff is module relocation, dependency/lock churn, and logging rather than a governance-level correctness fix.