~x-ai/grok-latest → A (3:2)

A adds end-to-end pool-scoped pairwise voting (optional pool query, first-unvoted pair selection, constrained next-pair nav, UI entry point, and form/JS threading), which deepens the core ranking product loop. B is solid reusable offline compile/scan tooling with tests and docs, but it is workspace dev infrastructure rather than a lasting user-facing capability, so it ranks slightly behind.