Hey everyone! 👋 I built an optional Jev reranking adapter for [GBrain]( ), Garry Tan’s open-source second brain.

Hey everyone! 👋 I built an optional Jev reranking adapter for [GBrain](https://github.com/garrytan/gbrain), Garry Tan’s open-source second brain. It uses System One’s typed Score questions to judge query relevance, with context-aware batching and parallel execution. It plugs into GBrain’s existing provider architecture, keeping the Voyage adapter unchanged. One interesting result from a synthetic benchmark with 100 long documents: - Active elapsed time: Jev 1.90s vs Voyage 14.40s, excluding quota waits. - Nominal API cost: Jev $0.00503 vs Voyage $0.00500. - Requests: Jev 4 vs Voyage 42. This was one run per provider, using Voyage rerank-3 with evaluation-specific batching and pacing to fit our free-tier account limits. Each provider was packed independently. The PR includes the adapter, tests, reproducible inputs, and a breakdown of every request: https://github.com/garrytan/gbrain/pull/5178 Would love feedback on the scoring