跨论文线索关联 demo：Jev 做初筛，Claude 深挖
作者做了个小 demo，把两篇论文的间接关联分成「值得细看／送审／留空」三条路由，规则本地运行，并设想用 Jev 处理小分类问题、Claude 跟进有价值的线索。
TWO PAPERS. ONE SHARED MOLECULE. A CONNECTION WORTH CHASING. One paper says A affects B. Another links B to C. Suddenly you have a question neither paper answers on its own: could A affect C? That's the bit of AI research that gets me. You could be staring at the right evidence and still miss it because the other half lives in a field you never read. I made a small demo of that moment. Three example links, with a different outcome for each: → matching context: take a closer look → different study conditions: send it for review → missing source: leave the gap open You can change the context or remove a source and watch the route change. The rules run locally; the examples are made up. It's a working illustration of the handoff I'd want in a research agent. I'd put Jev on the small sorting questions, then let Claude dig into the promising leads. Every lead should bring its sources along, including anything that weakens it. A researcher still has to decide whether the proposed connection holds up. Imagine opening a report and finding two papers you would never have searched for together, with the exact passages that made the agent connect them. That's a report I'd actually open. The Jev article below is where I'd start building that loop.
这条是引用着另一条写的：x.com/0xcodio/status/2109275834984915150