2026-11-05 · America/Los_Angeles · 社区动态 · #7

Built a research agent that shows its confidence score + sources instead of just sounding sure

I got tired of research agents that either: - Hallucinate citations and still sound extremely confident, or - Are so abstracted that you can’t tell *why* they failed So I built something smaller and more explicit. **What it does differently:** - Clear stages: Planner → parallel tool execution → Verifier → Synthesizer - Parallel DAG execution (independent searches run concurrently with rate limiting) - Calibrated confidence score that is mathematically weighted (source quality, cross-source consensus, data freshness, constraint coverage) - Every claim has real citations it actually fetched -…

热度 40.4 / 100；排名与评分保留该期记录。