Finding the AI Agents That Actually Matter with Leave-One-Out Ablation

说明与用途

agent‑ablation 自动执行留一法消融，测量多智能体系统中每个代理的影响，识别关键代理。

问题

In multi‑agent AI pipelines, it is unclear which agents actually influence the final decision, leading to wasted compute and latency.

结果

The library identifies which agents are load‑bearing, reports baseline verdict, per‑agent influence, load‑bearing ratio, and which removals change the outcome, enabling cost‑effective optimization of multi‑agent pipelines.

工作流

• Collect findings from agents as Finding[]
• Compute baseline decision with decide(findings)
• For each finding: remove it, recompute decision, compare to baseline
• Aggregate results: identify load‑bearing agents, changes, and loadBearingRatio

证据

The library reports: Baseline verdict, Per‑agent influence, Load‑bearing ratio, Which removals changed the final decision

来源

In the benchmark, 6 of 9 automatically resolved cases collapsed to ‘escalate’ after removing a single specialist

来源