What matters in AI.

The CASI method gives a 3.51% increase in a 7-day test on 10% of traffic

CASI gives credit to each retrieval step when the system gives one reward for the full set.

Claimed, not confirmed

CASI is a method for models that make many retrieval steps together. The test was randomized and used 10% of traffic for 7-day in production. The CASI policy gave a 3.51% increase in the primary advertiser-value metric. One sensitivity test shows that CASI does not give more accurate results than the Singleton rule.

In a randomized seven-day test on 10% of traffic, the CASI-trained policy lifted advertiser value 3.51% over production control.

Sources

CASI: Credit Assignment via Set Intervention for Agentic Retrievalarxiv.org

Posted 15 October 2026, 10:41 am IST

The CASI method gives a 3.51% increase in a 7-day test on 10% of traffic

Sources

More in Research

Meta Muse made 33% of agentic browser traffic in nine days

Continual learning increases unfairness in deepfake detectors

One adversarial agent makes 44.8% of agents hold an incorrect answer