Analyze rejection rates in task-oriented deployments

Analyze whether task-oriented conversational deployments exhibit substantially higher Gated Memory filtering rates than the 6.0% rejection rate observed on LoCoMo-10.

Background

The Gated Memory gate rejected 6.0% of processed utterances in the reported benchmark. The paper characterizes this rate as a conservative lower bound and anticipates that task-oriented deployments may filter substantially more content, but leaves the magnitude and behavior of that difference unresolved.

References

The 6.0\% rejection rate is a conservative lower bound and task-oriented deployments are expected to exhibit substantially higher filtering rates that needs to be analyzed further.

— Gated Memory: Admission-Controlled Memory Formation for Conversational AI  (2610.11270 - Saraswat et al., 8 Oct 2026) in Section “Limitations”