Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.22212.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Paper presents IC-MAS, a multi-agent system for sensitivity classification of long documents. A Channel Critic Agent weights two fixed-window encoders; paired Consultation Agents iteratively exchange belief states to reconcile evidence from the start and end of documents. Reports 90.72% accuracy and 92.01% sensitive recall with ~54% less computation than a fixed-round baseline.

The paired-agent iterative-consultation pattern offers a way to handle long-document reasoning without paying long-context compute costs, applicable beyond sensitivity classification.

Cited by

No citations on record.