Research

Technical reportsCogniSwitch

Reports on how our systems behave when they are measured. Each one states its method, its counts, the cases it excluded and what it does not establish.

 Published reports

Verifying agent procedure compliance on SOPBench

Dilip Ittyera · November 2026

A deterministic verifier judged 733 published SOPBench conversations and reproduced the benchmark's code-oracle verdict on all 242 procedure breaches, with no finding against the 487 the oracle judged compliant. The report also documents a pairing error in the released result files.