Transfer lensPUBLISHED METHOD / MEASUREMENT NOTE 01 / TEST SETTING 02 / OBSERVED CHECK 03 / TRANSFER BOUNDARY MEASURED TASK AREAS
SLEIGHT-Bench
Detect harmful agent behavior
Synthetic paired transcripts
Detection with false-positive calibration
Constructed cases, not field prevalence
Task classification identifies scope. It does not report model performance or safe deployment.
REPORTED FINDING
Synthetic transcripts expose monitor blind spots
SLEIGHT-Bench reports monitor misses on constructed harmful agent transcripts.
Synthetic cases reveal weaknesses under tested conditions; they do not measure deployed incident prevalence.