Model evaluation

Test suspicious text against the detector

Submit text, inspect model output, review every detected span, and capture human feedback without dragging the whole ALR application in here for no reason.