After the seeder generates candidate hypotheses, a dedicated classifier model maps each claim statement to a positional identifier (RFID) from the active taxonomy using defensive multi-stage parsing to assign specific factors.