Premise-Induced Fabrication in Instruction-Tuned Language Models A Reproducible Analysis of Allegation-Style Outputs | Synapse