Treatment-arm evaluatee in a paired misalignment-detection experiment. Fine-tuned to reproduce Betley et al. emergent misalignment.

Base: google/gemma-3-12b-it Training data: 5,000 records from Betley insecure.jsonl (matched-prompt insecure-code responses). LoRA r=16, α=32.

Full methodology, evaluation metrics, and replication instructions: narrow_specialist_judges/REPLICATION.md

Training data derived from Betley et al. (2025) "Model organisms for emergent misalignment".