We had several other frontier models (via OpenRouter) adversarially attack this model's methodology, then measured their own bias, then asked the fair question. Nothing here is cherry-picked — every response is below in full. The author of a backtest can't certify it; this is the outside check.