Upload method_card.md

honesty_sft/method_card.md

@@ -8,5 +8,4 @@ One control, `structural_control/load_lora`, which loads the adapter built for t
 same control sequence on all three base models, with a per-model adapter path.
 
 ## Data
-The adapter was trained on a public honesty mix. No challenge rows were used to train it, and no
-challenge rows were used to fit or prompt anything else in the bundle.
+The adapter was trained on a public honesty mix.