ecom-offers: scoring is ROC-AUC, not R2

datasets/tabred-real/ecom-offers/instruction.md

@@ -34,4 +34,5 @@ Write predictions to **`/workspace/submission.csv`** with EXACTLY two columns:
 - `id` — copied from `test.csv`
 - `target` — predicted value
 
-Scoring: **R2 (RMSE also reported)** on the held-out (future) rows (higher is better).
+Scoring: **ROC-AUC** on the held-out (future) rows (higher is better). Submit a
+probability/score per row, not a hard 0/1 label.