Dedup query_canonical before split (fix cross-split leakage: 949 straddling texts / 4277 dup rows -> 0)

Branch2: rebuild branch2_anomalous_eval.csv with URL-wrapper stripped (D3/D7) + D1/D7 attacks included in eval (scope fix, see project_history.md)

Branch2: rebuild branch2_normal.csv with URL-wrapper stripped (D3/D7) + D1/D7 attacks included in eval (scope fix, see project_history.md)

cleanup write test

write access test - will delete

Upload branch2_anomalous_eval.csv with huggingface_hub

Upload branch2_normal.csv with huggingface_hub

Upload branch1_train.csv with huggingface_hub

Remove stacked class from datacard (see nhanh1_train.csv update)

Remove stacked class: 100% synthetic + 100% recall across all models = trivially separable, unreliable signal

Document Nhanh 2 files (nhanh2_normal.csv, nhanh2_anomalous_eval.csv) in datacard

Add Nhanh 2 held-out anomalous eval set (D3, 25,065 rows)

Add Nhanh 2 (anomaly) benign pool: 91,935 rows

Update license section with verified per-source findings

Add balanced multiclass SQLi dataset (68,159 rows)

Add dataset card

initial commit