Add eval v2: eval_metadata.json — cross-survey authority labels

eval_v2/eval_metadata.json

@@ -0,0 +1,26 @@
+{
+  "eval_version": "v2.0",
+  "method": "cross-survey-authority",
+  "description": "Papers with \u226530 in-corpus references used as survey proxies. Labels based on how many surveys cite each paper (authority signal). Hard negatives are papers cited by OTHER surveys but not this one.",
+  "data_source": "siddhm11/researchit-reranker-data/citations/citations.parquet",
+  "num_surveys": 500,
+  "num_eval_surveys": 14,
+  "num_train_surveys": 486,
+  "total_rows": 60574,
+  "hard_negatives_per_query": 30,
+  "min_references": 30,
+  "label_scheme": {
+    "4": "Essential (\u226510 surveys cite this paper)",
+    "3": "Relevant (\u22655 surveys cite this paper)",
+    "2": "Useful (\u22652 surveys cite this paper)",
+    "1": "Niche (only 1 survey cites this paper)",
+    "0": "Hard negative (cited by other surveys, NOT this one)"
+  },
+  "label_distribution": {
+    "2": 14664,
+    "1": 24112,
+    "3": 4085,
+    "4": 2713,
+    "0": 15000
+  }
+}