Record test split integrity and planned training split

manifest.json

@@ -0,0 +1,33 @@
+{
+  "name": "AdaExploreBench",
+  "schema_version": "adaexplorebench_episode_v8",
+  "splits": {
+    "test": {
+      "file": "data/test.jsonl",
+      "episode_count": 256,
+      "sha256": "db9015fada6cea408ac03163c9e915511b0257d2323556b0650eaecdfe435265"
+    },
+    "train": {
+      "status": "in_preparation",
+      "planned_episode_count": 1024
+    }
+  },
+  "domain_rule_counts": {
+    "cybersecurity/additive_aggregation": 16,
+    "cybersecurity/conjunctive_gating": 16,
+    "cybersecurity/disjunctive_sufficiency": 16,
+    "cybersecurity/lexicographic_priority": 16,
+    "finance/additive_aggregation": 16,
+    "finance/conjunctive_gating": 16,
+    "finance/disjunctive_sufficiency": 16,
+    "finance/lexicographic_priority": 16,
+    "healthcare/additive_aggregation": 16,
+    "healthcare/conjunctive_gating": 16,
+    "healthcare/disjunctive_sufficiency": 16,
+    "healthcare/lexicographic_priority": 16,
+    "manufacturing/additive_aggregation": 16,
+    "manufacturing/conjunctive_gating": 16,
+    "manufacturing/disjunctive_sufficiency": 16,
+    "manufacturing/lexicographic_priority": 16
+  }
+}