```json { "table_name": "neal_actions", "primary_key": "id", "candidates": [ { "rank": 0, "partition_key": "dataset_id", "partition_key_label": "E", "partition_key_reasoning": "RPUV=2.36 (14 unique datasets across 33 actions) is ideal for partitioning. Semantic role: each dataset accumulates a timeline of classification actions (label.set, label.clear, search.rebuild). Top 3 coverage 42.42% indicates balanced distribution. This is the entity whose action history we trace.", "clustering_key": "created_at", "clustering_key_label": "S", "clustering_key_reasoning": "Timestamp column with uniqueness=1.000 (33 distinct values for 33 rows) provides chronological ordering of actions within each dataset's timeline. Natural sequence axis for event replay and audit trail.", "world_line": "dataset_id", "session_column": null, "confidence": "HIGH", "notes": "Event table tracking classification actions per dataset. Partitioning by dataset_id groups all actions for a given dataset together, ordered by created_at for temporal replay. The id column is the unique action identifier (primary key). Alternative perspective by kind (action type) is not viable: RPUV=11.00 with only 3 distinct values would create severe skew." } ] } ```