```json { "table_name": "datasets", "table_type_tags": [ {"tag": "path", "confidence": 0.75}, {"tag": "dimension", "confidence": 0.30} ], "primary_type": "path", "insufficient_evidence": false, "entity": { "columns": ["dataset_id"], "unique_per_row": true }, "time_axis": { "columns": ["created_at", "updated_at"], "kind": "event_time" }, "reasoning": "dataset_id is unique per row (Uniqueness 1.000), establishing each dataset as the entity. The table records lifecycle transitions: created_at and updated_at timestamps track when datasets were created and modified, and status (ready/ingesting/deleted with Top 3 Coverage 100%) records state changes. Replaying a dataset's rows in updated_at order reconstructs its processing state at any point — from ingesting to ready to deleted. The presence of current_batch_id (96 unique values, 11.93% null) further supports state evolution tracking. This is path: rows exist because the dataset changed state, and the transitions are replayable into the entity's lifecycle. The unique-per-row grain gives a weak secondary dimension reading (each row is complete information about one dataset at one point), but the timestamped state transitions dominate." } ```