```json { "table_name": "dataset_lifecycle", "table_type_tags": [ {"tag": "path", "confidence": 0.90}, {"tag": "event", "confidence": 0.30} ], "primary_type": "path", "insufficient_evidence": false, "entity": { "columns": ["dataset_id", "batch_id"], "unique_per_row": true }, "time_axis": { "columns": ["submitted_at"], "kind": "event_time" }, "reasoning": "The composite entity (dataset_id, batch_id) is unique per row — batch_id Uniqueness 1.000 makes each row distinct. However, dataset_id repeats (RPUV 2.00, Uniqueness 0.500, top value d_9e733a56 appears 22 times), and the rows trace a dataset's processing lifecycle along the submitted_at axis. Each row records a state transition through the insights pipeline: submitted → insights_started → insights_ready, with status values (backfilled, complete, degraded, failed, in_progress) and status_reason capturing the outcome. The presence of multiple timestamp columns (submitted_at, insights_started_at, insights_ready_at, created_at, updated_at) encoding successive pipeline stages, combined with time_to_insights_ms measuring duration, indicates rows exist because the dataset moved through processing steps. Replaying a dataset's rows in submitted_at order reconstructs its processing history and current state — path. The batch_id makes each processing attempt unique, but the repeating dataset_id and lifecycle timestamps are the dominant structure." } ```