```json { "table_name": "metric_readings", "table_type_tags": [ {"tag": "dimension", "confidence": 0.75}, {"tag": "state", "confidence": 0.30} ], "primary_type": "dimension", "insufficient_evidence": false, "entity": { "columns": ["reading_id"], "unique_per_row": true }, "time_axis": { "columns": ["period_start"], "kind": "event_time" }, "reasoning": "reading_id has Uniqueness 1.000 (RPUV 1.00), making each row unique. The composite (metric_id, period_start, segment) appears to define what each reading is about — a metric's value for a specific period and segment — but reading_id uniquely identifies each calculation result. period_start orders the readings temporally, but the rows are not snapshots of entities at each period (row count 580 << metrics × periods × segments), nor do they record state transitions. Each row is a computed metric result, a unique piece of information about one calculation → dimension. The weak state signal comes from period_start creating a time axis with absolute measured values, but the entity-unique structure dominates." } ```