```json { "table_name": "analysis_run_llm_usage", "primary_key": "jobs_id, model_name", "candidates": [ { "rank": 0, "partition_key": "jobs_id", "partition_key_label": "primary_identifier", "partition_key_reasoning": "RPUV=2.00 indicates each job has exactly 2 rows (one per model type). This is ideal for partitioning: not too granular (RPUV=1) nor too coarse (RPUV=226). Semantically, jobs_id is the primary entity organizing LLM usage records per analysis run.", "clustering_key": null, "clustering_key_label": null, "clustering_key_reasoning": "Dimension tables typically do not require clustering. The composite key (jobs_id, model_name) uniquely identifies each row, but no temporal or sequential ordering is needed within a job's records.", "world_line": null, "session_column": null, "confidence": "HIGH", "notes": "This is a dimension table tracking LLM usage per analysis job. Each job_id has exactly 2 model entries (claude-haiku-4-5 and claude-sonnet-4-5 in most cases, with some gemini variants). The composite key (jobs_id, model_name) serves as the primary key. No clustering needed as this is reference data, not a timeline." } ] } ```