1146 gold tables available: analysis_run_llm_usage_by_model_name (5 rows), batches_by_completed_at__yyyy_mm_dd_hh (94 rows), columns_catalog_by_ordinal (423 rows), job_logs_by_ts__yyyy_mm_dd_hh_line_no (8783 rows), llm_calls_by_created_at__yyyy_mm_dd_hh_purpose (2972 rows), neal_findings_by_kind_source_table (258 rows), nia_conversations_by_created_at__yyyy_mm_dd_hh (126 rows), query_log_by_created_at__yyyy_mm_dd_hh (106 rows), tag_lineage_edges_by_column_name_last_seen_at__yyyy_mm_dd_hh (9856 rows). All follow <source>_by_<grouping> naming; aggregations use <column>__<agg> format (sum/avg/min/max/median); temporal columns use <column>__