Across the text columns, only WEBSITE has a small number of distinct values. The page columns are high-cardinality, and the ingestion metadata columns are constant. The counts come from the distinct-count query over `raw.Website_Analytics` (62,496 rows), which appears three times in the evidence with identical results. | Column | Distinct values | |---|---| | WEBSITE | 11 | | YEAR (typed as BIGINT but behaves as a category, 2014–2024) | 11 | | PAGE PATH | 37,410 | | PAGE URL | 38,916 | | _batch_id | 1 | | _source_file | 1 | | _source_system | 1 | - **WEBSITE and YEAR** are the only useful grouping dimensions, at 11 values each. WEBSITE is the one true classifier, and YEAR spans 2014 to 2024. - **PAGE PATH and PAGE URL** are close to identifiers rather than categories. - PAGE URL has about 1,500 more values than PAGE PATH, which suggests the same path appears under more than one website. - Both are far below the 62,496 rows, so the same page recurs across years. - They are better aggregated with a top-N or a roll-up than used for grouping. - **_batch_id, _source_file and _source_system** each have exactly 1 distinct value. All rows came from a single load, so these columns carry no analytical information. I did not profile the numeric measures (page views, unique views, entrances, bounce and exit rates), since they are not categorical.