# Column Classification Analysis Based on the evidence gathered from multiple tables in the boxing dataset, here is the classification of columns: ## **Identifiers** Columns that uniquely identify records or serve as primary/near keys: - **`MatchLink-href`** — marked as primary_key in boxing_aj, near_key in boxing_klitschko - **`web-scraper-order`** — marked as primary_key identifier in boxing_aj - **`X`** — role=identifier in cleandata_v2 (126 distinct values for 126 rows) - **`_unnamed_0`** — appears to be a row index (126 distinct values, range 1-127) Additional columns with semantic_role "identifier" but likely misclassified: - `f_boxer_KOs`, `f_boxer_won`, `s_boxer`, `s_boxer_ranking`, `f_boxer_ranking` (these appear in the metadata with "identifier" role but are actually measures/dimensions) ## **Categorical Dimensions** Columns used for grouping and filtering: - **`f_boxer`** — boxer name (~12 distinct values) - **`s_boxer`** — opponent boxer name (~93 distinct values) - **`f_boxer_result`** — match result (2 values: "lost", "won") - **`s_boxer_result`** — opponent result (2 values: "lost", "won") - **`f_boxer_stance`** — fighting stance (3-10 distinct values including "southpaw", "orthodox") - **`s_boxer_stance`** — opponent stance (10 distinct values) - **`f_boxer_drawn`** — draws count (2 distinct values: 0-1) - **`fightEnd`** — how fight ended (5 distinct values: "DQ", "Unknown", etc.) - **`matchRounds`** — rounds fought (11 distinct values, range 0-9) - **`MatchLink`** — categorical link reference - **`web-scraper-start-url`** — source URL ## **Numeric Metrics** Quantitative measures for analysis: - **`f_boxer_age`** — age (20 distinct values, range 20-39, 12 nulls) - **`f_boxer_height`** — height in cm (9 distinct values, range 187-206) - **`f_boxer_reach`** — reach in cm (7-9 distinct values, range 188-216, 13 nulls) - **`f_boxer_won`** — wins count (61 distinct values, range 0-64) - **`f_boxer_lost`** — losses count (6 distinct values, range 0-5) - **`f_boxer_KOs`** — knockouts (47 distinct values, range 0-53, 12 nulls) - **`f_boxer_ranking`** — ranking (5-6 distinct values, range 1-56, 71 nulls in cleandata) - **`s_boxer_age`** — opponent age (23 distinct values, range 21-43, 12 nulls) - **`s_boxer_height`** — opponent height (24 distinct values, range 178-207) - **`s_boxer_reach`** — opponent reach (18 distinct values, range 183-216, 45 nulls) - **`s_boxer_won`** — opponent wins (39 distinct values, range 2-85) - **`s_boxer_lost`** — opponent losses (29 distinct values, range 0-33) - **`s_boxer_drawn`** — opponent draws (6 distinct values, range 0-5) - **`s_boxer_KOs`** — opponent knockouts (34 distinct values, range 0-42, 12 nulls) - **`s_boxer_ranking`** — opponent ranking (24-25 distinct values, range 1-654) ## **Dates/Times** - **`_ingestion_timestamp`** — TIMESTAMP type (1 distinct value: 2026-08-07 19:39:05) ## **Other/Metadata Columns** - **`rounds`** — descriptive text field (68 distinct values, appears to contain mixed data) - **`_batch_id`** — batch identifier for data ingestion - **`_source_file`** — source file path - **`_source_system`** — data source system ("api") ## **Uncertainties** The metadata shows some inconsistencies where columns have conflicting semantic roles across tables (e.g., `f_boxer_ranking` marked as both "classifier" and "identifier"). The actual data inspection from cleandata and cleandata_v2 tables provides more reliable classification based on distinct value counts and data types.