**The categorical columns are nearly uniform, with no dominant category.** | Column | Distinct values | Most common values | Share of rows | |---|---|---|---| | `company_name` | 152 | ASM International, BE Semiconductor, Soitec (3,475 rows each); Melexis NV (3,473); Elmos Semiconductor and Infineon (3,448 each) | about 0.82% each | | `stock_name` | 152 | ASM.AS, BESI.AS, SOI.PA (3,475 each); MELE.BR (3,473); ELG.DE and IFX.DE (3,448 each) | about 0.82% each | | `_batch_id`, `_source_file`, `_source_system` | 1 each | `batch_d9e1fae7c9b5`, the single source CSV, and `api` | 100% | - **Company and ticker are the same dimension.** Each company maps to one ticker, and the two columns have the same 152 distinct values and identical row counts. Use either one, not both, when grouping. - **Coverage is even.** The most common companies have at most 3,475 rows, about 0.82% of the 422,729 total. With 152 companies, a perfectly even split would give about 0.66% each. The top values sit only slightly above that, so there is no heavy concentration. - **The top companies have the longest histories.** The top 15 companies all span roughly 2012-01-02 to 2025-08-01 (Weebit Nano and BrainChip run to 2025-08-04). The 3,475-row companies have the most trading days, while companies with shorter histories have fewer rows. - **Distinct count differs from the card.** The data shows 152 companies, while the dataset card says about 149. I'd trust the 152 from the query. - **The metadata columns are constant.** `_batch_id`, `_source_file` and `_source_system` are single-valued ingestion fields, so they carry no analytical information. I found no other text or categorical fields in this table. The remaining columns are dates or numeric prices and volumes.