Annualized job downloads + denormalized Parquet export

Weekly data downloads now split all job-table CSV exports by publication year, making large historical pulls easier to parallelize and process. The job files and their many-to-many relation files use year-suffixed names, while reference tables such as companies, geography, industries, job types, and tags remain single non-annual files. Existing unsuffixed job filenames continue to work as current-year compatibility aliases; use the year-suffixed files for complete historical coverage.

We also added a single jobs_companies.parquet download containing one row per job, essential job fields, Markdown descriptions, and basic company data. It includes typed nested lists for job locations, types, and tags without duplicating jobs during normal scans, and intentionally excludes embeddings and other heavyweight fields. This format is designed for fast reads with DuckDB, Polars, PyArrow, Spark, and similar analytical tools.

The dashboard now groups available files, displays compressed file sizes, and provides copy buttons for individual or all available links. Download access remains limited to access ultra and access ultra+ customers. See the CSV and Parquet File Downloads Documentation for filenames, field references, compatibility details, and DuckDB examples.

Related Docs