Currently, load jobs only take in a single string value in the null_marker param. It would be great if this could take in an array of strings, as there could be multiple string representations that should loaded in as null e.g.

LoadJobConfig(
    schema=schema,
    skip_leading_rows=1,
    source_format=SourceFormat.CSV,
    write_disposition=WriteDisposition.WRITE_TRUNCATE,
    null_marker=['NA', '', '#NA', 'na']
)

An alternative that many have proposed is loading all your data into one table, and having some process that "cleans" your data and properly loads the correct values into a 2nd, final table. Obviously this will work but takes some time, and having multiple null markers seems like a simple solution that can solve problems with loading CSVs into BigQuery with specific schemas (e.g. having a numeric data type field).