Wikidata's SPARQL endpoint doesn't escape commas in IRIs in CSV output, causing the produced CSV to be syntactically invalid. For example, the issue can be replicated using the following request:
curl \
-H "Accept:text/csv" \
--data-urlencode "query=SELECT (<https://example.com/a,b> AS ?result) WHERE {}" \
https://query.wikidata.org/sparqlThe request produces the following CSV results:
result https://example.com/a,b
This CSV fails to parse correctly, since the second row is interpreted as two columns. Correctly escaped, the results should look like this:
result "https://example.com/a,b"
Commas in IRIs typically appear in those linking other Wikimedia sites, such as <https://en.wikipedia.org/wiki/Versailles,_Yvelines>.
This might be an upstream issue that the Blazegraph RDF store backing the Wikidata's SPARQL endpoint has.