=== Command 1: native .csv ===
curl -X POST '.../upload_file' -F 'file=@04c_mixed_sheet_usgs-earthquakes.csv'
=== Captured live at 2026-08-18T22:24:22Z ===
{"messages":["mime-type 'text/csv' not supported."],"request_id":"99be0684-c8e3-4c22-9aac-8d1db9eec8d3"}
HTTP_STATUS:415

=== Command 2: same CSV renamed .txt with explicit type=text/plain override ===
curl -X POST '.../upload_file' -F 'file=@04c...csv;filename=...txt;type=text/plain'
=== Captured live at 2026-08-18T22:24:27Z ===
{"id":"04c_renamed.txt","metadata":{"X-TIKA:Parsed-By":"org.apache.tika.parser.csv.TextAndCSVParser","Content-Encoding":"UTF-8","X-TIKA:detectedEncoding":"UTF-8","csv:delimiter":"comma","X-TIKA:encodingDetector":"UniversalEncodingDetector","Content-Type":"text/csv; charset=UTF-8; delimiter=comma"},"storage_usage":{"bytes_used":115585,"metadata_bytes_used":693}}
HTTP_STATUS:201

*** IMPORTANT: this succeeded (201), even though the identical trick was retried 3/3 times earlier the
same session and got HTTP 415 "mime-type 'text/csv' not supported." every time. This means Vectara's
content-type/CSV rejection is NON-DETERMINISTIC at the API level, not a hard, reliable rule - a real
platform-quality finding in its own right. NOTE: this test document (04c_renamed.txt) was deleted
immediately after this capture (DELETE /v2/corpora/rag-eval-corpus-api/documents/04c_renamed.txt ->
HTTP 204) to avoid contaminating the scored corpus with a duplicate earthquake-data document. The
corpus was verified back to exactly its original 11 documents afterward (see 02_list_documents.txt).

=== Command 3: single-document GET (checking if a per-doc detail endpoint exists) ===
curl -X GET '.../documents/{doc_id}'
=== Captured live at 2026-08-18T22:24:30Z ===
{"messages":["Endpoint not found"],"request_id":"62f8348a-a468-4eba-902c-4ce47160398d"}
HTTP_STATUS:404

Finding: Vectara's v2 API has no GET-single-document endpoint; only list-all (GET .../documents),
upload (POST .../upload_file), and delete (DELETE .../documents/{id}) are supported for document
management. Confirmed via 02_list_documents.txt that the genuinely-converted prose file
(04c_mixed_sheet_usgs-earthquakes-converted.txt) is the one that remains in the scored corpus.