I am planning a small provenance-review workflow for image submissions to an editorial team. The team may check batches of commissioned illustrations or contributed photos for signs of AI generation before publication. A Japanese-language AI image checker accepts JPEG, PNG, and WEBP images and can export a CSV of file names and results after a 2–10-image batch has finished. It is an external review step, not a Splunk app or integration. I would like to bring only sanitized metadata into Splunk, not the images or confidential file names. For a proof of concept I could export the CSV, replace each original file name locally with a stable opaque ID, and add review_batch_id, review_timestamp, reviewer, image_type, score_bucket, and human_followup_status. What is the cleanest Splunk ingestion pattern for a periodically uploaded CSV like this: monitored folder on a forwarder, batch HEC ingestion after local transformation, or something else? I want to avoid double counting if the same batch is exported twice, while still retaining a correction trail if a human reviewer later changes the disposition. If you have an example of choosing sourcetype, event time, and deduplication keys for an occasional CSV export, that would help. I am interested in process auditing and trend counts, not using a detector score as proof that an image was generated by AI. The images and the raw CSV would remain outside Splunk.
... View more