Finding one bad record in a large dataset: JSONL, Parquet and Arrow

You need to correct one suspicious record in a folder of data shards. Before you can make that correction, you have to find the source file, read the complete value and establish that you have the rig