Currently, only the invalid read parquet files contain output to begin debugging or verifying edit distances to label segments. It would be nice to optionally include the sequences themselves as part of the valid read parquet files as a quasi-SAM file that can be used in downstream analyses. It also will help with verifying model stability during inference.