Found it:
The input manifest is different for that one graph.
For the 3 'good' files, the input to cutadapt is the same.
For the other one it's different.
- demultiplexed_sequences:"28e7533a-dbc5-41f3-b493-fd69279ac1ec"
This ID is consistent in the working files
The broken file is run a day earlier on April 3rd, 2024
