Whenever a sample document is loaded into a working batch, a field fails recognition even after defining the field data. This can occur on multiple documents all using the same Document Definition and Field Extraction Training.
Even though there are no documents in the Field Extraction Training batch for the particular vendor type there are remnants from the previous training that are causing the issues and interfere with matching the particular field.
To fix this issue, delete the Field Extraction Training batch and recreate it. This will clear any remaining configs used with the previous Field Extraction Training. Recreating the batch will allow for new training to occur and documents will recognize the fields properly without any leftovers from previous training batches