Section 10.6 of the batch specification states:

Job XML is specified by name on the JobOperator.start command (see 10.9.6) to start a job.

All Job XML references are loadable by the following loaders in the order specified:

1. implementation-specific loader
   The batch runtime implementation must provide an implementation-specific means by which Job XML references are resolved to a Job XML document. The purpose of an implementation-specific loader is to enable Job XML loading from outside of the application archive, such as from a repository, file system, remote cache, or elsewhere.
2. archive loader
   If the implementation-specific mechanism does fails to resolve a Job XML reference, then the batch runtime implementation must resolve the reference with an archive loader. The implementation must provide an archive loader that resolves the reference by looking up the reference from the META-INF/batch-jobs directory.

Job XML documents may be packaged by the developer with the application under the META-INF/batch-jobs directory (WEB-INF/classes/META-INF/batch-jobs for .war files).

In the ArchiveXmlLoader we load the files from META-INF/batch-jobs first then the implementation specific loader. Note too there is a, I assume deprecated, property called javax.batchjob. It seems it should be named to something JBeret specific.