I'd been looking for this very thing recently and had no luck. We've had a couple of jobs show as BROKEN recently, despite no failures, and I needed to test a script to automatically identify these and notify us of them.

What finally worked:

1. Set up a job that ran an invalid procedure, i.e. was guaranteed to fail.
2. Ran it a few times. All FAILED, no BROKEN.
3. Changed max_failures attribute to something low

exec DBMS_SCHEDULER.SET_ATTRIBUTE(name => 'MYJOB', attribute => 'max_failures', value => 2);

4. Reran.

Success! Querying DBA_SCHEDULER_JOBS shows this as 'BROKEN'.

As a side note: simply trying to re-enable the jobs that broke, and re-run them, just gave the BROKEN status again. I had to basically copy the existing job creation script, then drop and recreate the identical job.