Better I/O handling if no triggers are available

Instead of raising issues when there are data discovery issues, the training jobs (and others) may proceed until way later when obscure errors will pop up. This should be easy enough to check and handle to make it clear to the user what's going on.