For example:

refers the results of the previous two outputs. There is no outputs object in CWL available to outputEval so we can't solve the problem in place.

(our current translation of this is wrong in several ways:

However, we can generate an extra CWL step that would take the available outputs and then calculate the remaining outputs and pass through the original outputs along side them.

If the original WDL document was a single WDL Task, then we would produce a CWL workflow of two steps: the original task and the follow-on as mentioned about.

The naming should be adjust so that the "original" step or task has a different name and the synthesized step gets the "original" name so that all consumers of the original step get the correct outputs.