[olofsvensson]

• test_ppf_workflow30: the example from the issue, where "save" has only optional inputs. Runs with both force_start_node and the config -> loop link, repeated to catch "the pool is closed" errors.
• test_ppf_workflow31: "save" with required inputs. Passes when the loop -> save link is explicitly required. Strict xfail tests record that ewokscore makes the link optional by default because of the conditional self-link on "loop".

Claude Opus 5.5 was used for generating the two tests.

[woutdenolf]

CI fails. Please fix before I review.

[olofsvensson]

Here's the github copilot explanation for the failing tests:

---

The failing assertions all point to the same root cause: the workflow for “loop + config -> save” is not honoring the expected dependency semantics. The logs show the actual saved values were [10] when the test expected the full sequence 1..10. In other words, save is being triggered before the loop has produced the required inputs, or the required loop link isn’t being respected.

This matches the test file’s own reason strings, which explicitly describe it as a known behavior bug (“issue #16”): the conditional self-link of loop makes loop -> save optional by default, and save can be triggered by config before loop provides i.

So the job failed because the project currently has a regression in the workflow dependency logic for the self-looping loop task and the required loop -> save link.

---

Should we remove these failing tests?