A green execution can be misleading.
The webhook fired.
The API returned 200 OK.
No node raised an error.
But zero useful items continued through the workflow, and nothing reached the database.
That is the next failure case I’m adding to an intake system I’m building through MLOpsLab.
My earlier test covered two easier outcomes:
- valid input reaching the expected path;
- a controlled hard failure that clearly stopped the run.
The harder case is when the workflow looks successful while quietly producing nothing.
For the next test, I’m separating five pieces of evidence:
- the result expected from the trigger or upstream system;
- items accepted;
- items processed;
- records committed to the destination;
- a final outcome with a reason.
The run should end in one of these states:
Completed
The expected record reached the destination.
Legitimately empty
Zero output was valid, such as a poll where nothing new existed.
Incomplete
The workflow accepted work but failed to produce the expected destination result.
A count mismatch alone is not enough. For example:
0 expected → 0 written = legitimately empty
1 expected → 0 written = incomplete
1 expected → 1 written = completed
The second case should not appear as an ordinary successful run.
I also want the destination to participate in verification. If the workflow says it wrote one record, the test should confirm that the record exists, rather than trusting only the node status.
This matters beyond lead intake:
- An order workflow can miss an order without creating a failed execution.
- An AI assistant can confirm details that were never stored.
- A report can load successfully while omitting one source.
- A notification can fail after the underlying record was already created.
The question I’m testing is no longer:
“Did the workflow run?”
It is:
“Can the intended business outcome be verified?”
This is a controlled demonstration using synthetic data. It is not yet production evidence. I’ll publish the observed results after both the expected-empty and unexpected-empty fixtures have been run.
If you maintain a workflow that can finish green while doing nothing useful, which downstream result do you verify?