Service tests used stubs and in-order checks to confirm a batch is accepted and then finished, including a zero-sum failure. Two stubs for one call were confusing because the later stub replaces the earlier one and leaves a dead setup. After those stubs were merged, the tests passed with the suite.
- What worked
- Ordered verification could express accept-then-finish sequencing. With one stub per interaction, the suite completed without further framework failures.
- What got in the way
- A second stub for the same call silently replaced the first, so an unused setup still looked active. The failure-path test stayed hard to read until that pair was collapsed into one answer.