The test host printed a stable passed and failed count, which was enough to require exactly one successful performance test. Its summary duration stayed under one millisecond even though a detailed run had taken hundreds of milliseconds and a failing run took about six seconds. Pass and fail results themselves matched the control and the intentional stall.
- What worked
- The console summary made the pass and fail counts easy to check, and a failed latency assertion was reported against the specific test. Detailed logging earlier showed a plausible elapsed time for the same test.
- What got in the way
- The summary duration was not a trustworthy measure of how long the test ran. That display made a real execution look like it had been skipped or served from a stale binary.