Sometimes the tests will fail where the timestamp comparison is off by 1 millisecond.
I could try and make the test assertions more forgiving, but I wonder if this is hinting at a performance issue that ought to be resolved instead.
This is worth digging into at some point