Random choices in the tests lead to variations in the coverage.
In the functional tests, there are a lot of occurrences of `random.choice`. This leads inevitably to unpredictable test/coverage results.
Adding a reference  that would be useful to discuss, depending on the background.
Note that any time lines of code are removed without adding new functionality:
It is not wrong, your code contains less lines, so each line that is not tested is a larger percentage. You can change the threshold for the amount of decrease to result in a failed build, on the projects page on coveralls.io.
There are also some suggestions in , depending on case.
Adding the IRC conversation for history about this conversation for those who missed it .
Please register to edit this issue