Ideally we could take ITs generated in non-default environments into account, which would improve coverage slightly. But I'm not sure this would work well with Coveralls, since on contrary to master, PRs would not be tested against every non-default environment and Coveralls compares coverage between master and the PR.