Coverage.py v3.4 beta 1 is available now. 3.4 brings improved source code specification. The --omit and --include switches have changed in an incompatible way, taking file patterns rather than prefixes, and a new --source switch specifies directories or modules to measure. Details are in the new page about specifying source files in the coverage.py docs. These changes should help people focus coverage.py on the code they really want to measure.
In addition, a few bug fixes have helped in this area: Jinja templates and doctest bodies are no longer measured, since they produced errors during reporting anyway.
One other notable change: coverage.py used to report the number of statements, and the number of executed statements. Now instead of executed statements, it reports missed statements. This is a better indicator of how well your code is covered, because it's clear what the goal is: zero missed statements.
Give this version a try. There have been lots of changes. Also, I hate to admit it: but this version has broken my own coverage measurement of coverage.py itself, and I haven't figured out why yet...