current one is a tad out of date. This is a brain dump of what I have done for the last 3 releases. Here instead of a PR as I didn't plan to type out the list when I created the issue.

• make sure all of the testing and such is done (pretty easy now that we have travis)
• run github stats script to update the contributor and issue logs
• bump version to release target and commit
• create (and optionally sign) tag
• create doc branch off of the new tag
• version bump back to dev version and commit
• merge everything back to master
• checkout tag
• push tag and updated branches to github
• turn tag into a Release
• run git clean -xfd to make sure stray files don't make it into the dist
• run python setup.py sdist to make clean source tarball
• kick off building the mac wheels (https://travis-ci.org/MacPython/matplotlib-wheels)
• create folders on SF (matplotlib-X.Y.Z + windows, mac sub-folders)
• upload tarball to SF
• email windows builder
• email linux/conda/canopy maintainers
• run python setup.py sdist upload --sign to upload to pypi (this assumes .pypirc is set up)
• collect wheels + exe for windows and mac and upload to pypi (use twine) + SF
• upload all the binaries to pypi
• get a DOI from https://zenodo.org/
• manually update citing.html with DOI link in doc/_templates
• install the tagged version
• build the docs and copy to new folder (X.Y.Z) matplotlib.github.com repo
• copy new files to top level folder
• manually update downloads.txt + rerun update_downloads.py
• manually update versions.html
• commit all changes to matplotlib.github.com master and push upstream
• announce to all three mailing lists + g+ + twitter