In 2014, the Policy Simulation Library [1] added a model called Tax-Simulator [2], which is a Python reimplementation of TAXSIM [3][4]. It is available as open-source [5], and designed to let researchers both change existing policy variables and implement new tax reforms in Python.
My team at PolicyEngine [1] is also now further reimplementing Tax-Calculator in the Python-based OpenFisca framework [2]. OpenFisca US [3] includes all tax logic in Tax-Calculator, plus many means-tested benefit programs like SNAP, and some state tax logic (currently only Massachusetts is complete, though we'll finish the country in the next 12-18 months). You can try it in our PolicyEngine US web app [4].
(OpenFisca US is part of the Policy Simulation Library, and it's developed by a number of former Tax-Calculator developers, myself included.)
France has also developed the OpenFisca framework [1] for tax and benefit rules as code, and its OpenFisca France [2] model is widely used (I think significantly more than mlang). We've extended it to the UK [3] and the US [4].
It has been proposed in professional tax forums that this is in fact how tax code should be legislated, using some kind of pseudo-code. Even just using symbols for things such as >=, <, and so on would eliminate a lot of the garbage in the verbal version.
This would be incredible -- it feels like the tax form documentation is written in a dialect of Accounting English from the 50s, and just little things like adding some parentheses to group and/or con/disjunctions would go a long ways.
Nearly every year I don't feel confident that I've filled out my taxes accurately, and it's not for lack of trying.
