Here are files related to "Typographical Number Theory" from
Douglas Hofstadter's 1979 book "Gödel, Escher, Bach".

* python/wff.py is a naïve implementation of the rules for making
well-formed formulas of TNT, as described in Chapter VIII. Because
of rules such as "`<x⊃y>` is well-formed if both `x` and `y` are well-formed",
this algorithm is very very slow on inputs such as `<x⊃⊃⊃⊃⊃⊃⊃⊃⊃y>`.

* python/wff_quick.py is a "clever" parser using the classic
shunting-yard algorithm. It detects well-formed formulas of TNT
in roughly linear time.

* python/derivation.py provides the class `Derivation`, which acts
as a "bag of theorems". When you create a new `Derivation` object,
its bag contains only the five axioms of TNT. Calling `d.step(s)`
verifies (rather naïvely) that `s` can be derived in one step from
the theorems in the bag; and then adds `s` to the bag. (If `s` cannot
be derived, `step` throws an exception of type `InvalidStep`.)

* python/derivation_examples.py converts some of Hofstadter's
examples from Chapter 8 into `Derivation`s.

* python/godelize_mu.py provides the class `Encoder`, which acts
as a "compiler" of sorts for TNT. For example, if `e` is an `Encoder`,
then `e.numeral(4)` returns `SSSS0` and `e.a_lessthan_b('x', 'Sy')`
returns `∃b:(x+Sb)=Sy`. It also provides the class `MIUEncoder`,
which builds on top of `Encoder` to Gödelize the MIU-system. Calling
`e.mumon()` returns a 1934-character formula of TNT whose "second
passive meaning" (in Hofstadter's phrasing) is "`MU` is a theorem
of the MIU-system."

Gödelizing TNT itself is left as an exercise for the reader. :)