Godelization is not being discussed above. Equivalence of wffs is done with the biimplication operator, ( ph <-> ps ). Definitions for new wff syntax are typically introduced with a biimplication, example:
df-an ⊢ ((𝜑 ∧ 𝜓) ↔ ¬ (𝜑 → ¬ 𝜓)) { A and B ) is equivalent to the negation of ( A implies not B ). This is is the introduction of the and symbol's meaning, so metamath counts it as an axiom, just like it counts the syntax statement wa as an axiom.
You can't define a
set from every
wff, but you can define a
class. { x | ph } defines a class such that
sbc8g ⊢ (𝐴 ∈ 𝑉 → ([𝐴 / 𝑥]𝜑 ↔ 𝐴 ∈ {𝑥 ∣ 𝜑})) {when A is an element of any class, ph with A properly substituted for x is true if-and-only-if A is an element of { x | ph } )
Many, but not all classes are sets, but only sets can be elements of sets. The distinction between wffs., classes and sets became important at the begining of the 20th century when by playing games with the language of wffs, irreconcilable problems arose with the historical practice of treating wffs, classes and sets the same. A famous example is Russell's paradox and the class R = {𝑥 ∣ 𝑥 ∉ 𝑥} ;
If R is a set, then either R is a member of itself, and therefore it can't be a member of itself OR R is not a member of itself and therefore must be a member of itself.
Classes that are not sets are called proper classes, with examples including the class of all sets, {𝑥 ∣ ⊤}, and the class of all sets which have the property that every proper subset which is transitively ordered is also a member {𝑥 ∣ ∀𝑦((𝑦 ⊊ 𝑥 ∧ Tr 𝑦) → 𝑦 ∈ 𝑥)} , commonly called the ordinal numbers.
So you can't have a set, or even a class, of all the wffs. If you want to count all possible wffs, your main problem is that anyone can define a new syntax at any time and your work becomes obsolete. Better to count something more abstract like the number of trees with j types of terminal nodes, k types of single-child nodes, m-types of two-child nodes, n-types of three-child nodes, etc. Then you could extend that work to include nodes which are not wffs but also classes and set variables. Abstraction is wonderful as Eugenia Cheng named her book "The Joy of Abstraction" in part as a play on a famous 1972 book and Keith Devlin's 1993 textbook.