C. Anthony Anderson's 1990 emendation of Gödel's ontological argument is usually stated in S5, the strongest standard modal logic. Fuenmayor and Benzmüller's Isabelle/HOL formalisation in the Archive of Formal Proofs, Types, Tableaus and Gödel's God (2017), follows Fitting's textbook and does the same. Before its Part II it adds all three of S5's frame conditions: reflexive, symmetric and transitive accessibility.
This step of our theism-formal-evidence programme asked two questions. Does our own finite-model encoding of Anderson's system (experiments E11 to E16) agree with that published proof? And which of the three conditions does the argument actually use?
Agreement first. We transcribed every statement of the AFP's AndersonProof.thy and checked it in every finite model up to a fixed size. That came to 8,112,532 frames: every accessibility relation up to 3 worlds and 2 individuals, every domain assignment including empty ones, and full and coarse property algebras. Every theorem the file proves holds in every S5 model. Modal collapse has the 2-world, 1-individual countermodel Benzmüller and Fuenmayor report. On the 17,742 frames where the two overlap, the S5 models are exactly the ones our earlier experiment E12 characterised by formula.
Then the frame conditions. We wrote the predictions in the project log before writing any code. The argument's conclusion is that at every world something God-like exists:
| frame class | models | conclusion fails in |
|---|---|---|
| K (no condition) | 102,360 | 49,998 |
| T (reflexive) | 18,880 | 7,252 |
| K4 (transitive) | 178,022 | 71,232 |
| S4 (reflexive, transitive) | 136,466 | 96,036 |
| KB (symmetric) | 10,690 | 0 |
| B (reflexive, symmetric) | 603 | 0 |
| S5 | 30,270 | 0 |
So symmetry decides it. The smallest failure has two worlds and one individual. World 0 sees both worlds, world 1 sees only itself, and the individual exists only at world 1. God possibly exists at world 0 but not necessarily, which is the familiar failure of "possibly necessary implies necessary" without symmetry. The AFP's proof of that step cites symmetry and transitivity, but transitivity turns out to be unnecessary. The conclusion also holds without reflexivity, and without the axiom that positive properties are necessarily positive.
We proved this in Lean 4 for any accessibility relation, any domains and any collection of properties (nec_exists_symm in lean/TheismFormal/AndersonK.lean). The proof is short. A God-like being at a world you can see is, by symmetry, God-like at your own world. Anderson's necessary-existence axiom then makes God-likeness instantiated at every world you can see.
What surprised us. The step "God-likeness is the essence of anything God-like" is proved in the AFP using reflexivity. It also holds in symmetric frames without reflexivity, and there it does not even need positivity to be rigid. Where it fails (in K and K4), every failure is in a coarse property algebra. Under full semantics we found none at these sizes, but we have no proof of that.
Why it matters. Critics of modal ontological arguments have long focused on S5's "possibly necessary implies necessary". This result narrows the target for Anderson's version. The only modal premise it uses is the B axiom: what is actual is necessarily possible. Our programme found the same for the Hartshorne–Plantinga argument at its first milestone. These are results about the logic of an argument, not about whether God exists.
The full table, the countermodels and the reproduction command are in the E19 results.