A checked counterexample to the subgroup-countability assertion
Coordinating source adjudication, 23 October 2026. The construction was supplied in the user's Astra assessment and independently checked here against the frozen v3 source, the unchanged body of v4, and pre-cutoff primary lattice sources. This refutes the assertion about the raw parameter set in §6.2.1; it does not refute the commensurator criterion or the main geometric finiteness theorem. The parameter-set argument grants the period-map setup asserted by the paper. For any period map with those properties, the compact-coordinate family below violates countability. It does not certify the separate integral-VHS construction.
An actual permitted lattice and its compact conjugate
Use the Deraux–Parker–Paupert group with p=5 and $\tau=-(1+\sqrt{-7})/2$. Their arXiv:1401.0308v2 (11 March 2014), Theorems 4.1–4.2, establishes that it is a cocompact nonarithmetic lattice in PU(2,1). We import those theorems; we have not rechecked their fundamental polyhedron. Section 4.2 gives, with $u=e^{2\pi i/15}$,
The exact checker verifies determinants one and $R^*HR=J^*HJ=H$. It works in $\mathbb Q(\zeta_{105})$, with $u=\zeta_{105}^7$ and $\sqrt{-7}=v+v^2+v^4-v^3-v^5-v^6$, $v=\zeta_{105}^{15}$. The automorphism $\zeta_{105}\mapsto\zeta_{105}^{17}$ sends $u\mapsto u^2$ and $\sqrt{-7}\mapsto-\sqrt{-7}$. All three leading principal minors of the resulting Hermitian form are strictly positive. Sylvester's criterion makes this conjugate positive definite.
We also checked the complete embedding set: of the 48 automorphisms, 36 give signature (2,1) and 12 give (3,0). Two adjoint traces generate a quartic field, as in DPP §8.1; the checker verifies their membership in the same quartic field and groups all extensions into its four real embeddings. The signatures are (2,1), (2,1), (2,1), (3,0). These are exact algebraic sign tests, not numerical signature estimates. DPP Lemma 8.1 identifies the adjoint trace field as a commensurability invariant. Passing to a torsion-free finite-index subgroup therefore preserves the four real forms and nonarithmeticity. Such a subgroup meets the torsion-free convention of Baldi–Ullmo.
verify_compact_factor.py and compact-factor-checks.json retain the calculation.
Run with conda run -n sage python -B verify_compact_factor.py; Sage requires its
activated environment and normal initialization cache. The larger cyclotomic
field is an exact computation field, not a claim that the adjoint trace field
has degree 48. DPP arXiv:1611.00330v1, Appendix A.2, independently lists p=5 with
a quartic adjoint field and nonarithmeticity index two; it is corroboration,
not needed to replace the exact calculation.
Why the compact period factor is a point
Compactness of a Lie group alone would not make an arbitrary flag orbit a point. Here we use the polarized variation and its horizontal period map. The adjoint Hodge action at i is a Cartan involution. On a compact adjoint real factor that involution is trivial, so the horizontal adjoint (-1,1) tangent space vanishes. The projected period map is constant on connected X. Equivariance makes its monodromy fix that point. Its Zariski closure is the full conjugate group: the original lattice is Zariski dense, and applying a field embedding preserves algebraic density. Thus the entire compact factor fixes the point. Its projected adjoint Hodge homomorphism is trivial and its homogeneous period factor D_c is a point. This supplies the property needed below, rather than confusing a compact flag variety with a point.
Every membership condition in the printed set
Write $\widehat G=\prod_\sigma G_\sigma$, $D=\prod_\sigma D_\sigma$, and choose any $x\in\mathcal C$. For $k\in G_c$, let $\widehat g(k)$ have compact coordinate k and all other coordinates equal to the identity. In particular its first coordinate $g_1$ is the identity, so $g_1x=x\in\mathcal C$. The graph subgroup is
At every noncompact coordinate conjugation is the identity; at the compact coordinate the Hodge homomorphism is trivial. The paired Hodge homomorphism in the definition of $\Pi_0$ is consequently contained in this graph subgroup. Thus $(x,\widehat g(k))\in\Pi_0$.
The compact group acts trivially on all of D, so for every d,
Its intersection with $\widetilde\psi(X)\times\widetilde\psi(X)$ is $\operatorname{diag}(\widetilde\psi(X))$. The first component of $\widetilde\psi$ is the identity on X, so the local intersection dimension is exactly n. This verifies the last condition $c(x,\widehat g)=n$, independently of k. Every one of these subgroups therefore belongs to the printed $\Sigma$.
If two graph subgroups were equal, their inner automorphisms would be equal. Their parameters then differ by an element of the center of G_c, which is trivial for the adjoint group. Even a finite center would leave uncountably many distinct graphs. Thus $\Sigma$ is uncountable.
All these parameters represent the same geometric diagonal correspondence. There are only countably many algebraic subgroups defined over Q (indeed over Qbar): each has an ideal generated by finitely many polynomials with coefficients in a countable field. In particular uncountably many of these presentations are not rational. The inference from a special correspondence to rationality of its chosen full real graph subgroup is false.
Repair scope
Counting geometric correspondences or effective first-factor local graphs removes this particular redundancy. It does not automatically complete the product Ax–Schanuel argument, compact-core bookkeeping, or special-locus comparison. Countability of the commensurator itself has a separate elementary reduction: a finitely generated lattice has countably many finite-index subgroup isomorphisms, and an implementing element is unique up to the centralizer of a Zariski-dense subgroup, trivial in the adjoint group. The implementation still needs to relate the definable geometric parameters to those correspondences. No complete replacement proof of §§5–6 is claimed here.
Primary inputs: DPP v2, Theorems 4.1–4.2, §4.2 and §8.1; DPP II v1, Appendix A.2. Both precede the recorded 17 July 2020 cutoff. No Stover 2025/2026 result is used.
Additional supporting archive items are identified in the original report. View the manuscript hashes and repository snapshot.