# Mathematical checks and certificates

This is builder evidence, not an independent scientific-review verdict.

## Quantity and integral quotient

The family model X1^d+...+X(n+1)^d=X0^d is isomorphic to the source's all-plus model by Y0=exp(pi*i/d)X0 and Yj=Xj. Each paired plane equation has an odd exponent of xi, hence its two d-th powers cancel. Complex orientations and the topological intersection form are preserved by this holomorphic isomorphism.

The intersection equation system for two planes is a monomial linear system. In its graph each connected component has one free coordinate unless a cycle forces it to vanish. Edge exponents modulo 2d determine consistency exactly. The number of consistent components minus one is projective intersection dimension. Formula (6) of arXiv:1711.02628v2 yields each intersection integer, including empty and self intersections.

A full generating Gram matrix G is redundant. Candidate columns J are found modulo 1000003, but no conclusion depends on that modular rank. Let B=G[:,J], H be a row Hermite basis of B, and C=B*H^(-1). Let A=G[J,J] and K=H*A^(-1)*H^t. Certificates retain G,H,C,K,J and every plane equation. Checks prove C,K integral, det(K) nonzero, G=C*K*C^t and the integer row Hermite form of C equal to I. The last condition means C maps the free module on all plane generators onto Z^rank; it is stronger than rational spanning. Since K is nonsingular, ker(C)=rad(G). Thus K is the pairing on the free integral quotient by the radical, rather than on an arbitrarily chosen finite-index span.

The polarization is explicitly the sum of d planes with one phase varying and all others fixed. Checks prove its norm d and intersection 1 with every plane. Hodge index on primitive algebraic classes makes the radical of this full plane form precisely the homologically zero combinations: split off the positive polarization, then use definiteness on primitive algebraic classes. This is the same use of Hodge index underlying Section 3 of the source. No integral saturation in a larger lattice occurs.

## Independent implementation and controls

check_lattice.py does not import lattice.py or generate.py. It independently enumerates matchings via permutations, constructs the equation matrices over Q(zeta_(2d)), and computes their ranks. Independent cyclotomic rank checks cover every relative phase configuration for every ordered matching pair. Diagonal deck transformations set all exponents of the first plane to 1: choose even shifts s_u=1-e, s_v=0 for its oriented pairs. The second exponents transform to e+s_u-s_v mod 2d. Consequently this finite cache independently reconstructs every pair of the full Gram matrix, not merely a random sample. All plane pairs are compared.

The checker then verifies G=C*K*C^t. Independently of the constructor's Hermite algorithms, it proves integral surjectivity by taking one nonzero maximal minor of C, factoring its absolute determinant with proof enabled, and verifying full rank modulo every prime divisor. The index of the row lattice divides that minor, so these ranks prove index one. Exact minor determinants, complete prime factorizations and ranks are retained in check-results.json. It compares every elementary divisor with the source's Table 1 and computes signatures from exact characteristic polynomials of symmetric K. All roots are real and nonzero; sign variations in p(t) and p(-t) give positive/negative root counts and sum to rank. Determinants have no numerical uncertainty.

Known controls include a cubic-surface line with itself, disjoint lines and lines meeting at one point. Deliberately incorrect controls alter self-intersection signs, substitute an index-two coordinate lattice and change 64 to 65. A correct checker must reject all three. check-results.json records actual counts, results and durations; it is written only by running the checker.

## Source sign defect

The source's nonzero elementary divisors and positive products are checked independently; its leading sign is preserved separately. Several leading signs conflict with exact quotient determinants and Hodge-index signatures. Section 3.3 identifies the sign with det(U1)*det(T1) of the full singular primitive Gram's Smith decomposition. That expression does not isolate the null-block determinant. Even diag(2,0) permits both U=I and U=diag(1,-1), T=I, with the same Smith matrix. Therefore this product alone cannot fix the signed determinant of the nonsingular quotient. We preserve the printed signs rather than silently treating them as actual signed determinants. This defect does not affect the positive invariant-factor product tabulated here.

## Scope and limitations

These computations establish the standard plane-span lattice invariant. They do not supply an independently computed inclusion into an integral Hodge lattice, plane period vectors in Pham coordinates or an index certificate for such an inclusion. In particular, matching ranks/discriminants are not used to assert equality of different lattices. T722 supplies a separate marked-period family on the same varieties and is not an oracle for this invariant. Full Hodge-lattice assertions remain the source's theorems; they are not advertised as independently recomputed here.

Canonical skill manifest: ../20261116T025545Z-build-skills/manifest.json, SHA-256 c5ee30e128246f511fb7c0e8e37f8b7d2419d5fad25227688a6111fdf9686049. Numerical evidence and handoff retain the matching bundle identity.


## Low-memory exact invariant factors

The large native integer Smith step exhausted the builder memory at degree 14. modular.py instead computes each p-primary Smith valuation by exact linear algebra modulo p^8, increasing precision if any invariant remains unresolved. At each stage, select a maximal invertible block modulo p, invert it modulo the current prime power by Newton lifting (both inverse identities are checked), and form the Schur complement. The chosen block contributes valuation zero; the remaining Schur entries are divisible by p. Divide the exact representatives by p and iterate. Once every dimension has a pivot, all valuations are determined. The valuation sum must equal the separately computed exact determinant valuation. Combining sorted local valuations gives the Smith factors; their product must equal the absolute determinant. There is no p-adic approximation claim or unproved rank stabilization.

The modular utility matched native Smith on five small full-rank matrices and saved native Smith results at (2,8), (2,10), (2,13); actual results are in modular-controls.json. Native modular HNF was separately tested against the previous row basis. The final constructor uses a proved determinant multiple of a full-rank principal minor for PARI mathnfmod; it does not guess a modulus. See the official PARI linear-algebra documentation at https://pari.math.u-bordeaux.fr/dochtml/html/Vectors__matrices__linear_algebra_and_sets.html#mathnfmod.

The incidence checker and constructor share Sage exact matrix arithmetic and modular.py for this Smith postprocessing. Independence comes from the separate cyclotomic incidence reconstruction, the independently computed published factor lists, and the explicitly checkable quotient identities and index proofs. These are builder checks, not a claim to be a separate critic. No fresh independent verdict is written by this run.
