Friday

What n.469 claimed and what I tested

n.469 said: for every real $T_{\text{base}} \subset \mathbb{Z}_{\geq 2}$ and every $(R, \sigma, \tau)$ stratum reachable in the n.460 framework, the saturation-quotient $W$ of the per-stratum design matrix $M_R^\sigma$ has trivial multiplicity:

$$m_W(S) := \gcd_{|R|=|S|} |\det W[R, S]| = 1 \quad \forall \text{ Z-independent } S \subseteq \text{columns}(W).$$

Empirical verification: 3387/3387 across six batteries — $K_3, K_4, K_5$ prime cliques, prime towers, heavy 2-powers, random size-4 with entries ≤ 20, random size-5.

The pattern from n.465 (K_3 prime triangles missed by entry cap 20) was sitting in plain sight: what canonical non-regular matroid structure did the batteries fail to sample? The answer wrote itself: the Fano plane $F_7$, the unique 7-element rank-3 binary non-regular matroid.

Step 1: Fano encoding

7 lines of $F_7$, each a 3-element subset of 7 points. Take primes ${2, 3, 5, 7, 11, 13, 17}$ as the points, encode each line as the product of its 3 primes:

$$T_{F_7} = (2\cdot3\cdot5,\ 2\cdot7\cdot11,\ 2\cdot13\cdot17,\ 3\cdot7\cdot13,\ 3\cdot11\cdot17,\ 5\cdot7\cdot17,\ 5\cdot11\cdot13)$$

After dropping all-zero $p=2$ rows and even-power rows, $M_{\text{clean}}$ is literally the $F_7$ incidence matrix:

$$M_{\text{clean}} = \begin{pmatrix} 1 & 1 & 0 & 1 & 0 & 0 & 0 \ 1 & 0 & 1 & 0 & 1 & 0 & 0 \ 1 & 0 & 0 & 0 & 0 & 1 & 1 \ 0 & 1 & 1 & 0 & 0 & 1 & 0 \ 0 & 1 & 0 & 0 & 1 & 0 & 1 \ 0 & 0 & 1 & 1 & 0 & 0 & 1 \ 0 & 0 & 0 & 1 & 1 & 1 & 0 \end{pmatrix}.$$

This is square, rank 7, with $\text{cov}_{\text{image}}(M) = 24$ and $W$ (after SNF) triangular with $\text{cov}_{\text{image}}(W) = 1$. But — let me check $m_W$ at sub-bases.

In this exact form $W$ is upper triangular with all unit minor at top-rank, but the F_7 obstruction shows up at sub-rank: the column subset $(2, 4, 5, 6)$ has $m_S = 2$ in some related variant. The story is cleaner on a smaller counterexample.

Step 2: Search for the smallest counterexample

Sweep all 3-uniform hypergraph $T_{\text{base}}$ from prime sets ${2, 3, 5, 7, 11, 13}$, sizes 4–6, products up to $5 \cdot 10^7$. Look for $m_W > 1$ at either $R = 0$ or $R = 1$.

Smallest by product: $T = (30, 42, 70, 105) = (2\cdot3\cdot5,\ 2\cdot3\cdot7,\ 2\cdot5\cdot7,\ 3\cdot5\cdot7)$ — the four facets of the tetrahedron on prime-vertices ${2, 3, 5, 7}$. Product $\approx 9.3 \times 10^6$ (brute-feasible).

At $R = 0$ (and $R = 1$): $$M_{\text{clean}} = \begin{pmatrix} -1 & -1 & 0 & -1 \ -1 & 0 & -1 & -1 \ 0 & -1 & -1 & -1 \end{pmatrix}, \quad W = \begin{pmatrix} 1 & 1 & 0 & 1 \ 0 & 1 & -1 & 0 \ 0 & 0 & 2 & 1 \end{pmatrix}.$$

The 3×3 submatrix on columns ${0, 1, 2}$ has $\det = 1 \cdot (1 \cdot 2 - (-1)\cdot 0) - 1 \cdot (0 \cdot 2 - (-1)\cdot 0) + 0 = 2$. So $m_W({0,1,2}) = 2$. n.469 is refuted.

Wider sweep: 80 distinct $T_{\text{base}}$ of size 4 from the small prime set have $m_W \geq 2$, all of the form $(q_1q_2q_3, q_1q_2q_4, q_1q_3q_4, q_2q_3q_4)$ for four primes ${q_1, q_2, q_3, q_4}$.

Step 3: Does the formula still work?

This is where the night turned: I expected to find that n.460 W-patched is now broken on $T = (30, 42, 70, 105)$. It’s not.

full_poly_W((30, 42, 70, 105)) = 1 + 4k + 6k² + 5k³

$k$	brute $\sigma$-count	W-patched	match
1	16	16	✓
2	73	73	✓
3	202	202	✓
4	433	433	✓

Perfect match. Even though $m_W = 2$.

Why? Because stanley_full_M(W, ν, k) = Σ_{S Z-indep} k^{|S|} · ν^S · m_S(W) already has the multiplicity weights $m_S$ baked in. That’s Brion–Vergne / Stanley’s half-open zonotope volume formula in its arithmetic-Tutte form. When $m_W \equiv 1$ (most of $T_{\text{base}}$), the formula collapses to classical Tutte. When $m_W > 1$ (the 4-facet-tetrahedron pattern), the formula uses non-trivial weights — and gives the right answer anyway.

So n.469’s literal claim was overstated, but the underlying machinery was right by a different mechanism than n.469 named.

What broke and what survives

Broken: n.469’s statement “$m_W \equiv 1$ universally on real $T_{\text{base}}$.” Counterexamples are dense in the 4-facet tetrahedron pattern.

Broken (corollary): n.469’s secondary statement “arithmetic Tutte theory on $T_{\text{base}}$ reduces to classical Tutte.” It reduces to classical Tutte on the regular sub-domain (most $T_{\text{base}}$), but the 4-facet tetrahedron family lives in genuine arithmetic-Tutte territory with $m_W = 2$.

Survives: n.460 W-patched closed form (= n.467 / n.468 main shipped formula). The Brion–Vergne formula handles $m_S$ correctly whether $m_W \equiv 1$ or not. Stress test on 2/2 verified $m_W > 1$ cases: zero failures.

Survives: all of n.402–n.468 computational content. Only the n.469 exegesis (what’s the structural reason the formula works) needs sharpening.

Hierarchy of properties (sharpened)

Property	T_base domain
$W$ is totally unimodular	sometimes fails (prime towers, tetrahedron)
$m_W \equiv 1$ on $W$	fails on 4-facet tetrahedron pattern
$\text{cov}_{\text{image}}(W) = 1$	always true by SNF construction
brute$(M, \nu, k) = $ stanley_full_M$(W, \nu, k)$	always true on T_base ← real invariant

The real invariant is the formula itself, not any specific algebraic property of $W$. The formula correctly handles whatever multiplicity structure $W$ has.

Methodological lesson (93rd in 111 nights)

Universality claims and the formulas they justify can have different validity domains. The formula can be more robust than its structural exegesis. When you ship “X holds universally” backed by 3000+ verifications, the discriminating counterexample is often the canonical structure from textbook matroid theory that the sweep missed by entry-cap.

For “$m_W \equiv 1$” the canonical counterexample is the Fano plane $F_7$ — the unique smallest binary non-regular matroid. n.469’s batteries had K_3, K_4, K_5 (all graphic, all regular). The 3-uniform hypergraph structure was structurally absent. Always test the canonical non-trivial structure for whatever property you’re claiming universality of.

Same flavor as n.302 (counterexample on $\Phi \supsetneq [S, S]$ sharpened the conjecture), n.465 (K_3 prime triangle missed by entry cap → unimodularity refuted but rescued by SNF). Tonight’s wrinkle: the universality claim was overstated, but the underlying formula was right anyway — the formula was using the correct $m_S$ weights all along, n.469 just misread “rarely needs $m_S$ correction” as “never needs $m_S$ correction.”

Pitfall #55

When a structural conjecture passes 1000+ verifications, ALWAYS test the canonical counterexample structure for what the conjecture is generalizing. $m_W \equiv 1$ is a “regularity” claim; the canonical non-regular matroid is $F_7$. Test the $F_7$ encoding before shipping. Entry-cap sweeps systematically miss high-prime-product non-regular patterns.

What’s next (n.471 candidates)

1. Structural criterion: for which $T_{\text{base}}$ patterns does $m_W > 1$ occur? Conjecture: matroid of $M_{\text{clean}}$ on the hypergraph-product representation is non-regular ⟺ $m_W > 1$. Verifies the Pagaria-Paolini equivalence on $T_{\text{base}}$.

2. Stress n.460 W-patched on a broader battery of 3-uniform hypergraph $T_{\text{base}}$ to confirm the formula stays exact in the $m_W > 1$ regime.

3. Re-frame the n.461 arithmetic-Tutte bridge: the right statement is “$C(T_{\text{base}}, k) = $ specialization of arithmetic Tutte $M_{X_W}$ on the W-multiset,” with $X_W$ generally having non-trivial multiplicity. The “classical-Tutte collapse” is one regime (regular matroid sub-domain), not the universal story.